The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/nfsclient/nfs_vnops.c

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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Rick Macklem at The University of Guelph.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: releng/6.4/sys/nfsclient/nfs_vnops.c 178459 2008-04-24 10:46:25Z dfr $");
   37 
   38 /*
   39  * vnode op calls for Sun NFS version 2 and 3
   40  */
   41 
   42 #include "opt_inet.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/kernel.h>
   46 #include <sys/systm.h>
   47 #include <sys/resourcevar.h>
   48 #include <sys/proc.h>
   49 #include <sys/mount.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/malloc.h>
   53 #include <sys/mbuf.h>
   54 #include <sys/namei.h>
   55 #include <sys/socket.h>
   56 #include <sys/vnode.h>
   57 #include <sys/dirent.h>
   58 #include <sys/fcntl.h>
   59 #include <sys/lockf.h>
   60 #include <sys/stat.h>
   61 #include <sys/sysctl.h>
   62 #include <sys/signalvar.h>
   63 
   64 #include <vm/vm.h>
   65 #include <vm/vm_object.h>
   66 #include <vm/vm_extern.h>
   67 #include <vm/vm_object.h>
   68 
   69 #include <fs/fifofs/fifo.h>
   70 
   71 #include <rpc/rpcclnt.h>
   72 
   73 #include <nfs/rpcv2.h>
   74 #include <nfs/nfsproto.h>
   75 #include <nfsclient/nfs.h>
   76 #include <nfsclient/nfsnode.h>
   77 #include <nfsclient/nfsmount.h>
   78 #include <nfsclient/nfs_lock.h>
   79 #include <nfs/xdr_subs.h>
   80 #include <nfsclient/nfsm_subs.h>
   81 
   82 #include <net/if.h>
   83 #include <netinet/in.h>
   84 #include <netinet/in_var.h>
   85 
   86 /* Defs */
   87 #define TRUE    1
   88 #define FALSE   0
   89 
   90 /*
   91  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   92  * calls are not in getblk() and brelse() so that they would not be necessary
   93  * here.
   94  */
   95 #ifndef B_VMIO
   96 #define vfs_busy_pages(bp, f)
   97 #endif
   98 
   99 static vop_read_t       nfsfifo_read;
  100 static vop_write_t      nfsfifo_write;
  101 static vop_close_t      nfsfifo_close;
  102 static int      nfs_flush(struct vnode *, int, struct thread *,
  103                     int);
  104 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  105                     struct thread *);
  106 static vop_lookup_t     nfs_lookup;
  107 static vop_create_t     nfs_create;
  108 static vop_mknod_t      nfs_mknod;
  109 static vop_open_t       nfs_open;
  110 static vop_close_t      nfs_close;
  111 static vop_access_t     nfs_access;
  112 static vop_getattr_t    nfs_getattr;
  113 static vop_setattr_t    nfs_setattr;
  114 static vop_read_t       nfs_read;
  115 static vop_fsync_t      nfs_fsync;
  116 static vop_remove_t     nfs_remove;
  117 static vop_link_t       nfs_link;
  118 static vop_rename_t     nfs_rename;
  119 static vop_mkdir_t      nfs_mkdir;
  120 static vop_rmdir_t      nfs_rmdir;
  121 static vop_symlink_t    nfs_symlink;
  122 static vop_readdir_t    nfs_readdir;
  123 static vop_strategy_t   nfs_strategy;
  124 static  int     nfs_lookitup(struct vnode *, const char *, int,
  125                     struct ucred *, struct thread *, struct nfsnode **);
  126 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  127                     struct componentname *);
  128 static vop_access_t     nfsspec_access;
  129 static vop_readlink_t   nfs_readlink;
  130 static vop_print_t      nfs_print;
  131 static vop_advlock_t    nfs_advlock;
  132 static vop_advlockasync_t nfs_advlockasync;
  133 
  134 /*
  135  * Global vfs data structures for nfs
  136  */
  137 struct vop_vector nfs_vnodeops = {
  138         .vop_default =          &default_vnodeops,
  139         .vop_access =           nfs_access,
  140         .vop_advlock =          nfs_advlock,
  141         .vop_advlockasync =     nfs_advlockasync,
  142         .vop_close =            nfs_close,
  143         .vop_create =           nfs_create,
  144         .vop_fsync =            nfs_fsync,
  145         .vop_getattr =          nfs_getattr,
  146         .vop_getpages =         nfs_getpages,
  147         .vop_putpages =         nfs_putpages,
  148         .vop_inactive =         nfs_inactive,
  149         .vop_lease =            VOP_NULL,
  150         .vop_link =             nfs_link,
  151         .vop_lookup =           nfs_lookup,
  152         .vop_mkdir =            nfs_mkdir,
  153         .vop_mknod =            nfs_mknod,
  154         .vop_open =             nfs_open,
  155         .vop_print =            nfs_print,
  156         .vop_read =             nfs_read,
  157         .vop_readdir =          nfs_readdir,
  158         .vop_readlink =         nfs_readlink,
  159         .vop_reclaim =          nfs_reclaim,
  160         .vop_remove =           nfs_remove,
  161         .vop_rename =           nfs_rename,
  162         .vop_rmdir =            nfs_rmdir,
  163         .vop_setattr =          nfs_setattr,
  164         .vop_strategy =         nfs_strategy,
  165         .vop_symlink =          nfs_symlink,
  166         .vop_write =            nfs_write,
  167 };
  168 
  169 struct vop_vector nfs_fifoops = {
  170         .vop_default =          &fifo_specops,
  171         .vop_access =           nfsspec_access,
  172         .vop_close =            nfsfifo_close,
  173         .vop_fsync =            nfs_fsync,
  174         .vop_getattr =          nfs_getattr,
  175         .vop_inactive =         nfs_inactive,
  176         .vop_print =            nfs_print,
  177         .vop_read =             nfsfifo_read,
  178         .vop_reclaim =          nfs_reclaim,
  179         .vop_setattr =          nfs_setattr,
  180         .vop_write =            nfsfifo_write,
  181 };
  182 
  183 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  184                              struct componentname *cnp, struct vattr *vap);
