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: releng/6.2/sys/nfsclient/nfs_vnops.c 165038 2006-12-09 17:49:56Z delphij $");
   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 
  133 /*
  134  * Global vfs data structures for nfs
  135  */
  136 struct vop_vector nfs_vnodeops = {
  137         .vop_default =          &default_vnodeops,
  138         .vop_access =           nfs_access,
  139         .vop_advlock =          nfs_advlock,
  140         .vop_close =            nfs_close,
  141         .vop_create =           nfs_create,
  142         .vop_fsync =            nfs_fsync,
  143         .vop_getattr =          nfs_getattr,
  144         .vop_getpages =         nfs_getpages,
  145         .vop_putpages =         nfs_putpages,
  146         .vop_inactive =         nfs_inactive,
  147         .vop_lease =            VOP_NULL,
  148         .vop_link =             nfs_link,
  149         .vop_lookup =           nfs_lookup,
  150         .vop_mkdir =            nfs_mkdir,
  151         .vop_mknod =            nfs_mknod,
  152         .vop_open =             nfs_open,
  153         .vop_print =            nfs_print,
  154         .vop_read =             nfs_read,
  155         .vop_readdir =          nfs_readdir,
  156         .vop_readlink =         nfs_readlink,
  157         .vop_reclaim =          nfs_reclaim,
  158         .vop_remove =           nfs_remove,
  159         .vop_rename =           nfs_rename,
  160         .vop_rmdir =            nfs_rmdir,
  161         .vop_setattr =          nfs_setattr,
  162         .vop_strategy =         nfs_strategy,
  163         .vop_symlink =          nfs_symlink,
  164         .vop_write =            nfs_write,
  165 };
  166 
  167 struct vop_vector nfs_fifoops = {
  168         .vop_default =          &fifo_specops,
  169         .vop_access =           nfsspec_access,
  170         .vop_close =            nfsfifo_close,
  171         .vop_fsync =            nfs_fsync,
  172         .vop_getattr =          nfs_getattr,
  173         .vop_inactive =         nfs_inactive,
  174         .vop_print =            nfs_print,
  175         .vop_read =             nfsfifo_read,
  176         .vop_reclaim =          nfs_reclaim,
  177         .vop_setattr =          nfs_setattr,
  178         .vop_write =            nfsfifo_write,
  179 };
  180 
  181 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  182                              struct componentname *cnp, struct vattr *vap);
  183 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  184                               struct ucred *cred, struct thread *td);
  185 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  186                               int fnamelen, struct vnode *tdvp,
  187                               const char *tnameptr, int tnamelen,
  188                               struct ucred *cred, struct thread *td);
  189 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  190                              struct sillyrename *sp);
  191 
  192 /*
  193  * Global variables
  194  */
  195 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  196 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  197 int              nfs_numasync = 0;
  198 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  199 
  200 SYSCTL_DECL(_vfs_nfs);
  201 
  202 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  203 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  204            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  205 
  206 static int      nfsv3_commit_on_close = 0;
  207 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  208            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  209 
  210 static int      nfs_clean_pages_on_close = 1;
  211 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  212            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  213 
  214 int nfs_directio_enable = 0;
  215 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  216            &nfs_directio_enable, 0, "Enable NFS directio");
  217 
  218 /*
  219  * This sysctl allows other processes to mmap a file that has been opened
  220  * O_DIRECT by a process.  In general, having processes mmap the file while
  221  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  222  * this by default to prevent DoS attacks - to prevent a malicious user from
  223  * opening up files O_DIRECT preventing other users from mmap'ing these
  224  * files.  "Protected" environments where stricter consistency guarantees are
  225  * required can disable this knob.  The process that opened the file O_DIRECT
  226  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  227  * meaningful.
  228  */
  229 int nfs_directio_allow_mmap = 1;
  230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  231            &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  232 
  233 #if 0
  234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  235            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  236 
  237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  238            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  239 #endif
  240 
  241 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  242                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  243                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  244 static int
  245 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  246     struct ucred *cred)
  247 {
  248         const int v3 = 1;
  249         u_int32_t *tl;
  250         int error = 0, attrflag;
  251 
  252         struct mbuf *mreq, *mrep, *md, *mb;
  253         caddr_t bpos, dpos;
  254         u_int32_t rmode;
  255         struct nfsnode *np = VTONFS(vp);
  256 
  257         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  258         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  259         mb = mreq;
  260         bpos = mtod(mb, caddr_t);
  261         nfsm_fhtom(vp, v3);
  262         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  263         *tl = txdr_unsigned(wmode);
  264         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  265         nfsm_postop_attr(vp, attrflag);
  266         if (!error) {
  267                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  268                 rmode = fxdr_unsigned(u_int32_t, *tl);
  269                 np->n_mode = rmode;
  270                 np->n_modeuid = cred->cr_uid;
  271                 np->n_modestamp = time_second;
  272         }
  273         m_freem(mrep);
  274 nfsmout:
  275         return (error);
  276 }
  277 
  278 /*
  279  * nfs access vnode op.
  280  * For nfs version 2, just return ok. File accesses may fail later.
  281  * For nfs version 3, use the access rpc to check accessibility. If file modes
  282  * are changed on the server, accesses might still fail later.
  283  */
  284 static int
  285 nfs_access(struct vop_access_args *ap)
  286 {
  287         struct vnode *vp = ap->a_vp;
  288         int error = 0;
  289         u_int32_t mode, wmode;
  290         int v3 = NFS_ISV3(vp);
  291         struct nfsnode *np = VTONFS(vp);
  292 
  293         /*
  294          * Disallow write attempts on filesystems mounted read-only;
  295          * unless the file is a socket, fifo, or a block or character
  296          * device resident on the filesystem.
  297          */
  298         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  299                 switch (vp->v_type) {
  300                 case VREG:
  301                 case VDIR:
  302                 case VLNK:
  303                         return (EROFS);
  304                 default:
  305                         break;
  306                 }
  307         }
  308         /*
  309          * For nfs v3, check to see if we have done this recently, and if
  310          * so return our cached result instead of making an ACCESS call.
  311          * If not, do an access rpc, otherwise you are stuck emulating
  312          * ufs_access() locally using the vattr. This may not be correct,
  313          * since the server may apply other access criteria such as
  314          * client uid-->server uid mapping that we do not know about.
  315          */
  316         if (v3) {
  317                 if (ap->a_mode & VREAD)
  318                         mode = NFSV3ACCESS_READ;
  319                 else
  320                         mode = 0;
  321                 if (vp->v_type != VDIR) {
  322                         if (ap->a_mode & VWRITE)
  323                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  324                         if (ap->a_mode & VEXEC)
  325                                 mode |= NFSV3ACCESS_EXECUTE;
  326                 } else {
  327                         if (ap->a_mode & VWRITE)
  328                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  329                                          NFSV3ACCESS_DELETE);
  330                         if (ap->a_mode & VEXEC)
  331                                 mode |= NFSV3ACCESS_LOOKUP;
  332                 }
  333                 /* XXX safety belt, only make blanket request if caching */
  334                 if (nfsaccess_cache_timeout > 0) {
  335                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  336                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  337                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  338                 } else {
  339                         wmode = mode;
  340                 }
  341 
  342                 /*
  343                  * Does our cached result allow us to give a definite yes to
  344                  * this request?
  345                  */
  346                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  347                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  348                     ((np->n_mode & mode) == mode)) {
  349                         nfsstats.accesscache_hits++;
  350                 } else {
  351                         /*
  352                          * Either a no, or a don't know.  Go to the wire.
  353                          */
  354                         nfsstats.accesscache_misses++;
  355                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  356                         if (!error) {
  357                                 if ((np->n_mode & mode) != mode) {
  358                                         error = EACCES;
  359                                 }
  360                         }
  361                 }
  362                 return (error);
  363         } else {
  364                 if ((error = nfsspec_access(ap)) != 0)
  365                         return (error);
  366 
  367                 /*
  368                  * Attempt to prevent a mapped root from accessing a file
  369                  * which it shouldn't.  We try to read a byte from the file
  370                  * if the user is root and the file is not zero length.
  371                  * After calling nfsspec_access, we should have the correct
  372                  * file size cached.
  373                  */
  374                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  375                     && VTONFS(vp)->n_size > 0) {
  376                         struct iovec aiov;
  377                         struct uio auio;
  378                         char buf[1];
  379 
  380                         aiov.iov_base = buf;
  381                         aiov.iov_len = 1;
  382                         auio.uio_iov = &aiov;
  383                         auio.uio_iovcnt = 1;
  384                         auio.uio_offset = 0;
  385                         auio.uio_resid = 1;
  386                         auio.uio_segflg = UIO_SYSSPACE;
  387                         auio.uio_rw = UIO_READ;
  388                         auio.uio_td = ap->a_td;
  389 
  390                         if (vp->v_type == VREG)
  391                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  392                         else if (vp->v_type == VDIR) {
  393                                 char* bp;
  394                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  395                                 aiov.iov_base = bp;
  396                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  397                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  398                                 free(bp, M_TEMP);
  399                         } else if (vp->v_type == VLNK)
  400                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  401                         else
  402                                 error = EACCES;
  403                 }
  404                 return (error);
  405         }
  406 }
  407 
  408 /*
  409  * nfs open vnode op
  410  * Check to see if the type is ok
  411  * and that deletion is not in progress.
  412  * For paged in text files, you will need to flush the page cache
  413  * if consistency is lost.
  414  */
  415 /* ARGSUSED */
  416 static int
  417 nfs_open(struct vop_open_args *ap)
  418 {
  419         struct vnode *vp = ap->a_vp;
  420         struct nfsnode *np = VTONFS(vp);
  421         struct vattr vattr;
  422         int error;
  423         int fmode = ap->a_mode;
  424 
  425         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  426                 return (EOPNOTSUPP);
  427 
  428         /*
  429          * Get a valid lease. If cached data is stale, flush it.
  430          */
  431         if (np->n_flag & NMODIFIED) {
  432                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  433                 if (error == EINTR || error == EIO)
  434                         return (error);
  435                 np->n_attrstamp = 0;
  436                 if (vp->v_type == VDIR)
  437                         np->n_direofoffset = 0;
  438                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  439                 if (error)
  440                         return (error);
  441                 np->n_mtime = vattr.va_mtime;
  442         } else {
  443                 np->n_attrstamp = 0;
  444                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  445                 if (error)
  446                         return (error);
  447                 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  448                         if (vp->v_type == VDIR)
  449                                 np->n_direofoffset = 0;
  450                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  451                         if (error == EINTR || error == EIO)
  452                                 return (error);
  453                         np->n_mtime = vattr.va_mtime;
  454                 }
  455         }
  456         /*
  457          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  458          */
  459         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  460                 if (np->n_directio_opens == 0) {
  461                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  462                         if (error)
  463                                 return (error);
  464                         np->n_flag |= NNONCACHE;
  465                 }
  466                 np->n_directio_opens++;
  467         }
  468         np->ra_expect_lbn = 0;
  469         vnode_create_vobject_off(vp, vattr.va_size, ap->a_td);
  470         return (0);
  471 }
  472 
  473 /*
  474  * nfs close vnode op
  475  * What an NFS client should do upon close after writing is a debatable issue.
  476  * Most NFS clients push delayed writes to the server upon close, basically for
  477  * two reasons:
  478  * 1 - So that any write errors may be reported back to the client process
  479  *     doing the close system call. By far the two most likely errors are
  480  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  481  * 2 - To put a worst case upper bound on cache inconsistency between
  482  *     multiple clients for the file.
  483  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  484  * not being able to tell if other clients are writing a file concurrently,
  485  * since there is no way of knowing if the changed modify time in the reply
  486  * is only due to the write for this client.
  487  * (NFS Version 3 provides weak cache consistency data in the reply that
  488  *  should be sufficient to detect and handle this case.)
  489  *
  490  * The current code does the following:
  491  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  492  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  493  *                     or commit them (this satisfies 1 and 2 except for the
  494  *                     case where the server crashes after this close but
  495  *                     before the commit RPC, which is felt to be "good
  496  *                     enough". Changing the last argument to nfs_flush() to
  497  *                     a 1 would force a commit operation, if it is felt a
  498  *                     commit is necessary now.