  185 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  186                               struct ucred *cred, struct thread *td);
  187 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  188                               int fnamelen, struct vnode *tdvp,
  189                               const char *tnameptr, int tnamelen,
  190                               struct ucred *cred, struct thread *td);
  191 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  192                              struct sillyrename *sp);
  193 
  194 /*
  195  * Global variables
  196  */
  197 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  198 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  199 int              nfs_numasync = 0;
  200 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  201 
  202 SYSCTL_DECL(_vfs_nfs);
  203 
  204 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  205 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  206            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  207 
  208 static int      nfsv3_commit_on_close = 0;
  209 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  210            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  211 
  212 static int      nfs_clean_pages_on_close = 1;
  213 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  214            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  215 
  216 int nfs_directio_enable = 0;
  217 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  218            &nfs_directio_enable, 0, "Enable NFS directio");
  219 
  220 /*
  221  * This sysctl allows other processes to mmap a file that has been opened
  222  * O_DIRECT by a process.  In general, having processes mmap the file while
  223  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  224  * this by default to prevent DoS attacks - to prevent a malicious user from
  225  * opening up files O_DIRECT preventing other users from mmap'ing these
  226  * files.  "Protected" environments where stricter consistency guarantees are
  227  * required can disable this knob.  The process that opened the file O_DIRECT
  228  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  229  * meaningful.
  230  */
  231 int nfs_directio_allow_mmap = 1;
  232 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  233            &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  234 
  235 #if 0
  236 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  237            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  238 
  239 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  240            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  241 #endif
  242 
  243 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  244                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  245                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  246 static int
  247 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  248     struct ucred *cred)
  249 {
  250         const int v3 = 1;
  251         u_int32_t *tl;
  252         int error = 0, attrflag;
  253 
  254         struct mbuf *mreq, *mrep, *md, *mb;
  255         caddr_t bpos, dpos;
  256         u_int32_t rmode;
  257         struct nfsnode *np = VTONFS(vp);
  258 
  259         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  260         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  261         mb = mreq;
  262         bpos = mtod(mb, caddr_t);
  263         nfsm_fhtom(vp, v3);
  264         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  265         *tl = txdr_unsigned(wmode);
  266         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  267         nfsm_postop_attr(vp, attrflag);
  268         if (!error) {
  269                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  270                 rmode = fxdr_unsigned(u_int32_t, *tl);
  271                 np->n_mode = rmode;
  272                 np->n_modeuid = cred->cr_uid;
  273                 np->n_modestamp = time_second;
  274         }
  275         m_freem(mrep);
  276 nfsmout:
  277         return (error);
  278 }
  279 
  280 /*
  281  * nfs access vnode op.
  282  * For nfs version 2, just return ok. File accesses may fail later.
  283  * For nfs version 3, use the access rpc to check accessibility. If file modes
  284  * are changed on the server, accesses might still fail later.
  285  */
  286 static int
  287 nfs_access(struct vop_access_args *ap)
  288 {
  289         struct vnode *vp = ap->a_vp;
  290         int error = 0;
  291         u_int32_t mode, wmode;
  292         int v3 = NFS_ISV3(vp);
  293         struct nfsnode *np = VTONFS(vp);
  294 
  295         /*
  296          * Disallow write attempts on filesystems mounted read-only;
  297          * unless the file is a socket, fifo, or a block or character
  298          * device resident on the filesystem.
  299          */
  300         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  301                 switch (vp->v_type) {
  302                 case VREG:
  303                 case VDIR:
  304                 case VLNK:
  305                         return (EROFS);
  306                 default:
  307                         break;
  308                 }
  309         }
  310         /*
  311          * For nfs v3, check to see if we have done this recently, and if
  312          * so return our cached result instead of making an ACCESS call.
  313          * If not, do an access rpc, otherwise you are stuck emulating
  314          * ufs_access() locally using the vattr. This may not be correct,
  315          * since the server may apply other access criteria such as
  316          * client uid-->server uid mapping that we do not know about.
  317          */
  318         if (v3) {
  319                 if (ap->a_mode & VREAD)
  320                         mode = NFSV3ACCESS_READ;
  321                 else
  322                         mode = 0;
  323                 if (vp->v_type != VDIR) {
  324                         if (ap->a_mode & VWRITE)
  325                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  326                         if (ap->a_mode & VEXEC)
  327                                 mode |= NFSV3ACCESS_EXECUTE;
  328                 } else {
  329                         if (ap->a_mode & VWRITE)
  330                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  331                                          NFSV3ACCESS_DELETE);
  332                         if (ap->a_mode & VEXEC)
  333                                 mode |= NFSV3ACCESS_LOOKUP;
  334                 }
  335                 /* XXX safety belt, only make blanket request if caching */
  336                 if (nfsaccess_cache_timeout > 0) {
  337                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  338                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  339                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  340                 } else {
  341                         wmode = mode;
  342                 }
  343 
  344                 /*
  345                  * Does our cached result allow us to give a definite yes to
  346                  * this request?
  347                  */
  348                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  349                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  350                     ((np->n_mode & mode) == mode)) {
  351                         nfsstats.accesscache_hits++;
  352                 } else {
  353                         /*
  354                          * Either a no, or a don't know.  Go to the wire.
  355                          */
  356                         nfsstats.accesscache_misses++;
  357                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  358                         if (!error) {
  359                                 if ((np->n_mode & mode) != mode) {
  360                                         error = EACCES;
  361                                 }
  362                         }
  363                 }
  364                 return (error);
  365         } else {
  366                 if ((error = nfsspec_access(ap)) != 0)
  367                         return (error);
  368 
  369                 /*
  370                  * Attempt to prevent a mapped root from accessing a file
  371                  * which it shouldn't.  We try to read a byte from the file
  372                  * if the user is root and the file is not zero length.
  373                  * After calling nfsspec_access, we should have the correct
  374                  * file size cached.
  375                  */
  376                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  377                     && VTONFS(vp)->n_size > 0) {
  378                         struct iovec aiov;
  379                         struct uio auio;
  380                         char buf[1];
  381 
  382                         aiov.iov_base = buf;
  383                         aiov.iov_len = 1;
  384                         auio.uio_iov = &aiov;
  385                         auio.uio_iovcnt = 1;
  386                         auio.uio_offset = 0;
  387                         auio.uio_resid = 1;
  388                         auio.uio_segflg = UIO_SYSSPACE;
  389                         auio.uio_rw = UIO_READ;
  390                         auio.uio_td = ap->a_td;
  391 
  392                         if (vp->v_type == VREG)
  393                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  394                         else if (vp->v_type == VDIR) {
  395                                 char* bp;
  396                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  397                                 aiov.iov_base = bp;
  398                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  399                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  400                                 free(bp, M_TEMP);
  401                         } else if (vp->v_type == VLNK)
  402                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  403                         else
  404                                 error = EACCES;
  405                 }
  406                 return (error);
  407         }
  408 }
  409 
  410 /*
  411  * nfs open vnode op
  412  * Check to see if the type is ok
  413  * and that deletion is not in progress.
  414  * For paged in text files, you will need to flush the page cache
  415  * if consistency is lost.
  416  */
  417 /* ARGSUSED */
  418 static int
  419 nfs_open(struct vop_open_args *ap)
  420 {
  421         struct vnode *vp = ap->a_vp;
  422         struct nfsnode *np = VTONFS(vp);
  423         struct vattr vattr;
  424         int error;
  425         int fmode = ap->a_mode;
  426 
  427         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  428                 return (EOPNOTSUPP);
  429 
  430         /*
  431          * Get a valid lease. If cached data is stale, flush it.
  432          */
  433         if (np->n_flag & NMODIFIED) {
  434                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  435                 if (error == EINTR || error == EIO)
  436                         return (error);
  437                 np->n_attrstamp = 0;
  438                 if (vp->v_type == VDIR)
  439                         np->n_direofoffset = 0;
  440                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  441                 if (error)
  442                         return (error);
  443                 np->n_mtime = vattr.va_mtime;
  444         } else {
  445                 np->n_attrstamp = 0;
  446                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  447                 if (error)
  448                         return (error);
  449                 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  450                         if (vp->v_type == VDIR)
  451                                 np->n_direofoffset = 0;
  452                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  453                         if (error == EINTR || error == EIO)
  454                                 return (error);
  455                         np->n_mtime = vattr.va_mtime;
  456                 }
  457         }
  458         /*
  459          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  460          */
  461         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  462                 if (np->n_directio_opens == 0) {
  463                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  464                         if (error)
  465                                 return (error);
  466                         np->n_flag |= NNONCACHE;
  467                 }
  468                 np->n_directio_opens++;
  469         }
  470         np->ra_expect_lbn = 0;
  471         vnode_create_vobject_off(vp, vattr.va_size, ap->a_td);
  472         return (0);
  473 }
  474 
  475 /*
  476  * nfs close vnode op
  477  * What an NFS client should do upon close after writing is a debatable issue.
  478  * Most NFS clients push delayed writes to the server upon close, basically for
  479  * two reasons:
  480  * 1 - So that any write errors may be reported back to the client process
  481  *     doing the close system call. By far the two most likely errors are
  482  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  483  * 2 - To put a worst case upper bound on cache inconsistency between
  484  *     multiple clients for the file.