  499  */
  500 /* ARGSUSED */
  501 static int
  502 nfs_close(struct vop_close_args *ap)
  503 {
  504         struct vnode *vp = ap->a_vp;
  505         struct nfsnode *np = VTONFS(vp);
  506         int error = 0;
  507         int fmode = ap->a_fflag;
  508 
  509         if (vp->v_type == VREG) {
  510             /*
  511              * Examine and clean dirty pages, regardless of NMODIFIED.
  512              * This closes a major hole in close-to-open consistency.
  513              * We want to push out all dirty pages (and buffers) on
  514              * close, regardless of whether they were dirtied by
  515              * mmap'ed writes or via write().
  516              */
  517             if (nfs_clean_pages_on_close && vp->v_object) {
  518                 VM_OBJECT_LOCK(vp->v_object);
  519                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  520                 VM_OBJECT_UNLOCK(vp->v_object);
  521             }
  522             if (np->n_flag & NMODIFIED) {
  523                 if (NFS_ISV3(vp)) {
  524                     /*
  525                      * Under NFSv3 we have dirty buffers to dispose of.  We
  526                      * must flush them to the NFS server.  We have the option
  527                      * of waiting all the way through the commit rpc or just
  528                      * waiting for the initial write.  The default is to only
  529                      * wait through the initial write so the data is in the
  530                      * server's cache, which is roughly similar to the state
  531                      * a standard disk subsystem leaves the file in on close().
  532                      *
  533                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  534                      * potential races with other processes, and certainly
  535                      * cannot clear it if we don't commit.
  536                      */
  537                     int cm = nfsv3_commit_on_close ? 1 : 0;
  538                     error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
  539                     /* np->n_flag &= ~NMODIFIED; */
  540                 } else
  541                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  542             }
  543             /* 
  544              * Invalidate the attribute cache in all cases.
  545              * An open is going to fetch fresh attrs any way, other procs
  546              * on this node that have file open will be forced to do an 
  547              * otw attr fetch, but this is safe.
  548              */
  549             np->n_attrstamp = 0;
  550             if (np->n_flag & NWRITEERR) {
  551                 np->n_flag &= ~NWRITEERR;
  552                 error = np->n_error;
  553             }
  554         }
  555         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  556                 KASSERT((np->n_directio_opens > 0), 
  557                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));           
  558                 np->n_directio_opens--;
  559                 if (np->n_directio_opens == 0)
  560                         np->n_flag &= ~NNONCACHE;
  561         }
  562         return (error);
  563 }
  564 
  565 /*
  566  * nfs getattr call from vfs.
  567  */
  568 static int
  569 nfs_getattr(struct vop_getattr_args *ap)
  570 {
  571         struct vnode *vp = ap->a_vp;
  572         struct nfsnode *np = VTONFS(vp);
  573         caddr_t bpos, dpos;
  574         int error = 0;
  575         struct mbuf *mreq, *mrep, *md, *mb;
  576         int v3 = NFS_ISV3(vp);
  577 
  578         /*
  579          * Update local times for special files.
  580          */
  581         if (np->n_flag & (NACC | NUPD))
  582                 np->n_flag |= NCHG;
  583         /*
  584          * First look in the cache.
  585          */
  586         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  587                 return (0);
  588 
  589         if (v3 && nfsaccess_cache_timeout > 0) {
  590                 nfsstats.accesscache_misses++;
  591                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  592                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  593                         return (0);
  594         }
  595 
  596         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  597         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  598         mb = mreq;
  599         bpos = mtod(mb, caddr_t);
  600         nfsm_fhtom(vp, v3);
  601         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  602         if (!error) {
  603                 nfsm_loadattr(vp, ap->a_vap);
  604         }
  605         m_freem(mrep);
  606 nfsmout:
  607         return (error);
  608 }
  609 
  610 /*
  611  * nfs setattr call.
  612  */
  613 static int
  614 nfs_setattr(struct vop_setattr_args *ap)
  615 {
  616         struct vnode *vp = ap->a_vp;
  617         struct nfsnode *np = VTONFS(vp);
  618         struct vattr *vap = ap->a_vap;
  619         int error = 0;
  620         u_quad_t tsize;
  621 
  622 #ifndef nolint
  623         tsize = (u_quad_t)0;
  624 #endif
  625 
  626         /*
  627          * Setting of flags and marking of atimes are not supported.
  628          */
  629         if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
  630                 return (EOPNOTSUPP);
  631 
  632         /*
  633          * Disallow write attempts if the filesystem is mounted read-only.
  634          */
  635         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  636             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  637             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  638             (vp->v_mount->mnt_flag & MNT_RDONLY))
  639                 return (EROFS);
  640         if (vap->va_size != VNOVAL) {
  641                 switch (vp->v_type) {
  642                 case VDIR:
  643                         return (EISDIR);
  644                 case VCHR:
  645                 case VBLK:
  646                 case VSOCK:
  647                 case VFIFO:
  648                         if (vap->va_mtime.tv_sec == VNOVAL &&
  649                             vap->va_atime.tv_sec == VNOVAL &&
  650                             vap->va_mode == (mode_t)VNOVAL &&
  651                             vap->va_uid == (uid_t)VNOVAL &&
  652                             vap->va_gid == (gid_t)VNOVAL)
  653                                 return (0);
  654                         vap->va_size = VNOVAL;
  655                         break;
  656                 default:
  657                         /*
  658                          * Disallow write attempts if the filesystem is
  659                          * mounted read-only.
  660                          */
  661                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  662                                 return (EROFS);
  663 
  664                         /*
  665                          *  We run vnode_pager_setsize() early (why?),
  666                          * we must set np->n_size now to avoid vinvalbuf
  667                          * V_SAVE races that might setsize a lower
  668                          * value.
  669                          */
  670 
  671                         tsize = np->n_size;
  672                         error = nfs_meta_setsize(vp, ap->a_cred, 
  673                                                 ap->a_td, vap->va_size);
  674 
  675                         if (np->n_flag & NMODIFIED) {
  676                             if (vap->va_size == 0)
  677                                 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
  678                             else
  679                                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  680                             if (error) {
  681                                 vnode_pager_setsize(vp, np->n_size);
  682                                 return (error);
  683                             }
  684                         }
  685                         /*
  686                          * np->n_size has already been set to vap->va_size
  687                          * in nfs_meta_setsize(). We must set it again since
  688                          * nfs_loadattrcache() could be called through
  689                          * nfs_meta_setsize() and could modify np->n_size.
  690                          */
  691                         np->n_vattr.va_size = np->n_size = vap->va_size;
  692                 };
  693         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  694                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  695                 vp->v_type == VREG &&
  696                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
  697                     (error == EINTR || error == EIO))
  698                 return (error);
  699         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  700         if (error && vap->va_size != VNOVAL) {
  701                 np->n_size = np->n_vattr.va_size = tsize;
  702                 vnode_pager_setsize(vp, np->n_size);
  703         }
  704         return (error);
  705 }
  706 
  707 /*
  708  * Do an nfs setattr rpc.
  709  */
  710 static int
  711 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  712     struct thread *td)
  713 {
  714         struct nfsv2_sattr *sp;
  715         struct nfsnode *np = VTONFS(vp);
  716         caddr_t bpos, dpos;
  717         u_int32_t *tl;
  718         int error = 0, wccflag = NFSV3_WCCRATTR;
  719         struct mbuf *mreq, *mrep, *md, *mb;
  720         int v3 = NFS_ISV3(vp);
  721 
  722         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  723         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  724         mb = mreq;
  725         bpos = mtod(mb, caddr_t);
  726         nfsm_fhtom(vp, v3);
  727         if (v3) {
  728                 nfsm_v3attrbuild(vap, TRUE);
  729                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  730                 *tl = nfs_false;
  731         } else {
  732                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  733                 if (vap->va_mode == (mode_t)VNOVAL)
  734                         sp->sa_mode = nfs_xdrneg1;
  735                 else
  736                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  737                 if (vap->va_uid == (uid_t)VNOVAL)
  738                         sp->sa_uid = nfs_xdrneg1;
  739                 else
  740                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  741                 if (vap->va_gid == (gid_t)VNOVAL)
  742                         sp->sa_gid = nfs_xdrneg1;
  743                 else
  744                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  745                 sp->sa_size = txdr_unsigned(vap->va_size);
  746                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  747                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  748         }
  749         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  750         if (v3) {
  751                 np->n_modestamp = 0;
  752                 nfsm_wcc_data(vp, wccflag);
  753         } else
  754                 nfsm_loadattr(vp, NULL);
  755         m_freem(mrep);
  756 nfsmout:
  757         return (error);
  758 }
  759 
  760 /*
  761  * nfs lookup call, one step at a time...
  762  * First look in cache
  763  * If not found, unlock the directory nfsnode and do the rpc
  764  */
  765 static int
  766 nfs_lookup(struct vop_lookup_args *ap)
  767 {
  768         struct componentname *cnp = ap->a_cnp;
  769         struct vnode *dvp = ap->a_dvp;
  770         struct vnode **vpp = ap->a_vpp;
  771         int flags = cnp->cn_flags;
  772         struct vnode *newvp;
  773         struct nfsmount *nmp;
  774         caddr_t bpos, dpos;
  775         struct mbuf *mreq, *mrep, *md, *mb;
  776         long len;
  777         nfsfh_t *fhp;
  778         struct nfsnode *np;
  779         int error = 0, attrflag, fhsize;
  780         int v3 = NFS_ISV3(dvp);
  781         struct thread *td = cnp->cn_thread;
  782 
  783         *vpp = NULLVP;
  784         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  785             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  786                 return (EROFS);
  787         if (dvp->v_type != VDIR)
  788                 return (ENOTDIR);
  789         nmp = VFSTONFS(dvp->v_mount);
  790         np = VTONFS(dvp);
  791         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  792                 *vpp = NULLVP;
  793                 return (error);
  794         }
  795         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  796                 struct vattr vattr;
  797 
  798                 newvp = *vpp;
  799                 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  800                  && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  801                      nfsstats.lookupcache_hits++;
  802                      if (cnp->cn_nameiop != LOOKUP &&
  803                          (flags & ISLASTCN))
  804                              cnp->cn_flags |= SAVENAME;
  805                      return (0);
  806                 }
  807                 cache_purge(newvp);
  808                 if (dvp != newvp)
  809                         vput(newvp);
  810                 else 
  811                         vrele(newvp);
  812                 *vpp = NULLVP;
  813         }
  814         error = 0;
  815         newvp = NULLVP;
  816         nfsstats.lookupcache_misses++;
  817         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  818         len = cnp->cn_namelen;
  819         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  820                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  821         mb = mreq;
  822         bpos = mtod(mb, caddr_t);
  823         nfsm_fhtom(dvp, v3);
  824         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  825         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  826         if (error) {
  827                 if (v3) {
  828                         nfsm_postop_attr(dvp, attrflag);
  829                         m_freem(mrep);
  830                 }
  831                 goto nfsmout;
  832         }
  833         nfsm_getfh(fhp, fhsize, v3);
  834 
  835         /*
  836          * Handle RENAME case...
  837          */
  838         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
  839                 if (NFS_CMPFH(np, fhp, fhsize)) {
  840                         m_freem(mrep);
  841                         return (EISDIR);
  842                 }
  843                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  844                 if (error) {
  845                         m_freem(mrep);
  846                         return (error);
  847                 }
  848                 newvp = NFSTOV(np);
  849                 if (v3) {
  850                         nfsm_postop_attr(newvp, attrflag);
  851                         nfsm_postop_attr(dvp, attrflag);
  852                 } else
  853                         nfsm_loadattr(newvp, NULL);
  854                 *vpp = newvp;
  855                 m_freem(mrep);
  856                 cnp->cn_flags |= SAVENAME;
  857                 return (0);
  858         }
  859 
  860         if (flags & ISDOTDOT) {
  861                 VOP_UNLOCK(dvp, 0, td);
  862                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  863                 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  864                 if (error)
  865                         return (error);
  866                 newvp = NFSTOV(np);
  867         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  868                 VREF(dvp);
  869                 newvp = dvp;
  870         } else {
  871                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  872                 if (error) {
  873                         m_freem(mrep);
  874                         return (error);
  875                 }
  876                 newvp = NFSTOV(np);
  877         }
  878         if (v3) {
  879                 nfsm_postop_attr(newvp, attrflag);
  880                 nfsm_postop_attr(dvp, attrflag);
  881         } else
  882                 nfsm_loadattr(newvp, NULL);
  883         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  884                 cnp->cn_flags |= SAVENAME;
  885         if ((cnp->cn_flags & MAKEENTRY) &&
  886             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  887                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  888                 cache_enter(dvp, newvp, cnp);
  889         }
  890         *vpp = newvp;
  891         m_freem(mrep);
  892 nfsmout:
  893         if (error) {
  894                 if (newvp != NULLVP) {
  895                         vput(newvp);
  896                         *vpp = NULLVP;
  897                 }
  898                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  899                     (flags & ISLASTCN) && error == ENOENT) {
  900                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  901                                 error = EROFS;
  902                         else
  903                                 error = EJUSTRETURN;
  904                 }
  905                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  906                         cnp->cn_flags |= SAVENAME;
  907         }
  908         return (error);
  909 }
  910 
  911 /*
  912  * nfs read call.