  485  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  486  * not being able to tell if other clients are writing a file concurrently,
  487  * since there is no way of knowing if the changed modify time in the reply
  488  * is only due to the write for this client.
  489  * (NFS Version 3 provides weak cache consistency data in the reply that
  490  *  should be sufficient to detect and handle this case.)
  491  *
  492  * The current code does the following:
  493  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  494  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  495  *                     or commit them (this satisfies 1 and 2 except for the
  496  *                     case where the server crashes after this close but
  497  *                     before the commit RPC, which is felt to be "good
  498  *                     enough". Changing the last argument to nfs_flush() to
  499  *                     a 1 would force a commit operation, if it is felt a
  500  *                     commit is necessary now.
  501  */
  502 /* ARGSUSED */
  503 static int
  504 nfs_close(struct vop_close_args *ap)
  505 {
  506         struct vnode *vp = ap->a_vp;
  507         struct nfsnode *np = VTONFS(vp);
  508         int error = 0;
  509         int fmode = ap->a_fflag;
  510 
  511         if (vp->v_type == VREG) {
  512             /*
  513              * Examine and clean dirty pages, regardless of NMODIFIED.
  514              * This closes a major hole in close-to-open consistency.
  515              * We want to push out all dirty pages (and buffers) on
  516              * close, regardless of whether they were dirtied by
  517              * mmap'ed writes or via write().
  518              */
  519             if (nfs_clean_pages_on_close && vp->v_object) {
  520                 VM_OBJECT_LOCK(vp->v_object);
  521                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  522                 VM_OBJECT_UNLOCK(vp->v_object);
  523             }
  524             if (np->n_flag & NMODIFIED) {
  525                 if (NFS_ISV3(vp)) {
  526                     /*
  527                      * Under NFSv3 we have dirty buffers to dispose of.  We
  528                      * must flush them to the NFS server.  We have the option
  529                      * of waiting all the way through the commit rpc or just
  530                      * waiting for the initial write.  The default is to only
  531                      * wait through the initial write so the data is in the
  532                      * server's cache, which is roughly similar to the state
  533                      * a standard disk subsystem leaves the file in on close().
  534                      *
  535                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  536                      * potential races with other processes, and certainly
  537                      * cannot clear it if we don't commit.
  538                      */
  539                     int cm = nfsv3_commit_on_close ? 1 : 0;
  540                     error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
  541                     /* np->n_flag &= ~NMODIFIED; */
  542                 } else
  543                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  544             }
  545             /* 
  546              * Invalidate the attribute cache in all cases.
  547              * An open is going to fetch fresh attrs any way, other procs
  548              * on this node that have file open will be forced to do an 
  549              * otw attr fetch, but this is safe.
  550              */
  551             np->n_attrstamp = 0;
  552             if (np->n_flag & NWRITEERR) {
  553                 np->n_flag &= ~NWRITEERR;
  554                 error = np->n_error;
  555             }
  556         }
  557         if (nfs_directio_enable)
  558                 KASSERT((np->n_directio_asyncwr == 0),
  559                         ("nfs_close: dirty unflushed (%d) directio buffers\n",
  560                          np->n_directio_asyncwr));
  561         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  562                 KASSERT((np->n_directio_opens > 0), 
  563                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));           
  564                 np->n_directio_opens--;
  565                 if (np->n_directio_opens == 0)
  566                         np->n_flag &= ~NNONCACHE;
  567         }
  568         return (error);
  569 }
  570 
  571 /*
  572  * nfs getattr call from vfs.
  573  */
  574 static int
  575 nfs_getattr(struct vop_getattr_args *ap)
  576 {
  577         struct vnode *vp = ap->a_vp;
  578         struct nfsnode *np = VTONFS(vp);
  579         caddr_t bpos, dpos;
  580         int error = 0;
  581         struct mbuf *mreq, *mrep, *md, *mb;
  582         int v3 = NFS_ISV3(vp);
  583 
  584         /*
  585          * Update local times for special files.
  586          */
  587         if (np->n_flag & (NACC | NUPD))
  588                 np->n_flag |= NCHG;
  589         /*
  590          * First look in the cache.
  591          */
  592         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  593                 return (0);
  594 
  595         if (v3 && nfsaccess_cache_timeout > 0) {
  596                 nfsstats.accesscache_misses++;
  597                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  598                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  599                         return (0);
  600         }
  601 
  602         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  603         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  604         mb = mreq;
  605         bpos = mtod(mb, caddr_t);
  606         nfsm_fhtom(vp, v3);
  607         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  608         if (!error) {
  609                 nfsm_loadattr(vp, ap->a_vap);
  610         }
  611         m_freem(mrep);
  612 nfsmout:
  613         return (error);
  614 }
  615 
  616 /*
  617  * nfs setattr call.
  618  */
  619 static int
  620 nfs_setattr(struct vop_setattr_args *ap)
  621 {
  622         struct vnode *vp = ap->a_vp;
  623         struct nfsnode *np = VTONFS(vp);
  624         struct vattr *vap = ap->a_vap;
  625         int error = 0;
  626         u_quad_t tsize;
  627 
  628 #ifndef nolint
  629         tsize = (u_quad_t)0;
  630 #endif
  631 
  632         /*
  633          * Setting of flags and marking of atimes are not supported.