  913  * Just call nfs_bioread() to do the work.
  914  */
  915 static int
  916 nfs_read(struct vop_read_args *ap)
  917 {
  918         struct vnode *vp = ap->a_vp;
  919 
  920         switch (vp->v_type) {
  921         case VREG:
  922                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  923         case VDIR:
  924                 return (EISDIR);
  925         default:
  926                 return (EOPNOTSUPP);
  927         }
  928 }
  929 
  930 /*
  931  * nfs readlink call
  932  */
  933 static int
  934 nfs_readlink(struct vop_readlink_args *ap)
  935 {
  936         struct vnode *vp = ap->a_vp;
  937 
  938         if (vp->v_type != VLNK)
  939                 return (EINVAL);
  940         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  941 }
  942 
  943 /*
  944  * Do a readlink rpc.
  945  * Called by nfs_doio() from below the buffer cache.
  946  */
  947 int
  948 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  949 {
  950         caddr_t bpos, dpos;
  951         int error = 0, len, attrflag;
  952         struct mbuf *mreq, *mrep, *md, *mb;
  953         int v3 = NFS_ISV3(vp);
  954 
  955         nfsstats.rpccnt[NFSPROC_READLINK]++;
  956         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  957         mb = mreq;
  958         bpos = mtod(mb, caddr_t);
  959         nfsm_fhtom(vp, v3);
  960         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  961         if (v3)
  962                 nfsm_postop_attr(vp, attrflag);
  963         if (!error) {
  964                 nfsm_strsiz(len, NFS_MAXPATHLEN);
  965                 if (len == NFS_MAXPATHLEN) {
  966                         struct nfsnode *np = VTONFS(vp);
  967                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
  968                                 len = np->n_size;
  969                 }
  970                 nfsm_mtouio(uiop, len);
  971         }
  972         m_freem(mrep);
  973 nfsmout:
  974         return (error);
  975 }
  976 
  977 /*
  978  * nfs read rpc call
  979  * Ditto above
  980  */
  981 int
  982 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  983 {
  984         u_int32_t *tl;
  985         caddr_t bpos, dpos;
  986         struct mbuf *mreq, *mrep, *md, *mb;
  987         struct nfsmount *nmp;
  988         int error = 0, len, retlen, tsiz, eof, attrflag;
  989         int v3 = NFS_ISV3(vp);
  990 
  991 #ifndef nolint
  992         eof = 0;
  993 #endif
  994         nmp = VFSTONFS(vp->v_mount);
  995         tsiz = uiop->uio_resid;
  996         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
  997                 return (EFBIG);
  998         while (tsiz > 0) {
  999                 nfsstats.rpccnt[NFSPROC_READ]++;
 1000                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1001                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1002                 mb = mreq;
 1003                 bpos = mtod(mb, caddr_t);
 1004                 nfsm_fhtom(vp, v3);
 1005                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1006                 if (v3) {
 1007                         txdr_hyper(uiop->uio_offset, tl);
 1008                         *(tl + 2) = txdr_unsigned(len);
 1009                 } else {
 1010                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1011                         *tl++ = txdr_unsigned(len);
 1012                         *tl = 0;
 1013                 }
 1014                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1015                 if (v3) {
 1016                         nfsm_postop_attr(vp, attrflag);
 1017                         if (error) {
 1018                                 m_freem(mrep);
 1019                                 goto nfsmout;
 1020                         }
 1021                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1022                         eof = fxdr_unsigned(int, *(tl + 1));
 1023                 } else
 1024                         nfsm_loadattr(vp, NULL);
 1025                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1026                 nfsm_mtouio(uiop, retlen);
 1027                 m_freem(mrep);
 1028                 tsiz -= retlen;
 1029                 if (v3) {
 1030                         if (eof || retlen == 0) {
 1031                                 tsiz = 0;
 1032                         }
 1033                 } else if (retlen < len) {
 1034                         tsiz = 0;
 1035                 }
 1036         }
 1037 nfsmout:
 1038         return (error);
 1039 }
 1040 
 1041 /*
 1042  * nfs write call
 1043  */
 1044 int
 1045 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1046     int *iomode, int *must_commit)
 1047 {
 1048         u_int32_t *tl;
 1049         int32_t backup;
 1050         caddr_t bpos, dpos;
 1051         struct mbuf *mreq, *mrep, *md, *mb;
 1052         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1053         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1054         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1055 
 1056 #ifndef DIAGNOSTIC
 1057         if (uiop->uio_iovcnt != 1)
 1058                 panic("nfs: writerpc iovcnt > 1");
 1059 #endif
 1060         *must_commit = 0;
 1061         tsiz = uiop->uio_resid;
 1062         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1063                 return (EFBIG);
 1064         while (tsiz > 0) {
 1065                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1066                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1067                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1068                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1069                 mb = mreq;
 1070                 bpos = mtod(mb, caddr_t);
 1071                 nfsm_fhtom(vp, v3);
 1072                 if (v3) {
 1073                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1074                         txdr_hyper(uiop->uio_offset, tl);
 1075                         tl += 2;
 1076                         *tl++ = txdr_unsigned(len);
 1077                         *tl++ = txdr_unsigned(*iomode);
 1078                         *tl = txdr_unsigned(len);
 1079                 } else {
 1080                         u_int32_t x;
 1081 
 1082                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1083                         /* Set both "begin" and "current" to non-garbage. */
 1084                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1085                         *tl++ = x;      /* "begin offset" */
 1086                         *tl++ = x;      /* "current offset" */
 1087                         x = txdr_unsigned(len);
 1088                         *tl++ = x;      /* total to this offset */
 1089                         *tl = x;        /* size of this write */
 1090                 }
 1091                 nfsm_uiotom(uiop, len);
 1092                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1093                 if (v3) {
 1094                         wccflag = NFSV3_WCCCHK;
 1095                         nfsm_wcc_data(vp, wccflag);
 1096                         if (!error) {
 1097                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1098                                         + NFSX_V3WRITEVERF);
 1099                                 rlen = fxdr_unsigned(int, *tl++);
 1100                                 if (rlen == 0) {
 1101                                         error = NFSERR_IO;
 1102                                         m_freem(mrep);
 1103                                         break;
 1104                                 } else if (rlen < len) {
 1105                                         backup = len - rlen;
 1106                                         uiop->uio_iov->iov_base =
 1107                                             (char *)uiop->uio_iov->iov_base -
 1108                                             backup;
 1109                                         uiop->uio_iov->iov_len += backup;
 1110                                         uiop->uio_offset -= backup;
 1111                                         uiop->uio_resid += backup;
 1112                                         len = rlen;
 1113                                 }
 1114                                 commit = fxdr_unsigned(int, *tl++);
 1115 
 1116                                 /*
 1117                                  * Return the lowest committment level
 1118                                  * obtained by any of the RPCs.
 1119                                  */
 1120                                 if (committed == NFSV3WRITE_FILESYNC)
 1121                                         committed = commit;
 1122                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1123                                         commit == NFSV3WRITE_UNSTABLE)
 1124                                         committed = commit;
 1125                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1126                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1127                                         NFSX_V3WRITEVERF);
 1128                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1129                                 } else if (bcmp((caddr_t)tl,
 1130                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1131                                     *must_commit = 1;
 1132                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1133                                         NFSX_V3WRITEVERF);
 1134                                 }
 1135                         }
 1136                 } else
 1137                     nfsm_loadattr(vp, NULL);
 1138                 if (wccflag)
 1139                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
 1140                 m_freem(mrep);
 1141                 if (error)
 1142                         break;
 1143                 tsiz -= len;
 1144         }
 1145 nfsmout:
 1146         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1147                 committed = NFSV3WRITE_FILESYNC;
 1148         *iomode = committed;
 1149         if (error)
 1150                 uiop->uio_resid = tsiz;
 1151         return (error);
 1152 }
 1153 
 1154 /*
 1155  * nfs mknod rpc
 1156  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1157  * mode set to specify the file type and the size field for rdev.
 1158  */
 1159 static int
 1160 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1161     struct vattr *vap)
 1162 {
 1163         struct nfsv2_sattr *sp;
 1164         u_int32_t *tl;
 1165         struct vnode *newvp = NULL;
 1166         struct nfsnode *np = NULL;
 1167         struct vattr vattr;
 1168         caddr_t bpos, dpos;
 1169         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1170         struct mbuf *mreq, *mrep, *md, *mb;
 1171         u_int32_t rdev;
 1172         int v3 = NFS_ISV3(dvp);
 1173 
 1174         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1175                 rdev = txdr_unsigned(vap->va_rdev);
 1176         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1177                 rdev = nfs_xdrneg1;
 1178         else {
 1179                 return (EOPNOTSUPP);
 1180         }
 1181         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1182                 return (error);
 1183         }
 1184         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1185         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1186                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1187         mb = mreq;
 1188         bpos = mtod(mb, caddr_t);
 1189         nfsm_fhtom(dvp, v3);
 1190         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1191         if (v3) {
 1192                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1193                 *tl++ = vtonfsv3_type(vap->va_type);
 1194                 nfsm_v3attrbuild(vap, FALSE);
 1195                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1196                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1197                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1198                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1199                 }
 1200         } else {
 1201                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1202                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1203                 sp->sa_uid = nfs_xdrneg1;
 1204                 sp->sa_gid = nfs_xdrneg1;
 1205                 sp->sa_size = rdev;
 1206                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1207                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1208         }
 1209         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1210         if (!error) {
 1211                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1212                 if (!gotvp) {
 1213                         if (newvp) {
 1214                                 vput(newvp);
 1215                                 newvp = NULL;
 1216                         }
 1217                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1218                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1219                         if (!error)
 1220                                 newvp = NFSTOV(np);
 1221                 }
 1222         }
 1223         if (v3)
 1224                 nfsm_wcc_data(dvp, wccflag);
 1225         m_freem(mrep);
 1226 nfsmout:
 1227         if (error) {
 1228                 if (newvp)
 1229                         vput(newvp);
 1230         } else {
 1231                 if (cnp->cn_flags & MAKEENTRY)
 1232                         cache_enter(dvp, newvp, cnp);
 1233                 *vpp = newvp;
 1234         }
 1235         VTONFS(dvp)->n_flag |= NMODIFIED;
 1236         if (!wccflag)
 1237                 VTONFS(dvp)->n_attrstamp = 0;
 1238         return (error);
 1239 }
 1240 
 1241 /*
 1242  * nfs mknod vop
 1243  * just call nfs_mknodrpc() to do the work.
 1244  */
 1245 /* ARGSUSED */
 1246 static int
 1247 nfs_mknod(struct vop_mknod_args *ap)
 1248 {
 1249 
 1250         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1251 }
 1252 
 1253 static u_long create_verf;
 1254 /*
 1255  * nfs file create call
 1256  */
 1257 static int
 1258 nfs_create(struct vop_create_args *ap)
 1259 {
 1260         struct vnode *dvp = ap->a_dvp;
 1261         struct vattr *vap = ap->a_vap;
 1262         struct componentname *cnp = ap->a_cnp;
 1263         struct nfsv2_sattr *sp;
 1264         u_int32_t *tl;
 1265         struct nfsnode *np = NULL;
 1266         struct vnode *newvp = NULL;
 1267         caddr_t bpos, dpos;
 1268         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1269         struct mbuf *mreq, *mrep, *md, *mb;
 1270         struct vattr vattr;
 1271         int v3 = NFS_ISV3(dvp);
 1272 
 1273         /*
 1274          * Oops, not for me..