  634          */
  635         if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
  636                 return (EOPNOTSUPP);
  637 
  638         /*
  639          * Disallow write attempts if the filesystem is mounted read-only.
  640          */
  641         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  642             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  643             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  644             (vp->v_mount->mnt_flag & MNT_RDONLY))
  645                 return (EROFS);
  646         if (vap->va_size != VNOVAL) {
  647                 switch (vp->v_type) {
  648                 case VDIR:
  649                         return (EISDIR);
  650                 case VCHR:
  651                 case VBLK:
  652                 case VSOCK:
  653                 case VFIFO:
  654                         if (vap->va_mtime.tv_sec == VNOVAL &&
  655                             vap->va_atime.tv_sec == VNOVAL &&
  656                             vap->va_mode == (mode_t)VNOVAL &&
  657                             vap->va_uid == (uid_t)VNOVAL &&
  658                             vap->va_gid == (gid_t)VNOVAL)
  659                                 return (0);
  660                         vap->va_size = VNOVAL;
  661                         break;
  662                 default:
  663                         /*
  664                          * Disallow write attempts if the filesystem is
  665                          * mounted read-only.
  666                          */
  667                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  668                                 return (EROFS);
  669 
  670                         /*
  671                          *  We run vnode_pager_setsize() early (why?),
  672                          * we must set np->n_size now to avoid vinvalbuf
  673                          * V_SAVE races that might setsize a lower
  674                          * value.
  675                          */
  676 
  677                         tsize = np->n_size;
  678                         error = nfs_meta_setsize(vp, ap->a_cred, 
  679                                                 ap->a_td, vap->va_size);
  680 
  681                         if (np->n_flag & NMODIFIED) {
  682                             if (vap->va_size == 0)
  683                                 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
  684                             else
  685                                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  686                             if (error) {
  687                                 vnode_pager_setsize(vp, np->n_size);
  688                                 return (error);
  689                             }
  690                         }
  691                         /*
  692                          * np->n_size has already been set to vap->va_size
  693                          * in nfs_meta_setsize(). We must set it again since
  694                          * nfs_loadattrcache() could be called through
  695                          * nfs_meta_setsize() and could modify np->n_size.
  696                          */
  697                         np->n_vattr.va_size = np->n_size = vap->va_size;
  698                 };
  699         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  700                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  701                 vp->v_type == VREG &&
  702                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
  703                     (error == EINTR || error == EIO))
  704                 return (error);
  705         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  706         if (error && vap->va_size != VNOVAL) {
  707                 np->n_size = np->n_vattr.va_size = tsize;
  708                 vnode_pager_setsize(vp, np->n_size);
  709         }
  710         return (error);
  711 }
  712 
  713 /*
  714  * Do an nfs setattr rpc.
  715  */
  716 static int
  717 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  718     struct thread *td)
  719 {
  720         struct nfsv2_sattr *sp;
  721         struct nfsnode *np = VTONFS(vp);
  722         caddr_t bpos, dpos;
  723         u_int32_t *tl;
  724         int error = 0, wccflag = NFSV3_WCCRATTR;
  725         struct mbuf *mreq, *mrep, *md, *mb;
  726         int v3 = NFS_ISV3(vp);
  727 
  728         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  729         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  730         mb = mreq;
  731         bpos = mtod(mb, caddr_t);
  732         nfsm_fhtom(vp, v3);
  733         if (v3) {
  734                 nfsm_v3attrbuild(vap, TRUE);
  735                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  736                 *tl = nfs_false;
  737         } else {
  738                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  739                 if (vap->va_mode == (mode_t)VNOVAL)
  740                         sp->sa_mode = nfs_xdrneg1;
  741                 else
  742                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  743                 if (vap->va_uid == (uid_t)VNOVAL)
  744                         sp->sa_uid = nfs_xdrneg1;
  745                 else
  746                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  747                 if (vap->va_gid == (gid_t)VNOVAL)
  748                         sp->sa_gid = nfs_xdrneg1;
  749                 else
  750                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  751                 sp->sa_size = txdr_unsigned(vap->va_size);
  752                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  753                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  754         }
  755         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  756         if (v3) {
  757                 np->n_modestamp = 0;
  758                 nfsm_wcc_data(vp, wccflag);
  759         } else
  760                 nfsm_loadattr(vp, NULL);
  761         m_freem(mrep);
  762 nfsmout:
  763         return (error);
  764 }
  765 
  766 /*
  767  * nfs lookup call, one step at a time...
  768  * First look in cache
  769  * If not found, unlock the directory nfsnode and do the rpc
  770  */
  771 static int
  772 nfs_lookup(struct vop_lookup_args *ap)
  773 {
  774         struct componentname *cnp = ap->a_cnp;
  775         struct vnode *dvp = ap->a_dvp;
  776         struct vnode **vpp = ap->a_vpp;
  777         int flags = cnp->cn_flags;
  778         struct vnode *newvp;
  779         struct nfsmount *nmp;
  780         caddr_t bpos, dpos;
  781         struct mbuf *mreq, *mrep, *md, *mb;
  782         long len;
  783         nfsfh_t *fhp;
  784         struct nfsnode *np;
  785         int error = 0, attrflag, fhsize;
  786         int v3 = NFS_ISV3(dvp);
  787         struct thread *td = cnp->cn_thread;
  788 
  789         *vpp = NULLVP;
  790         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  791             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  792                 return (EROFS);
  793         if (dvp->v_type != VDIR)
  794                 return (ENOTDIR);
  795         nmp = VFSTONFS(dvp->v_mount);
  796         np = VTONFS(dvp);
  797         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  798                 *vpp = NULLVP;
  799                 return (error);
  800         }
  801         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  802                 struct vattr vattr;
  803 
  804                 newvp = *vpp;
  805                 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  806                  && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  807                      nfsstats.lookupcache_hits++;
  808                      if (cnp->cn_nameiop != LOOKUP &&
  809                          (flags & ISLASTCN))
  810                              cnp->cn_flags |= SAVENAME;
  811                      return (0);
  812                 }
  813                 cache_purge(newvp);
  814                 if (dvp != newvp)
  815                         vput(newvp);
  816                 else 
  817                         vrele(newvp);
  818                 *vpp = NULLVP;
  819         }
  820         error = 0;
  821         newvp = NULLVP;
  822         nfsstats.lookupcache_misses++;
  823         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  824         len = cnp->cn_namelen;
  825         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  826                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  827         mb = mreq;
  828         bpos = mtod(mb, caddr_t);
  829         nfsm_fhtom(dvp, v3);
  830         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  831         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  832         if (error) {
  833                 if (v3) {
  834                         nfsm_postop_attr(dvp, attrflag);
  835                         m_freem(mrep);
  836                 }
  837                 goto nfsmout;
  838         }
  839         nfsm_getfh(fhp, fhsize, v3);
  840 
  841         /*
  842          * Handle RENAME case...
  843          */
  844         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
  845                 if (NFS_CMPFH(np, fhp, fhsize)) {
  846                         m_freem(mrep);
  847                         return (EISDIR);
  848                 }
  849                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
  850                 if (error) {
  851                         m_freem(mrep);
  852                         return (error);
  853                 }
  854                 newvp = NFSTOV(np);
  855                 if (v3) {
  856                         nfsm_postop_attr(newvp, attrflag);
  857                         nfsm_postop_attr(dvp, attrflag);
  858                 } else
  859                         nfsm_loadattr(newvp, NULL);
  860                 *vpp = newvp;
  861                 m_freem(mrep);
  862                 cnp->cn_flags |= SAVENAME;
  863                 return (0);
  864         }
  865 
  866         if (flags & ISDOTDOT) {
  867                 VOP_UNLOCK(dvp, 0, td);
  868                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
  869                 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  870                 if (error)
  871                         return (error);
  872                 newvp = NFSTOV(np);
  873         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  874                 VREF(dvp);
  875                 newvp = dvp;
  876         } else {
  877                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
  878                 if (error) {
  879                         m_freem(mrep);
  880                         return (error);
  881                 }
  882                 newvp = NFSTOV(np);
  883         }
  884         if (v3) {
  885                 nfsm_postop_attr(newvp, attrflag);
  886                 nfsm_postop_attr(dvp, attrflag);
  887         } else
  888                 nfsm_loadattr(newvp, NULL);
  889         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  890                 cnp->cn_flags |= SAVENAME;
  891         if ((cnp->cn_flags & MAKEENTRY) &&
  892             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  893                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  894                 cache_enter(dvp, newvp, cnp);
  895         }
  896         *vpp = newvp;
  897         m_freem(mrep);
  898 nfsmout:
  899         if (error) {
  900                 if (newvp != NULLVP) {
  901                         vput(newvp);
  902                         *vpp = NULLVP;
  903                 }
  904                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  905                     (flags & ISLASTCN) && error == ENOENT) {
  906                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  907                                 error = EROFS;
  908                         else
  909                                 error = EJUSTRETURN;
  910                 }
  911                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  912                         cnp->cn_flags |= SAVENAME;
  913         }
  914         return (error);
  915 }
  916 
  917 /*
  918  * nfs read call.
  919  * Just call nfs_bioread() to do the work.
  920  */
  921 static int
  922 nfs_read(struct vop_read_args *ap)
  923 {
  924         struct vnode *vp = ap->a_vp;
  925 
  926         switch (vp->v_type) {
  927         case VREG:
  928                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  929         case VDIR:
  930                 return (EISDIR);
  931         default:
  932                 return (EOPNOTSUPP);
  933         }
  934 }
  935 
  936 /*
  937  * nfs readlink call
  938  */
  939 static int
  940 nfs_readlink(struct vop_readlink_args *ap)
  941 {
  942         struct vnode *vp = ap->a_vp;
  943 
  944         if (vp->v_type != VLNK)
  945                 return (EINVAL);
  946         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  947 }
  948 
  949 /*
  950  * Do a readlink rpc.
  951  * Called by nfs_doio() from below the buffer cache.