 1275          */
 1276         if (vap->va_type == VSOCK)
 1277                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1278 
 1279         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1280                 return (error);
 1281         }
 1282         if (vap->va_vaflags & VA_EXCLUSIVE)
 1283                 fmode |= O_EXCL;
 1284 again:
 1285         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1286         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1287                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1288         mb = mreq;
 1289         bpos = mtod(mb, caddr_t);
 1290         nfsm_fhtom(dvp, v3);
 1291         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1292         if (v3) {
 1293                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1294                 if (fmode & O_EXCL) {
 1295                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1296                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1297 #ifdef INET
 1298                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1299                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1300                         else
 1301 #endif
 1302                                 *tl++ = create_verf;
 1303                         *tl = ++create_verf;
 1304                 } else {
 1305                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1306                         nfsm_v3attrbuild(vap, FALSE);
 1307                 }
 1308         } else {
 1309                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1310                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1311                 sp->sa_uid = nfs_xdrneg1;
 1312                 sp->sa_gid = nfs_xdrneg1;
 1313                 sp->sa_size = 0;
 1314                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1315                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1316         }
 1317         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1318         if (!error) {
 1319                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1320                 if (!gotvp) {
 1321                         if (newvp) {
 1322                                 vput(newvp);
 1323                                 newvp = NULL;
 1324                         }
 1325                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1326                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1327                         if (!error)
 1328                                 newvp = NFSTOV(np);
 1329                 }
 1330         }
 1331         if (v3)
 1332                 nfsm_wcc_data(dvp, wccflag);
 1333         m_freem(mrep);
 1334 nfsmout:
 1335         if (error) {
 1336                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1337                         fmode &= ~O_EXCL;
 1338                         goto again;
 1339                 }
 1340                 if (newvp)
 1341                         vput(newvp);
 1342         } else if (v3 && (fmode & O_EXCL)) {
 1343                 /*
 1344                  * We are normally called with only a partially initialized
 1345                  * VAP.  Since the NFSv3 spec says that server may use the
 1346                  * file attributes to store the verifier, the spec requires
 1347                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1348                  * in atime, but we can't really assume that all servers will
 1349                  * so we ensure that our SETATTR sets both atime and mtime.
 1350                  */
 1351                 if (vap->va_mtime.tv_sec == VNOVAL)
 1352                         vfs_timestamp(&vap->va_mtime);
 1353                 if (vap->va_atime.tv_sec == VNOVAL)
 1354                         vap->va_atime = vap->va_mtime;
 1355                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1356         }
 1357         if (!error) {
 1358                 if (cnp->cn_flags & MAKEENTRY)
 1359                         cache_enter(dvp, newvp, cnp);
 1360                 *ap->a_vpp = newvp;
 1361         }
 1362         VTONFS(dvp)->n_flag |= NMODIFIED;
 1363         if (!wccflag)
 1364                 VTONFS(dvp)->n_attrstamp = 0;
 1365         return (error);
 1366 }
 1367 
 1368 /*
 1369  * nfs file remove call
 1370  * To try and make nfs semantics closer to ufs semantics, a file that has
 1371  * other processes using the vnode is renamed instead of removed and then
 1372  * removed later on the last close.
 1373  * - If v_usecount > 1
 1374  *        If a rename is not already in the works
 1375  *           call nfs_sillyrename() to set it up
 1376  *     else
 1377  *        do the remove rpc
 1378  */
 1379 static int
 1380 nfs_remove(struct vop_remove_args *ap)
 1381 {
 1382         struct vnode *vp = ap->a_vp;
 1383         struct vnode *dvp = ap->a_dvp;
 1384         struct componentname *cnp = ap->a_cnp;
 1385         struct nfsnode *np = VTONFS(vp);
 1386         int error = 0;
 1387         struct vattr vattr;
 1388 
 1389 #ifndef DIAGNOSTIC
 1390         if ((cnp->cn_flags & HASBUF) == 0)
 1391                 panic("nfs_remove: no name");
 1392         if (vrefcnt(vp) < 1)
 1393                 panic("nfs_remove: bad v_usecount");
 1394 #endif
 1395         if (vp->v_type == VDIR)
 1396                 error = EPERM;
 1397         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1398             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1399             vattr.va_nlink > 1)) {
 1400                 /*
 1401                  * Purge the name cache so that the chance of a lookup for
 1402                  * the name succeeding while the remove is in progress is
 1403                  * minimized. Without node locking it can still happen, such
 1404                  * that an I/O op returns ESTALE, but since you get this if
 1405                  * another host removes the file..
 1406                  */
 1407                 cache_purge(vp);
 1408                 /*
 1409                  * throw away biocache buffers, mainly to avoid
 1410                  * unnecessary delayed writes later.
 1411                  */
 1412                 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1413                 /* Do the rpc */
 1414                 if (error != EINTR && error != EIO)
 1415                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1416                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1417                 /*
 1418                  * Kludge City: If the first reply to the remove rpc is lost..
 1419                  *   the reply to the retransmitted request will be ENOENT
 1420                  *   since the file was in fact removed
 1421                  *   Therefore, we cheat and return success.
 1422                  */
 1423                 if (error == ENOENT)
 1424                         error = 0;
 1425         } else if (!np->n_sillyrename)
 1426                 error = nfs_sillyrename(dvp, vp, cnp);
 1427         np->n_attrstamp = 0;
 1428         return (error);
 1429 }
 1430 
 1431 /*
 1432  * nfs file remove rpc called from nfs_inactive
 1433  */
 1434 int
 1435 nfs_removeit(struct sillyrename *sp)
 1436 {
 1437 
 1438         /*
 1439          * Make sure that the directory vnode is still valid.
 1440          * XXX we should lock sp->s_dvp here.
 1441          */
 1442         if (sp->s_dvp->v_type == VBAD)
 1443                 return (0);
 1444         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1445                 NULL));
 1446 }
 1447 
 1448 /*
 1449  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1450  */
 1451 static int
 1452 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1453     struct ucred *cred, struct thread *td)
 1454 {
 1455         caddr_t bpos, dpos;
 1456         int error = 0, wccflag = NFSV3_WCCRATTR;
 1457         struct mbuf *mreq, *mrep, *md, *mb;
 1458         int v3 = NFS_ISV3(dvp);
 1459 
 1460         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1461         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1462                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1463         mb = mreq;
 1464         bpos = mtod(mb, caddr_t);
 1465         nfsm_fhtom(dvp, v3);
 1466         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1467         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1468         if (v3)
 1469                 nfsm_wcc_data(dvp, wccflag);
 1470         m_freem(mrep);
 1471 nfsmout:
 1472         VTONFS(dvp)->n_flag |= NMODIFIED;
 1473         if (!wccflag)
 1474                 VTONFS(dvp)->n_attrstamp = 0;
 1475         return (error);
 1476 }
 1477 
 1478 /*
 1479  * nfs file rename call
 1480  */
 1481 static int
 1482 nfs_rename(struct vop_rename_args *ap)
 1483 {
 1484         struct vnode *fvp = ap->a_fvp;
 1485         struct vnode *tvp = ap->a_tvp;
 1486         struct vnode *fdvp = ap->a_fdvp;
 1487         struct vnode *tdvp = ap->a_tdvp;
 1488         struct componentname *tcnp = ap->a_tcnp;
 1489         struct componentname *fcnp = ap->a_fcnp;
 1490         int error;
 1491 
 1492 #ifndef DIAGNOSTIC
 1493         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1494             (fcnp->cn_flags & HASBUF) == 0)
 1495                 panic("nfs_rename: no name");
 1496 #endif
 1497         /* Check for cross-device rename */
 1498         if ((fvp->v_mount != tdvp->v_mount) ||
 1499             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1500                 error = EXDEV;
 1501                 goto out;
 1502         }
 1503 
 1504         if (fvp == tvp) {
 1505                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1506                 error = 0;
 1507                 goto out;
 1508         }
 1509         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1510                 goto out;
 1511 
 1512         /*
 1513          * We have to flush B_DELWRI data prior to renaming
 1514          * the file.  If we don't, the delayed-write buffers
 1515          * can be flushed out later after the file has gone stale
 1516          * under NFSV3.  NFSV2 does not have this problem because
 1517          * ( as far as I can tell ) it flushes dirty buffers more
 1518          * often.
 1519          * 
 1520          * Skip the rename operation if the fsync fails, this can happen
 1521          * due to the server's volume being full, when we pushed out data
 1522          * that was written back to our cache earlier. Not checking for
 1523          * this condition can result in potential (silent) data loss.
 1524          */
 1525         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1526         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1527         if (!error && tvp)
 1528                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1529         if (error)
 1530                 goto out;
 1531 
 1532         /*
 1533          * If the tvp exists and is in use, sillyrename it before doing the
 1534          * rename of the new file over it.
 1535          * XXX Can't sillyrename a directory.
 1536          */
 1537         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1538                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1539                 vput(tvp);
 1540                 tvp = NULL;
 1541         }
 1542 
 1543         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1544                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1545                 tcnp->cn_thread);
 1546 
 1547         if (fvp->v_type == VDIR) {
 1548                 if (tvp != NULL && tvp->v_type == VDIR)
 1549                         cache_purge(tdvp);
 1550                 cache_purge(fdvp);
 1551         }
 1552 
 1553 out:
 1554         if (tdvp == tvp)
 1555                 vrele(tdvp);
 1556         else
 1557                 vput(tdvp);
 1558         if (tvp)
 1559                 vput(tvp);
 1560         vrele(fdvp);
 1561         vrele(fvp);
 1562         /*
 1563          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1564          */
 1565         if (error == ENOENT)
 1566                 error = 0;
 1567         return (error);
 1568 }
 1569 
 1570 /*
 1571  * nfs file rename rpc called from nfs_remove() above
 1572  */
 1573 static int
 1574 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1575     struct sillyrename *sp)
 1576 {
 1577 
 1578         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1579             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1580 }
 1581 
 1582 /*
 1583  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1584  */
 1585 static int
 1586 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1587     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1588     struct thread *td)
 1589 {
 1590         caddr_t bpos, dpos;
 1591         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1592         struct mbuf *mreq, *mrep, *md, *mb;
 1593         int v3 = NFS_ISV3(fdvp);
 1594 
 1595         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1596         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1597                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1598                 nfsm_rndup(tnamelen));
 1599         mb = mreq;
 1600         bpos = mtod(mb, caddr_t);
 1601         nfsm_fhtom(fdvp, v3);
 1602         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1603         nfsm_fhtom(tdvp, v3);
 1604         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1605         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1606         if (v3) {
 1607                 nfsm_wcc_data(fdvp, fwccflag);
 1608                 nfsm_wcc_data(tdvp, twccflag);
 1609         }
 1610         m_freem(mrep);
 1611 nfsmout:
 1612         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1613         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1614         if (!fwccflag)
 1615                 VTONFS(fdvp)->n_attrstamp = 0;
 1616         if (!twccflag)
 1617                 VTONFS(tdvp)->n_attrstamp = 0;
 1618         return (error);
 1619 }
 1620 
 1621 /*
 1622  * nfs hard link create call
 1623  */
 1624 static int
 1625 nfs_link(struct vop_link_args *ap)
 1626 {
 1627         struct vnode *vp = ap->a_vp;
 1628         struct vnode *tdvp = ap->a_tdvp;
 1629         struct componentname *cnp = ap->a_cnp;
 1630         caddr_t bpos, dpos;
 1631         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1632         struct mbuf *mreq, *mrep, *md, *mb;
 1633         int v3;
 1634 
 1635         if (vp->v_mount != tdvp->v_mount) {
 1636                 return (EXDEV);
 1637         }
 1638 
 1639         /*
 1640          * Push all writes to the server, so that the attribute cache
 1641          * doesn't get "out of sync" with the server.
 1642          * XXX There should be a better way!