  952  */
  953 int
  954 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  955 {
  956         caddr_t bpos, dpos;
  957         int error = 0, len, attrflag;
  958         struct mbuf *mreq, *mrep, *md, *mb;
  959         int v3 = NFS_ISV3(vp);
  960 
  961         nfsstats.rpccnt[NFSPROC_READLINK]++;
  962         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  963         mb = mreq;
  964         bpos = mtod(mb, caddr_t);
  965         nfsm_fhtom(vp, v3);
  966         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  967         if (v3)
  968                 nfsm_postop_attr(vp, attrflag);
  969         if (!error) {
  970                 nfsm_strsiz(len, NFS_MAXPATHLEN);
  971                 if (len == NFS_MAXPATHLEN) {
  972                         struct nfsnode *np = VTONFS(vp);
  973                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
  974                                 len = np->n_size;
  975                 }
  976                 nfsm_mtouio(uiop, len);
  977         }
  978         m_freem(mrep);
  979 nfsmout:
  980         return (error);
  981 }
  982 
  983 /*
  984  * nfs read rpc call
  985  * Ditto above
  986  */
  987 int
  988 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  989 {
  990         u_int32_t *tl;
  991         caddr_t bpos, dpos;
  992         struct mbuf *mreq, *mrep, *md, *mb;
  993         struct nfsmount *nmp;
  994         int error = 0, len, retlen, tsiz, eof, attrflag;
  995         int v3 = NFS_ISV3(vp);
  996 
  997 #ifndef nolint
  998         eof = 0;
  999 #endif
 1000         nmp = VFSTONFS(vp->v_mount);
 1001         tsiz = uiop->uio_resid;
 1002         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1003                 return (EFBIG);
 1004         while (tsiz > 0) {
 1005                 nfsstats.rpccnt[NFSPROC_READ]++;
 1006                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
 1007                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1008                 mb = mreq;
 1009                 bpos = mtod(mb, caddr_t);
 1010                 nfsm_fhtom(vp, v3);
 1011                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1012                 if (v3) {
 1013                         txdr_hyper(uiop->uio_offset, tl);
 1014                         *(tl + 2) = txdr_unsigned(len);
 1015                 } else {
 1016                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1017                         *tl++ = txdr_unsigned(len);
 1018                         *tl = 0;
 1019                 }
 1020                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1021                 if (v3) {
 1022                         nfsm_postop_attr(vp, attrflag);
 1023                         if (error) {
 1024                                 m_freem(mrep);
 1025                                 goto nfsmout;
 1026                         }
 1027                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1028                         eof = fxdr_unsigned(int, *(tl + 1));
 1029                 } else
 1030                         nfsm_loadattr(vp, NULL);
 1031                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1032                 nfsm_mtouio(uiop, retlen);
 1033                 m_freem(mrep);
 1034                 tsiz -= retlen;
 1035                 if (v3) {
 1036                         if (eof || retlen == 0) {
 1037                                 tsiz = 0;
 1038                         }
 1039                 } else if (retlen < len) {
 1040                         tsiz = 0;
 1041                 }
 1042         }
 1043 nfsmout:
 1044         return (error);
 1045 }
 1046 
 1047 /*
 1048  * nfs write call
 1049  */
 1050 int
 1051 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1052     int *iomode, int *must_commit)
 1053 {
 1054         u_int32_t *tl;
 1055         int32_t backup;
 1056         caddr_t bpos, dpos;
 1057         struct mbuf *mreq, *mrep, *md, *mb;
 1058         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1059         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1060         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1061 
 1062 #ifndef DIAGNOSTIC
 1063         if (uiop->uio_iovcnt != 1)
 1064                 panic("nfs: writerpc iovcnt > 1");
 1065 #endif
 1066         *must_commit = 0;
 1067         tsiz = uiop->uio_resid;
 1068         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1069                 return (EFBIG);
 1070         while (tsiz > 0) {
 1071                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1072                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1073                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1074                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1075                 mb = mreq;
 1076                 bpos = mtod(mb, caddr_t);
 1077                 nfsm_fhtom(vp, v3);
 1078                 if (v3) {
 1079                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1080                         txdr_hyper(uiop->uio_offset, tl);
 1081                         tl += 2;
 1082                         *tl++ = txdr_unsigned(len);
 1083                         *tl++ = txdr_unsigned(*iomode);
 1084                         *tl = txdr_unsigned(len);
 1085                 } else {
 1086                         u_int32_t x;
 1087 
 1088                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1089                         /* Set both "begin" and "current" to non-garbage. */
 1090                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1091                         *tl++ = x;      /* "begin offset" */
 1092                         *tl++ = x;      /* "current offset" */
 1093                         x = txdr_unsigned(len);
 1094                         *tl++ = x;      /* total to this offset */
 1095                         *tl = x;        /* size of this write */
 1096                 }
 1097                 nfsm_uiotom(uiop, len);
 1098                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1099                 if (v3) {
 1100                         wccflag = NFSV3_WCCCHK;
 1101                         nfsm_wcc_data(vp, wccflag);
 1102                         if (!error) {
 1103                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1104                                         + NFSX_V3WRITEVERF);
 1105                                 rlen = fxdr_unsigned(int, *tl++);
 1106                                 if (rlen == 0) {
 1107                                         error = NFSERR_IO;
 1108                                         m_freem(mrep);
 1109                                         break;
 1110                                 } else if (rlen < len) {
 1111                                         backup = len - rlen;
 1112                                         uiop->uio_iov->iov_base =
 1113                                             (char *)uiop->uio_iov->iov_base -
 1114                                             backup;
 1115                                         uiop->uio_iov->iov_len += backup;
 1116                                         uiop->uio_offset -= backup;
 1117                                         uiop->uio_resid += backup;
 1118                                         len = rlen;
 1119                                 }
 1120                                 commit = fxdr_unsigned(int, *tl++);
 1121 
 1122                                 /*
 1123                                  * Return the lowest committment level
 1124                                  * obtained by any of the RPCs.