 1643          */
 1644         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1645 
 1646         v3 = NFS_ISV3(vp);
 1647         nfsstats.rpccnt[NFSPROC_LINK]++;
 1648         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1649                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1650         mb = mreq;
 1651         bpos = mtod(mb, caddr_t);
 1652         nfsm_fhtom(vp, v3);
 1653         nfsm_fhtom(tdvp, v3);
 1654         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1655         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1656         if (v3) {
 1657                 nfsm_postop_attr(vp, attrflag);
 1658                 nfsm_wcc_data(tdvp, wccflag);
 1659         }
 1660         m_freem(mrep);
 1661 nfsmout:
 1662         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1663         if (!attrflag)
 1664                 VTONFS(vp)->n_attrstamp = 0;
 1665         if (!wccflag)
 1666                 VTONFS(tdvp)->n_attrstamp = 0;
 1667         /*
 1668          * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
 1669          */
 1670         if (error == EEXIST)
 1671                 error = 0;
 1672         return (error);
 1673 }
 1674 
 1675 /*
 1676  * nfs symbolic link create call
 1677  */
 1678 static int
 1679 nfs_symlink(struct vop_symlink_args *ap)
 1680 {
 1681         struct vnode *dvp = ap->a_dvp;
 1682         struct vattr *vap = ap->a_vap;
 1683         struct componentname *cnp = ap->a_cnp;
 1684         struct nfsv2_sattr *sp;
 1685         caddr_t bpos, dpos;
 1686         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1687         struct mbuf *mreq, *mrep, *md, *mb;
 1688         struct vnode *newvp = NULL;
 1689         int v3 = NFS_ISV3(dvp);
 1690 
 1691         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1692         slen = strlen(ap->a_target);
 1693         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1694             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1695         mb = mreq;
 1696         bpos = mtod(mb, caddr_t);
 1697         nfsm_fhtom(dvp, v3);
 1698         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1699         if (v3) {
 1700                 nfsm_v3attrbuild(vap, FALSE);
 1701         }
 1702         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1703         if (!v3) {
 1704                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1705                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1706                 sp->sa_uid = nfs_xdrneg1;
 1707                 sp->sa_gid = nfs_xdrneg1;
 1708                 sp->sa_size = nfs_xdrneg1;
 1709                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1710                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1711         }
 1712 
 1713         /*
 1714          * Issue the NFS request and get the rpc response.
 1715          *
 1716          * Only NFSv3 responses returning an error of 0 actually return
 1717          * a file handle that can be converted into newvp without having
 1718          * to do an extra lookup rpc.
 1719          */
 1720         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1721         if (v3) {
 1722                 if (error == 0)
 1723                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1724                 nfsm_wcc_data(dvp, wccflag);
 1725         }
 1726 
 1727         /*
 1728          * out code jumps -> here, mrep is also freed.
 1729          */
 1730 
 1731         m_freem(mrep);
 1732 nfsmout:
 1733 
 1734         /*
 1735          * If we get an EEXIST error, silently convert it to no-error
 1736          * in case of an NFS retry.
 1737          */
 1738         if (error == EEXIST)
 1739                 error = 0;
 1740 
 1741         /*
 1742          * If we do not have (or no longer have) an error, and we could
 1743          * not extract the newvp from the response due to the request being
 1744          * NFSv2 or the error being EEXIST.  We have to do a lookup in order
 1745          * to obtain a newvp to return.
 1746          */
 1747         if (error == 0 && newvp == NULL) {
 1748                 struct nfsnode *np = NULL;
 1749 
 1750                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1751                     cnp->cn_cred, cnp->cn_thread, &np);
 1752                 if (!error)
 1753                         newvp = NFSTOV(np);
 1754         }
 1755         if (error) {
 1756                 if (newvp)
 1757                         vput(newvp);
 1758         } else {
 1759                 *ap->a_vpp = newvp;
 1760         }
 1761         VTONFS(dvp)->n_flag |= NMODIFIED;
 1762         if (!wccflag)
 1763                 VTONFS(dvp)->n_attrstamp = 0;
 1764         return (error);
 1765 }
 1766 
 1767 /*
 1768  * nfs make dir call
 1769  */
 1770 static int
 1771 nfs_mkdir(struct vop_mkdir_args *ap)
 1772 {
 1773         struct vnode *dvp = ap->a_dvp;
 1774         struct vattr *vap = ap->a_vap;
 1775         struct componentname *cnp = ap->a_cnp;
 1776         struct nfsv2_sattr *sp;
 1777         int len;
 1778         struct nfsnode *np = NULL;
 1779         struct vnode *newvp = NULL;
 1780         caddr_t bpos, dpos;
 1781         int error = 0, wccflag = NFSV3_WCCRATTR;
 1782         int gotvp = 0;
 1783         struct mbuf *mreq, *mrep, *md, *mb;
 1784         struct vattr vattr;
 1785         int v3 = NFS_ISV3(dvp);
 1786 
 1787         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1788                 return (error);
 1789         }
 1790         len = cnp->cn_namelen;
 1791         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1792         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1793           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1794         mb = mreq;
 1795         bpos = mtod(mb, caddr_t);
 1796         nfsm_fhtom(dvp, v3);
 1797         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1798         if (v3) {
 1799                 nfsm_v3attrbuild(vap, FALSE);
 1800         } else {
 1801                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1802                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1803                 sp->sa_uid = nfs_xdrneg1;
 1804                 sp->sa_gid = nfs_xdrneg1;
 1805                 sp->sa_size = nfs_xdrneg1;
 1806                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1807                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1808         }
 1809         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1810         if (!error)
 1811                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1812         if (v3)
 1813                 nfsm_wcc_data(dvp, wccflag);
 1814         m_freem(mrep);
 1815 nfsmout:
 1816         VTONFS(dvp)->n_flag |= NMODIFIED;
 1817         if (!wccflag)
 1818                 VTONFS(dvp)->n_attrstamp = 0;
 1819         /*
 1820          * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
 1821          * if we can succeed in looking up the directory.
 1822          */
 1823         if (error == EEXIST || (!error && !gotvp)) {
 1824                 if (newvp) {
 1825                         vput(newvp);
 1826                         newvp = NULL;
 1827                 }
 1828                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1829                         cnp->cn_thread, &np);
 1830                 if (!error) {
 1831                         newvp = NFSTOV(np);
 1832                         if (newvp->v_type != VDIR)
 1833                                 error = EEXIST;
 1834                 }
 1835         }
 1836         if (error) {
 1837                 if (newvp)
 1838                         vput(newvp);
 1839         } else
 1840                 *ap->a_vpp = newvp;
 1841         return (error);
 1842 }
 1843 
 1844 /*
 1845  * nfs remove directory call
 1846  */
 1847 static int
 1848 nfs_rmdir(struct vop_rmdir_args *ap)
 1849 {
 1850         struct vnode *vp = ap->a_vp;
 1851         struct vnode *dvp = ap->a_dvp;
 1852         struct componentname *cnp = ap->a_cnp;
 1853         caddr_t bpos, dpos;
 1854         int error = 0, wccflag = NFSV3_WCCRATTR;
 1855         struct mbuf *mreq, *mrep, *md, *mb;
 1856         int v3 = NFS_ISV3(dvp);
 1857 
 1858         if (dvp == vp)
 1859                 return (EINVAL);
 1860         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1861         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1862                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1863         mb = mreq;
 1864         bpos = mtod(mb, caddr_t);
 1865         nfsm_fhtom(dvp, v3);
 1866         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1867         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1868         if (v3)
 1869                 nfsm_wcc_data(dvp, wccflag);
 1870         m_freem(mrep);
 1871 nfsmout:
 1872         VTONFS(dvp)->n_flag |= NMODIFIED;
 1873         if (!wccflag)
 1874                 VTONFS(dvp)->n_attrstamp = 0;
 1875         cache_purge(dvp);
 1876         cache_purge(vp);
 1877         /*
 1878          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1879          */
 1880         if (error == ENOENT)
 1881                 error = 0;
 1882         return (error);
 1883 }
 1884 
 1885 /*
 1886  * nfs readdir call
 1887  */
 1888 static int
 1889 nfs_readdir(struct vop_readdir_args *ap)
 1890 {
 1891         struct vnode *vp = ap->a_vp;
 1892         struct nfsnode *np = VTONFS(vp);
 1893         struct uio *uio = ap->a_uio;
 1894         int tresid, error;
 1895         struct vattr vattr;
 1896 
 1897         if (vp->v_type != VDIR)
 1898                 return (EPERM);
 1899         /*
 1900          * First, check for hit on the EOF offset cache
 1901          */
 1902         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1903             (np->n_flag & NMODIFIED) == 0) {
 1904                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1905                     !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 1906                         nfsstats.direofcache_hits++;
 1907                         return (0);
 1908                 }
 1909         }
 1910 
 1911         /*
 1912          * Call nfs_bioread() to do the real work.
 1913          */
 1914         tresid = uio->uio_resid;
 1915         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1916 
 1917         if (!error && uio->uio_resid == tresid)
 1918                 nfsstats.direofcache_misses++;
 1919         return (error);
 1920 }
 1921 
 1922 /*
 1923  * Readdir rpc call.
 1924  * Called from below the buffer cache by nfs_doio().
 1925  */
 1926 int
 1927 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1928 {
 1929         int len, left;
 1930         struct dirent *dp = NULL;
 1931         u_int32_t *tl;
 1932         caddr_t cp;
 1933         nfsuint64 *cookiep;
 1934         caddr_t bpos, dpos;
 1935         struct mbuf *mreq, *mrep, *md, *mb;
 1936         nfsuint64 cookie;
 1937         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1938         struct nfsnode *dnp = VTONFS(vp);
 1939         u_quad_t fileno;
 1940         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1941         int attrflag;
 1942         int v3 = NFS_ISV3(vp);
 1943 
 1944 #ifndef DIAGNOSTIC
 1945         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1946                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1947                 panic("nfs readdirrpc bad uio");
 1948 #endif
 1949 
 1950         /*
 1951          * If there is no cookie, assume directory was stale.
 1952          */
 1953         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1954         if (cookiep)
 1955                 cookie = *cookiep;
 1956         else
 1957                 return (NFSERR_BAD_COOKIE);
 1958         /*
 1959          * Loop around doing readdir rpc's of size nm_readdirsize
 1960          * truncated to a multiple of DIRBLKSIZ.
 1961          * The stopping criteria is EOF or buffer full.
 1962          */
 1963         while (more_dirs && bigenough) {
 1964                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1965                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1966                         NFSX_READDIR(v3));
 1967                 mb = mreq;
 1968                 bpos = mtod(mb, caddr_t);
 1969                 nfsm_fhtom(vp, v3);
 1970                 if (v3) {
 1971                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1972                         *tl++ = cookie.nfsuquad[0];
 1973                         *tl++ = cookie.nfsuquad[1];
 1974                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 1975                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 1976                 } else {
 1977                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1978                         *tl++ = cookie.nfsuquad[0];
 1979                 }
 1980                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 1981                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 1982                 if (v3) {
 1983                         nfsm_postop_attr(vp, attrflag);
 1984                         if (!error) {
 1985                                 tl = nfsm_dissect(u_int32_t *,
 1986                                     2 * NFSX_UNSIGNED);
 1987                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 1988                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 1989                         } else {
 1990                                 m_freem(mrep);
 1991                                 goto nfsmout;
 1992                         }
 1993                 }
 1994                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 1995                 more_dirs = fxdr_unsigned(int, *tl);
 1996 
 1997                 /* loop thru the dir entries, doctoring them to 4bsd form */
 1998                 while (more_dirs && bigenough) {
 1999                         if (v3) {
 2000                                 tl = nfsm_dissect(u_int32_t *,
 2001                                     3 * NFSX_UNSIGNED);
 2002                                 fileno = fxdr_hyper(tl);
 2003                                 len = fxdr_unsigned(int, *(tl + 2));
 2004                         } else {
 2005                                 tl = nfsm_dissect(u_int32_t *,
 2006                                     2 * NFSX_UNSIGNED);
 2007                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 2008                                 len = fxdr_unsigned(int, *tl);
 2009                         }
 2010                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2011                                 error = EBADRPC;
 2012                                 m_freem(mrep);
 2013                                 goto nfsmout;
 2014                         }
 2015                         tlen = nfsm_rndup(len);
 2016                         if (tlen == len)
 2017                                 tlen += 4;      /* To ensure null termination */
 2018                         left = DIRBLKSIZ - blksiz;
 2019                         if ((tlen + DIRHDSIZ) > left) {
 2020                                 dp->d_reclen += left;
 2021                                 uiop->uio_iov->iov_base =
 2022                                     (char *)uiop->uio_iov->iov_base + left;
 2023                                 uiop->uio_iov->iov_len -= left;
 2024                                 uiop->uio_offset += left;
 2025                                 uiop->uio_resid -= left;
 2026                                 blksiz = 0;
 2027                         }
 2028                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2029                                 bigenough = 0;
 2030                         if (bigenough) {
 2031                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2032                                 dp->d_fileno = (int)fileno;
 2033                                 dp->d_namlen = len;
 2034                                 dp->d_reclen = tlen + DIRHDSIZ;
 2035                                 dp->d_type = DT_UNKNOWN;
 2036                                 blksiz += dp->d_reclen;
 2037                                 if (blksiz == DIRBLKSIZ)
 2038                                         blksiz = 0;
 2039                                 uiop->uio_offset += DIRHDSIZ;
 2040                                 uiop->uio_resid -= DIRHDSIZ;
 2041                                 uiop->uio_iov->iov_base =
 2042                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2043                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2044                                 nfsm_mtouio(uiop, len);
 2045                                 cp = uiop->uio_iov->iov_base;
 2046                                 tlen -= len;
 2047                                 *cp = '\0';     /* null terminate */
 2048                                 uiop->uio_iov->iov_base =
 2049                                     (char *)uiop->uio_iov->iov_base + tlen;
 2050                                 uiop->uio_iov->iov_len -= tlen;
 2051                                 uiop->uio_offset += tlen;
 2052                                 uiop->uio_resid -= tlen;
 2053                         } else
 2054                                 nfsm_adv(nfsm_rndup(len));
 2055                         if (v3) {
 2056                                 tl = nfsm_dissect(u_int32_t *,
 2057                                     3 * NFSX_UNSIGNED);
 2058                         } else {
 2059                                 tl = nfsm_dissect(u_int32_t *,
 2060                                     2 * NFSX_UNSIGNED);
 2061                         }
 2062                         if (bigenough) {
 2063                                 cookie.nfsuquad[0] = *tl++;
 2064                                 if (v3)
 2065                                         cookie.nfsuquad[1] = *tl++;
 2066                         } else if (v3)
 2067                                 tl += 2;
 2068                         else
 2069                                 tl++;
 2070                         more_dirs = fxdr_unsigned(int, *tl);
 2071                 }
 2072                 /*
 2073                  * If at end of rpc data, get the eof boolean
 2074                  */
 2075                 if (!more_dirs) {
 2076                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2077                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2078                 }
 2079                 m_freem(mrep);
 2080         }
 2081         /*
 2082          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2083          * by increasing d_reclen for the last record.
 2084          */
 2085         if (blksiz > 0) {
 2086                 left = DIRBLKSIZ - blksiz;
 2087                 dp->d_reclen += left;
 2088                 uiop->uio_iov->iov_base =
 2089                     (char *)uiop->uio_iov->iov_base + left;
 2090                 uiop->uio_iov->iov_len -= left;
 2091                 uiop->uio_offset += left;
 2092                 uiop->uio_resid -= left;
 2093         }
 2094 
 2095         /*
 2096          * We are now either at the end of the directory or have filled the
 2097          * block.
 2098          */
 2099         if (bigenough)
 2100                 dnp->n_direofoffset = uiop->uio_offset;
 2101         else {
 2102                 if (uiop->uio_resid > 0)
 2103                         printf("EEK! readdirrpc resid > 0\n");
 2104                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2105                 *cookiep = cookie;
 2106         }
 2107 nfsmout:
 2108         return (error);
 2109 }
 2110 
 2111 /*
 2112  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2113  */
 2114 int
 2115 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2116 {
 2117         int len, left;
 2118         struct dirent *dp;
 2119         u_int32_t *tl;
 2120         caddr_t cp;
 2121         struct vnode *newvp;
 2122         nfsuint64 *cookiep;
 2123         caddr_t bpos, dpos, dpossav1, dpossav2;
 2124         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2125         struct nameidata nami, *ndp = &nami;
 2126         struct componentname *cnp = &ndp->ni_cnd;
 2127         nfsuint64 cookie;
 2128         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2129         struct nfsnode *dnp = VTONFS(vp), *np;
 2130         nfsfh_t *fhp;
 2131         u_quad_t fileno;
 2132         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2133         int attrflag, fhsize;
 2134 
 2135 #ifndef nolint
 2136         dp = NULL;
 2137 #endif
 2138 #ifndef DIAGNOSTIC
 2139         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2140                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2141                 panic("nfs readdirplusrpc bad uio");
 2142 #endif
 2143         ndp->ni_dvp = vp;
 2144         newvp = NULLVP;
 2145 
 2146         /*
 2147          * If there is no cookie, assume directory was stale.
 2148          */
 2149         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2150         if (cookiep)
 2151                 cookie = *cookiep;
 2152         else
 2153                 return (NFSERR_BAD_COOKIE);
 2154         /*
 2155          * Loop around doing readdir rpc's of size nm_readdirsize
 2156          * truncated to a multiple of DIRBLKSIZ.
 2157          * The stopping criteria is EOF or buffer full.
 2158          */
 2159         while (more_dirs && bigenough) {
 2160                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2161                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2162                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2163                 mb = mreq;
 2164                 bpos = mtod(mb, caddr_t);
 2165                 nfsm_fhtom(vp, 1);
 2166                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2167                 *tl++ = cookie.nfsuquad[0];
 2168                 *tl++ = cookie.nfsuquad[1];
 2169                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2170                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2171                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2172                 *tl = txdr_unsigned(nmp->nm_rsize);
 2173                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2174                 nfsm_postop_attr(vp, attrflag);
 2175                 if (error) {
 2176                         m_freem(mrep);
 2177                         goto nfsmout;
 2178                 }
 2179                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2180                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2181                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2182                 more_dirs = fxdr_unsigned(int, *tl);
 2183 
 2184                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2185                 while (more_dirs && bigenough) {
 2186                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2187                         fileno = fxdr_hyper(tl);
 2188                         len = fxdr_unsigned(int, *(tl + 2));
 2189                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2190                                 error = EBADRPC;
 2191                                 m_freem(mrep);
 2192                                 goto nfsmout;
 2193                         }
 2194                         tlen = nfsm_rndup(len);
 2195                         if (tlen == len)
 2196                                 tlen += 4;      /* To ensure null termination*/
 2197                         left = DIRBLKSIZ - blksiz;
 2198                         if ((tlen + DIRHDSIZ) > left) {
 2199                                 dp->d_reclen += left;
 2200                                 uiop->uio_iov->iov_base =
 2201                                     (char *)uiop->uio_iov->iov_base + left;
 2202                                 uiop->uio_iov->iov_len -= left;
 2203                                 uiop->uio_offset += left;
 2204                                 uiop->uio_resid -= left;
 2205                                 blksiz = 0;
 2206                         }
 2207                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2208                                 bigenough = 0;
 2209                         if (bigenough) {
 2210                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2211                                 dp->d_fileno = (int)fileno;
 2212                                 dp->d_namlen = len;
 2213                                 dp->d_reclen = tlen + DIRHDSIZ;
 2214                                 dp->d_type = DT_UNKNOWN;
 2215                                 blksiz += dp->d_reclen;
 2216                                 if (blksiz == DIRBLKSIZ)
 2217                                         blksiz = 0;
 2218                                 uiop->uio_offset += DIRHDSIZ;
 2219                                 uiop->uio_resid -= DIRHDSIZ;
 2220                                 uiop->uio_iov->iov_base =
 2221                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2222                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2223                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2224                                 cnp->cn_namelen = len;
 2225                                 nfsm_mtouio(uiop, len);
 2226                                 cp = uiop->uio_iov->iov_base;
 2227                                 tlen -= len;
 2228                                 *cp = '\0';
 2229                                 uiop->uio_iov->iov_base =
 2230                                     (char *)uiop->uio_iov->iov_base + tlen;
 2231                                 uiop->uio_iov->iov_len -= tlen;
 2232                                 uiop->uio_offset += tlen;
 2233                                 uiop->uio_resid -= tlen;
 2234                         } else
 2235                                 nfsm_adv(nfsm_rndup(len));
 2236                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2237                         if (bigenough) {
 2238                                 cookie.nfsuquad[0] = *tl++;
 2239                                 cookie.nfsuquad[1] = *tl++;
 2240                         } else
 2241                                 tl += 2;
 2242 
 2243                         /*
 2244                          * Since the attributes are before the file handle
 2245                          * (sigh), we must skip over the attributes and then
 2246                          * come back and get them.
 2247                          */
 2248                         attrflag = fxdr_unsigned(int, *tl);
 2249                         if (attrflag) {
 2250                             dpossav1 = dpos;
 2251                             mdsav1 = md;
 2252                             nfsm_adv(NFSX_V3FATTR);
 2253                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2254                             doit = fxdr_unsigned(int, *tl);
 2255                             /*
 2256                              * Skip loading the attrs for "..". There's a 
 2257                              * race between loading the attrs here and 
 2258                              * lookups that look for the directory currently
 2259                              * being read (in the parent). We try to acquire
 2260                              * the exclusive lock on ".." here, owning the 
 2261                              * lock on the directory being read. Lookup will
 2262                              * hold the lock on ".." and try to acquire the 
 2263                              * lock on the directory being read.
 2264                              * 
 2265                              * There are other ways of fixing this, one would
 2266                              * be to do a trylock on the ".." vnode and skip
 2267                              * loading the attrs on ".." if it happens to be 
 2268                              * locked by another process. But skipping the
 2269                              * attrload on ".." seems the easiest option.
 2270                              */
 2271                             if (strcmp(dp->d_name, "..") == 0) {
 2272                                     doit = 0;
 2273                                     /*
 2274                                      * We've already skipped over the attrs, 
 2275                                      * skip over the filehandle. And store d_type
 2276                                      * as VDIR.
 2277                                      */
 2278                                     tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2279                                     i = fxdr_unsigned(int, *tl);
 2280                                     nfsm_adv(nfsm_rndup(i));
 2281                                     dp->d_type = IFTODT(VTTOIF(VDIR));
 2282                             }       
 2283                             if (doit) {
 2284                                 nfsm_getfh(fhp, fhsize, 1);
 2285                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2286                                     VREF(vp);
 2287                                     newvp = vp;
 2288                                     np = dnp;
 2289                                 } else {
 2290                                     error = nfs_nget(vp->v_mount, fhp,
 2291                                         fhsize, &np);
 2292                                     if (error)
 2293                                         doit = 0;
 2294                                     else
 2295                                         newvp = NFSTOV(np);
 2296                                 }
 2297                             }
 2298                             if (doit && bigenough) {
 2299                                 dpossav2 = dpos;
 2300                                 dpos = dpossav1;
 2301                                 mdsav2 = md;
 2302                                 md = mdsav1;
 2303                                 nfsm_loadattr(newvp, NULL);
 2304                                 dpos = dpossav2;
 2305                                 md = mdsav2;
 2306                                 dp->d_type =
 2307                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2308                                 ndp->ni_vp = newvp;
 2309                                 /* Update n_ctime, so subsequent lookup doesn't purge entry */
 2310                                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
 2311                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2312                             }
 2313                         } else {
 2314                             /* Just skip over the file handle */
 2315                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2316                             i = fxdr_unsigned(int, *tl);
 2317                             if (i) {
 2318                                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2319                                 fhsize = fxdr_unsigned(int, *tl);
 2320                                 nfsm_adv(nfsm_rndup(fhsize));
 2321                             }
 2322                         }
 2323                         if (newvp != NULLVP) {
 2324                             if (newvp == vp)
 2325                                 vrele(newvp);
 2326                             else
 2327                                 vput(newvp);
 2328                             newvp = NULLVP;
 2329                         }
 2330                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2331                         more_dirs = fxdr_unsigned(int, *tl);
 2332                 }
 2333                 /*
 2334                  * If at end of rpc data, get the eof boolean
 2335                  */
 2336                 if (!more_dirs) {
 2337                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2338                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2339                 }
 2340                 m_freem(mrep);
 2341         }
 2342         /*
 2343          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2344          * by increasing d_reclen for the last record.
 2345          */
 2346         if (blksiz > 0) {
 2347                 left = DIRBLKSIZ - blksiz;
 2348                 dp->d_reclen += left;
 2349                 uiop->uio_iov->iov_base =
 2350                     (char *)uiop->uio_iov->iov_base + left;
 2351                 uiop->uio_iov->iov_len -= left;
 2352                 uiop->uio_offset += left;
 2353                 uiop->uio_resid -= left;
 2354         }
 2355 
 2356         /*
 2357          * We are now either at the end of the directory or have filled the
 2358          * block.
 2359          */
 2360         if (bigenough)
 2361                 dnp->n_direofoffset = uiop->uio_offset;
 2362         else {
 2363                 if (uiop->uio_resid > 0)
 2364                         printf("EEK! readdirplusrpc resid > 0\n");
 2365                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2366                 *cookiep = cookie;
 2367         }
 2368 nfsmout:
 2369         if (newvp != NULLVP) {
 2370                 if (newvp == vp)
 2371                         vrele(newvp);
 2372                 else
 2373                         vput(newvp);
 2374                 newvp = NULLVP;
 2375         }
 2376         return (error);
 2377 }
 2378 
 2379 /*
 2380  * Silly rename. To make the NFS filesystem that is stateless look a little
 2381  * more like the "ufs" a remove of an active vnode is translated to a rename
 2382  * to a funny looking filename that is removed by nfs_inactive on the
 2383  * nfsnode. There is the potential for another process on a different client
 2384  * to create the same funny name between the nfs_lookitup() fails and the
 2385  * nfs_rename() completes, but...
 2386  */
 2387 static int
 2388 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2389 {
 2390         struct sillyrename *sp;
 2391         struct nfsnode *np;
 2392         int error;
 2393         short pid;
 2394         unsigned int lticks;
 2395 
 2396         cache_purge(dvp);
 2397         np = VTONFS(vp);
 2398 #ifndef DIAGNOSTIC
 2399         if (vp->v_type == VDIR)
 2400                 panic("nfs: sillyrename dir");
 2401 #endif
 2402         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2403                 M_NFSREQ, M_WAITOK);
 2404         sp->s_cred = crhold(cnp->cn_cred);
 2405         sp->s_dvp = dvp;
 2406         sp->s_removeit = nfs_removeit;
 2407         VREF(dvp);
 2408 
 2409         /* 
 2410          * Fudge together a funny name.
 2411          * Changing the format of the funny name to accomodate more 
 2412          * sillynames per directory.
 2413          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2414          * CPU ticks since boot.
 2415          */
 2416         pid = cnp->cn_thread->td_proc->p_pid;
 2417         lticks = (unsigned int)ticks;
 2418         for ( ; ; ) {
 2419                 sp->s_namlen = sprintf(sp->s_name, 
 2420                                        ".nfs.%08x.%04x4.4", lticks, 
 2421                                        pid);
 2422                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2423                                  cnp->cn_thread, NULL))
 2424                         break;
 2425                 lticks++;
 2426         }
 2427         error = nfs_renameit(dvp, cnp, sp);
 2428         if (error)
 2429                 goto bad;
 2430         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2431                 cnp->cn_thread, &np);
 2432         np->n_sillyrename = sp;
 2433         return (0);
 2434 bad:
 2435         vrele(sp->s_dvp);
 2436         crfree(sp->s_cred);
 2437         free((caddr_t)sp, M_NFSREQ);
 2438         return (error);
 2439 }
 2440 
 2441 /*
 2442  * Look up a file name and optionally either update the file handle or
 2443  * allocate an nfsnode, depending on the value of npp.
 2444  * npp == NULL  --> just do the lookup
 2445  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2446  *                      handled too
 2447  * *npp != NULL --> update the file handle in the vnode
 2448  */
 2449 static int
 2450 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2451     struct thread *td, struct nfsnode **npp)
 2452 {
 2453         struct vnode *newvp = NULL;
 2454         struct nfsnode *np, *dnp = VTONFS(dvp);
 2455         caddr_t bpos, dpos;
 2456         int error = 0, fhlen, attrflag;
 2457         struct mbuf *mreq, *mrep, *md, *mb;
 2458         nfsfh_t *nfhp;
 2459         int v3 = NFS_ISV3(dvp);
 2460 
 2461         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2462         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2463                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2464         mb = mreq;
 2465         bpos = mtod(mb, caddr_t);
 2466         nfsm_fhtom(dvp, v3);
 2467         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2468         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2469         if (npp && !error) {
 2470                 nfsm_getfh(nfhp, fhlen, v3);
 2471                 if (*npp) {
 2472                     np = *npp;
 2473                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2474                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2475                         np->n_fhp = &np->n_fh;
 2476                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2477                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2478                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2479                     np->n_fhsize = fhlen;
 2480                     newvp = NFSTOV(np);
 2481                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2482                     VREF(dvp);
 2483                     newvp = dvp;
 2484                 } else {
 2485                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
 2486                     if (error) {
 2487                         m_freem(mrep);
 2488                         return (error);
 2489                     }
 2490                     newvp = NFSTOV(np);
 2491                 }
 2492                 if (v3) {
 2493                         nfsm_postop_attr(newvp, attrflag);
 2494                         if (!attrflag && *npp == NULL) {
 2495                                 m_freem(mrep);
 2496                                 if (newvp == dvp)
 2497                                         vrele(newvp);
 2498                                 else
 2499                                         vput(newvp);
 2500                                 return (ENOENT);
 2501                         }
 2502                 } else
 2503                         nfsm_loadattr(newvp, NULL);
 2504         }
 2505         m_freem(mrep);
 2506 nfsmout:
 2507         if (npp && *npp == NULL) {
 2508                 if (error) {
 2509                         if (newvp) {
 2510                                 if (newvp == dvp)
 2511                                         vrele(newvp);
 2512                                 else
 2513                                         vput(newvp);
 2514                         }
 2515                 } else
 2516                         *npp = np;
 2517         }
 2518         return (error);
 2519 }
 2520 
 2521 /*
 2522  * Nfs Version 3 commit rpc
 2523  */
 2524 int
 2525 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2526     struct thread *td)
 2527 {
 2528         u_int32_t *tl;
 2529         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2530         caddr_t bpos, dpos;
 2531         int error = 0, wccflag = NFSV3_WCCRATTR;
 2532         struct mbuf *mreq, *mrep, *md, *mb;
 2533 
 2534         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2535                 return (0);
 2536         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2537         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2538         mb = mreq;
 2539         bpos = mtod(mb, caddr_t);
 2540         nfsm_fhtom(vp, 1);
 2541         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2542         txdr_hyper(offset, tl);
 2543         tl += 2;
 2544         *tl = txdr_unsigned(cnt);
 2545         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2546         nfsm_wcc_data(vp, wccflag);
 2547         if (!error) {
 2548                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2549                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2550                         NFSX_V3WRITEVERF)) {
 2551                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2552                                 NFSX_V3WRITEVERF);
 2553                         error = NFSERR_STALEWRITEVERF;
 2554                 }
 2555         }
 2556         m_freem(mrep);
 2557 nfsmout:
 2558         return (error);
 2559 }
 2560 
 2561 /*
 2562  * Strategy routine.
 2563  * For async requests when nfsiod(s) are running, queue the request by
 2564  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2565  * request.
 2566  */
 2567 static int
 2568 nfs_strategy(struct vop_strategy_args *ap)
 2569 {
 2570         struct buf *bp = ap->a_bp;
 2571         struct ucred *cr;
 2572 
 2573         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2574         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2575 
 2576         if (bp->b_iocmd == BIO_READ)
 2577                 cr = bp->b_rcred;
 2578         else
 2579                 cr = bp->b_wcred;
 2580 
 2581         /*
 2582          * If the op is asynchronous and an i/o daemon is waiting
 2583          * queue the request, wake it up and wait for completion
 2584          * otherwise just do it ourselves.
 2585          */
 2586         if ((bp->b_flags & B_ASYNC) == 0 ||
 2587             nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
 2588                 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
 2589         return (0);
 2590 }
 2591 
 2592 /*
 2593  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2594  */
 2595 /* ARGSUSED */
 2596 static int
 2597 nfs_fsync(struct vop_fsync_args *ap)
 2598 {
 2599 
 2600         return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
 2601 }
 2602 
 2603 /*
 2604  * Flush all the blocks associated with a vnode.
 2605  *      Walk through the buffer pool and push any dirty pages
 2606  *      associated with the vnode.
 2607  */
 2608 static int
 2609 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
 2610     int commit)
 2611 {
 2612         struct nfsnode *np = VTONFS(vp);
 2613         struct buf *bp;
 2614         int i;
 2615         struct buf *nbp;
 2616         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2617         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2618         int passone = 1;
 2619         u_quad_t off, endoff, toff;
 2620         struct ucred* wcred = NULL;
 2621         struct buf **bvec = NULL;
 2622 #ifndef NFS_COMMITBVECSIZ
 2623 #define NFS_COMMITBVECSIZ       20
 2624 #endif
 2625         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2626         int bvecsize = 0, bveccount;
 2627 
 2628         if (nmp->nm_flag & NFSMNT_INT)
 2629                 slpflag = PCATCH;
 2630         if (!commit)
 2631                 passone = 0;
 2632         /*
 2633          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2634          * server, but has not been committed to stable storage on the server
 2635          * yet. On the first pass, the byte range is worked out and the commit
 2636          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2637          * job.
 2638          */
 2639 again:
 2640         off = (u_quad_t)-1;
 2641         endoff = 0;
 2642         bvecpos = 0;
 2643         if (NFS_ISV3(vp) && commit) {
 2644                 s = splbio();
 2645                 if (bvec != NULL && bvec != bvec_on_stack)
 2646                         free(bvec, M_TEMP);
 2647                 /*
 2648                  * Count up how many buffers waiting for a commit.
 2649                  */
 2650                 bveccount = 0;
 2651                 VI_LOCK(vp);
 2652                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2653                         if (BUF_REFCNT(bp) == 0 &&
 2654                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2655                                 == (B_DELWRI | B_NEEDCOMMIT))
 2656                                 bveccount++;
 2657                 }
 2658                 /*
 2659                  * Allocate space to remember the list of bufs to commit.  It is
 2660                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2661                  * If we can't get memory (for whatever reason), we will end up
 2662                  * committing the buffers one-by-one in the loop below.
 2663                  */
 2664                 if (bveccount > NFS_COMMITBVECSIZ) {
 2665                         /*
 2666                          * Release the vnode interlock to avoid a lock
 2667                          * order reversal.
 2668                          */
 2669                         VI_UNLOCK(vp);
 2670                         bvec = (struct buf **)
 2671                                 malloc(bveccount * sizeof(struct buf *),
 2672                                        M_TEMP, M_NOWAIT);
 2673                         VI_LOCK(vp);
 2674                         if (bvec == NULL) {
 2675                                 bvec = bvec_on_stack;
 2676                                 bvecsize = NFS_COMMITBVECSIZ;
 2677                         } else
 2678                                 bvecsize = bveccount;
 2679                 } else {
 2680                         bvec = bvec_on_stack;
 2681                         bvecsize = NFS_COMMITBVECSIZ;
 2682                 }
 2683                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2684                         if (bvecpos >= bvecsize)
 2685                                 break;
 2686                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2687                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2688                                 continue;
 2689                         }
 2690                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2691                             (B_DELWRI | B_NEEDCOMMIT)) {
 2692                                 BUF_UNLOCK(bp);
 2693                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2694                                 continue;
 2695                         }
 2696                         VI_UNLOCK(vp);
 2697                         bremfree(bp);
 2698                         /*
 2699                          * Work out if all buffers are using the same cred
 2700                          * so we can deal with them all with one commit.
 2701                          *
 2702                          * NOTE: we are not clearing B_DONE here, so we have
 2703                          * to do it later on in this routine if we intend to
 2704                          * initiate I/O on the bp.
 2705                          *
 2706                          * Note: to avoid loopback deadlocks, we do not
 2707                          * assign b_runningbufspace.
 2708                          */
 2709                         if (wcred == NULL)
 2710                                 wcred = bp->b_wcred;
 2711                         else if (wcred != bp->b_wcred)
 2712                                 wcred = NOCRED;
 2713                         vfs_busy_pages(bp, 1);
 2714 
 2715                         VI_LOCK(vp);
 2716                         /*
 2717                          * bp is protected by being locked, but nbp is not
 2718                          * and vfs_busy_pages() may sleep.  We have to
 2719                          * recalculate nbp.
 2720                          */
 2721                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2722 
 2723                         /*
 2724                          * A list of these buffers is kept so that the
 2725                          * second loop knows which buffers have actually
 2726                          * been committed. This is necessary, since there
 2727                          * may be a race between the commit rpc and new
 2728                          * uncommitted writes on the file.
 2729                          */
 2730                         bvec[bvecpos++] = bp;
 2731                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2732                                 bp->b_dirtyoff;
 2733                         if (toff < off)
 2734                                 off = toff;
 2735                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2736                         if (toff > endoff)
 2737                                 endoff = toff;
 2738                 }
 2739                 splx(s);
 2740                 VI_UNLOCK(vp);
 2741         }
 2742         if (bvecpos > 0) {
 2743                 /*
 2744                  * Commit data on the server, as required.
 2745                  * If all bufs are using the same wcred, then use that with
 2746                  * one call for all of them, otherwise commit each one
 2747                  * separately.
 2748                  */
 2749                 if (wcred != NOCRED)
 2750                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2751                                           wcred, td);
 2752                 else {
 2753                         retv = 0;
 2754                         for (i = 0; i < bvecpos; i++) {
 2755                                 off_t off, size;
 2756                                 bp = bvec[i];
 2757                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2758                                         bp->b_dirtyoff;
 2759                                 size = (u_quad_t)(bp->b_dirtyend
 2760                                                   - bp->b_dirtyoff);
 2761                                 retv = nfs_commit(vp, off, (int)size,
 2762                                                   bp->b_wcred, td);
 2763                                 if (retv) break;
 2764                         }
 2765                 }
 2766 
 2767                 if (retv == NFSERR_STALEWRITEVERF)
 2768                         nfs_clearcommit(vp->v_mount);
 2769 
 2770                 /*
 2771                  * Now, either mark the blocks I/O done or mark the
 2772                  * blocks dirty, depending on whether the commit
 2773                  * succeeded.
 2774                  */
 2775                 for (i = 0; i < bvecpos; i++) {
 2776                         bp = bvec[i];
 2777                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2778                         if (retv) {
 2779                                 /*
 2780                                  * Error, leave B_DELWRI intact
 2781                                  */
 2782                                 vfs_unbusy_pages(bp);
 2783                                 brelse(bp);
 2784                         } else {
 2785                                 /*
 2786                                  * Success, remove B_DELWRI ( bundirty() ).
 2787                                  *
 2788                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2789                                  * specific.  We should probably move that
 2790                                  * into bundirty(). XXX
 2791                                  */
 2792                                 s = splbio();
 2793                                 bufobj_wref(&vp->v_bufobj);
 2794                                 bp->b_flags |= B_ASYNC;
 2795                                 bundirty(bp);
 2796                                 bp->b_flags &= ~B_DONE;
 2797                                 bp->b_ioflags &= ~BIO_ERROR;
 2798                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2799                                 splx(s);
 2800                                 bufdone(bp);
 2801                         }
 2802                 }
 2803         }
 2804 
 2805         /*
 2806          * Start/do any write(s) that are required.
 2807          */
 2808 loop:
 2809         s = splbio();
 2810         VI_LOCK(vp);
 2811         TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2812                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2813                         if (waitfor != MNT_WAIT || passone)
 2814                                 continue;
 2815 
 2816                         error = BUF_TIMELOCK(bp,
 2817                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2818                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2819                         splx(s);
 2820                         if (error == 0)
 2821                                 panic("nfs_fsync: inconsistent lock");
 2822                         if (error == ENOLCK)
 2823                                 goto loop;
 2824                         if (nfs_sigintr(nmp, NULL, td)) {
 2825                                 error = EINTR;
 2826                                 goto done;
 2827                         }
 2828                         if (slpflag == PCATCH) {
 2829                                 slpflag = 0;
 2830                                 slptimeo = 2 * hz;
 2831                         }
 2832                         goto loop;
 2833                 }
 2834                 if ((bp->b_flags & B_DELWRI) == 0)
 2835                         panic("nfs_fsync: not dirty");
 2836                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2837                         BUF_UNLOCK(bp);
 2838                         continue;
 2839                 }
 2840                 VI_UNLOCK(vp);
 2841                 bremfree(bp);
 2842                 if (passone || !commit)
 2843                     bp->b_flags |= B_ASYNC;
 2844                 else
 2845                     bp->b_flags |= B_ASYNC;
 2846                 splx(s);
 2847                 bwrite(bp);
 2848                 if (nfs_sigintr(nmp, NULL, td)) {
 2849                         error = EINTR;
 2850                         goto done;
 2851                 }
 2852                 goto loop;
 2853         }
 2854         splx(s);
 2855         if (passone) {
 2856                 passone = 0;
 2857                 VI_UNLOCK(vp);
 2858                 goto again;
 2859         }
 2860         if (waitfor == MNT_WAIT) {
 2861                 while (vp->v_bufobj.bo_numoutput) {
 2862                         error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
 2863                         if (error) {
 2864                             VI_UNLOCK(vp);
 2865                             error = nfs_sigintr(nmp, NULL, td);
 2866                             if (error)
 2867                                 goto done;
 2868                             if (slpflag == PCATCH) {
 2869                                 slpflag = 0;
 2870                                 slptimeo = 2 * hz;
 2871                             }
 2872                             VI_LOCK(vp);
 2873                         }
 2874                 }
 2875                 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
 2876                         VI_UNLOCK(vp);
 2877                         goto loop;
 2878                 }
 2879         }
 2880         VI_UNLOCK(vp);
 2881         if (np->n_flag & NWRITEERR) {
 2882                 error = np->n_error;
 2883                 np->n_flag &= ~NWRITEERR;
 2884         }
 2885         if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0)
 2886                 np->n_flag &= ~NMODIFIED;
 2887 done:
 2888         if (bvec != NULL && bvec != bvec_on_stack)
 2889                 free(bvec, M_TEMP);
 2890         return (error);
 2891 }
 2892 
 2893 /*
 2894  * NFS advisory byte-level locks.
 2895  */
 2896 static int
 2897 nfs_advlock(struct vop_advlock_args *ap)
 2898 {
 2899 
 2900         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2901                 struct nfsnode *np = VTONFS(ap->a_vp);
 2902 
 2903                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2904         }
 2905         return (nfs_dolock(ap));
 2906 }
 2907 
 2908 /*
 2909  * Print out the contents of an nfsnode.
 2910  */
 2911 static int
 2912 nfs_print(struct vop_print_args *ap)
 2913 {
 2914         struct vnode *vp = ap->a_vp;
 2915         struct nfsnode *np = VTONFS(vp);
 2916 
 2917         printf("\tfileid %ld fsid 0x%x",
 2918            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2919         if (vp->v_type == VFIFO)
 2920                 fifo_printinfo(vp);
 2921         printf("\n");
 2922         return (0);
 2923 }
 2924 
 2925 /*
 2926  * This is the "real" nfs::bwrite(struct buf*).
 2927  * We set B_CACHE if this is a VMIO buffer.
 2928  */
 2929 int
 2930 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
 2931 {
 2932         int s;
 2933         int oldflags = bp->b_flags;
 2934 #if 0
 2935         int retv = 1;
 2936         off_t off;
 2937 #endif
 2938 
 2939         if (BUF_REFCNT(bp) == 0)
 2940                 panic("bwrite: buffer is not locked???");
 2941 
 2942         if (bp->b_flags & B_INVAL) {
 2943                 brelse(bp);
 2944                 return(0);
 2945         }
 2946 
 2947         bp->b_flags |= B_CACHE;
 2948 
 2949         /*
 2950          * Undirty the bp.  We will redirty it later if the I/O fails.
 2951          */
 2952 
 2953         s = splbio();
 2954         bundirty(bp);
 2955         bp->b_flags &= ~B_DONE;
 2956         bp->b_ioflags &= ~BIO_ERROR;
 2957         bp->b_iocmd = BIO_WRITE;
 2958 
 2959         bufobj_wref(bp->b_bufobj);
 2960         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2961         splx(s);
 2962 
 2963         /*
 2964          * Note: to avoid loopback deadlocks, we do not
 2965          * assign b_runningbufspace.
 2966          */
 2967         vfs_busy_pages(bp, 1);
 2968 
 2969         BUF_KERNPROC(bp);
 2970         bp->b_iooffset = dbtob(bp->b_blkno);
 2971         bstrategy(bp);
 2972 
 2973         if( (oldflags & B_ASYNC) == 0) {
 2974                 int rtval = bufwait(bp);
 2975 
 2976                 if (oldflags & B_DELWRI) {
 2977                         s = splbio();
 2978                         reassignbuf(bp);
 2979                         splx(s);
 2980                 }
 2981 
 2982                 brelse(bp);
 2983                 return (rtval);
 2984         }
 2985 
 2986         return (0);
 2987 }
 2988 
 2989 /*
 2990  * nfs special file access vnode op.
 2991  * Essentially just get vattr and then imitate iaccess() since the device is
 2992  * local to the client.
 2993  */
 2994 static int
 2995 nfsspec_access(struct vop_access_args *ap)
 2996 {
 2997         struct vattr *vap;
 2998         struct ucred *cred = ap->a_cred;
 2999         struct vnode *vp = ap->a_vp;
 3000         mode_t mode = ap->a_mode;
 3001         struct vattr vattr;
 3002         int error;
 3003 
 3004         /*
 3005          * Disallow write attempts on filesystems mounted read-only;
 3006          * unless the file is a socket, fifo, or a block or character
 3007          * device resident on the filesystem.
 3008          */
 3009         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3010                 switch (vp->v_type) {
 3011                 case VREG:
 3012                 case VDIR:
 3013                 case VLNK:
 3014                         return (EROFS);
 3015                 default:
 3016                         break;
 3017                 }
 3018         }
 3019         vap = &vattr;
 3020         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3021         if (error)
 3022                 return (error);
 3023         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3024             mode, cred, NULL));
 3025 }
 3026 
 3027 /*
 3028  * Read wrapper for fifos.
 3029  */
 3030 static int
 3031 nfsfifo_read(struct vop_read_args *ap)
 3032 {
 3033         struct nfsnode *np = VTONFS(ap->a_vp);
 3034 
 3035         /*
 3036          * Set access flag.
 3037          */
 3038         np->n_flag |= NACC;
 3039         getnanotime(&np->n_atim);
 3040         return (fifo_specops.vop_read(ap));
 3041 }
 3042 
 3043 /*
 3044  * Write wrapper for fifos.
 3045  */
 3046 static int
 3047 nfsfifo_write(struct vop_write_args *ap)
 3048 {
 3049         struct nfsnode *np = VTONFS(ap->a_vp);
 3050 
 3051         /*
 3052          * Set update flag.
 3053          */
 3054         np->n_flag |= NUPD;
 3055         getnanotime(&np->n_mtim);
 3056         return (fifo_specops.vop_write(ap));
 3057 }
 3058 
 3059 /*
 3060  * Close wrapper for fifos.
 3061  *
 3062  * Update the times on the nfsnode then do fifo close.
 3063  */
 3064 static int
 3065 nfsfifo_close(struct vop_close_args *ap)
 3066 {
 3067         struct vnode *vp = ap->a_vp;
 3068         struct nfsnode *np = VTONFS(vp);
 3069         struct vattr vattr;
 3070         struct timespec ts;
 3071 
 3072         if (np->n_flag & (NACC | NUPD)) {
 3073                 getnanotime(&ts);
 3074                 if (np->n_flag & NACC)
 3075                         np->n_atim = ts;
 3076                 if (np->n_flag & NUPD)
 3077                         np->n_mtim = ts;
 3078                 np->n_flag |= NCHG;
 3079                 if (vrefcnt(vp) == 1 &&
 3080                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3081                         VATTR_NULL(&vattr);
 3082                         if (np->n_flag & NACC)
 3083                                 vattr.va_atime = np->n_atim;
 3084                         if (np->n_flag & NUPD)
 3085                                 vattr.va_mtime = np->n_mtim;
 3086                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3087                 }
 3088         }
 3089         return (fifo_specops.vop_close(ap));
 3090 }
 3091 
 3092 /*
 3093  * Just call nfs_writebp() with the force argument set to 1.
 3094  *
 3095  * NOTE: B_DONE may or may not be set in a_bp on call.
 3096  */
 3097 static int
 3098 nfs_bwrite(struct buf *bp)
 3099 {
 3100 
 3101         return (nfs_writebp(bp, 1, curthread));
 3102 }
 3103 
 3104 struct buf_ops buf_ops_nfs = {
 3105         .bop_name       =       "buf_ops_nfs",
 3106         .bop_write      =       nfs_bwrite,
 3107         .bop_strategy   =       bufstrategy,
 3108         .bop_sync       =       bufsync,
 3109 };

Cache object: ad82e5e9d114a3fa5efd1daa0c6c80bc


[ 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.