 1125                                  */
 1126                                 if (committed == NFSV3WRITE_FILESYNC)
 1127                                         committed = commit;
 1128                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1129                                         commit == NFSV3WRITE_UNSTABLE)
 1130                                         committed = commit;
 1131                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1132                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1133                                         NFSX_V3WRITEVERF);
 1134                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1135                                 } else if (bcmp((caddr_t)tl,
 1136                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1137                                     *must_commit = 1;
 1138                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1139                                         NFSX_V3WRITEVERF);
 1140                                 }
 1141                         }
 1142                 } else
 1143                     nfsm_loadattr(vp, NULL);
 1144                 if (wccflag)
 1145                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
 1146                 m_freem(mrep);
 1147                 if (error)
 1148                         break;
 1149                 tsiz -= len;
 1150         }
 1151 nfsmout:
 1152         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1153                 committed = NFSV3WRITE_FILESYNC;
 1154         *iomode = committed;
 1155         if (error)
 1156                 uiop->uio_resid = tsiz;
 1157         return (error);
 1158 }
 1159 
 1160 /*
 1161  * nfs mknod rpc
 1162  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1163  * mode set to specify the file type and the size field for rdev.
 1164  */
 1165 static int
 1166 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1167     struct vattr *vap)
 1168 {
 1169         struct nfsv2_sattr *sp;
 1170         u_int32_t *tl;
 1171         struct vnode *newvp = NULL;
 1172         struct nfsnode *np = NULL;
 1173         struct vattr vattr;
 1174         caddr_t bpos, dpos;
 1175         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1176         struct mbuf *mreq, *mrep, *md, *mb;
 1177         u_int32_t rdev;
 1178         int v3 = NFS_ISV3(dvp);
 1179 
 1180         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1181                 rdev = txdr_unsigned(vap->va_rdev);
 1182         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1183                 rdev = nfs_xdrneg1;
 1184         else {
 1185                 return (EOPNOTSUPP);
 1186         }
 1187         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1188                 return (error);
 1189         }
 1190         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1191         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1192                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1193         mb = mreq;
 1194         bpos = mtod(mb, caddr_t);
 1195         nfsm_fhtom(dvp, v3);
 1196         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1197         if (v3) {
 1198                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1199                 *tl++ = vtonfsv3_type(vap->va_type);
 1200                 nfsm_v3attrbuild(vap, FALSE);
 1201                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1202                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1203                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1204                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1205                 }
 1206         } else {
 1207                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1208                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1209                 sp->sa_uid = nfs_xdrneg1;
 1210                 sp->sa_gid = nfs_xdrneg1;
 1211                 sp->sa_size = rdev;
 1212                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1213                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1214         }
 1215         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1216         if (!error) {
 1217                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1218                 if (!gotvp) {
 1219                         if (newvp) {
 1220                                 vput(newvp);
 1221                                 newvp = NULL;
 1222                         }
 1223                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1224                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1225                         if (!error)
 1226                                 newvp = NFSTOV(np);
 1227                 }
 1228         }
 1229         if (v3)
 1230                 nfsm_wcc_data(dvp, wccflag);
 1231         m_freem(mrep);
 1232 nfsmout:
 1233         if (error) {
 1234                 if (newvp)
 1235                         vput(newvp);
 1236         } else {
 1237                 if (cnp->cn_flags & MAKEENTRY)
 1238                         cache_enter(dvp, newvp, cnp);
 1239                 *vpp = newvp;
 1240         }
 1241         VTONFS(dvp)->n_flag |= NMODIFIED;
 1242         if (!wccflag)
 1243                 VTONFS(dvp)->n_attrstamp = 0;
 1244         return (error);
 1245 }
 1246 
 1247 /*
 1248  * nfs mknod vop
 1249  * just call nfs_mknodrpc() to do the work.
 1250  */
 1251 /* ARGSUSED */
 1252 static int
 1253 nfs_mknod(struct vop_mknod_args *ap)
 1254 {
 1255 
 1256         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1257 }
 1258 
 1259 static u_long create_verf;
 1260 /*
 1261  * nfs file create call
 1262  */
 1263 static int
 1264 nfs_create(struct vop_create_args *ap)
 1265 {
 1266         struct vnode *dvp = ap->a_dvp;
 1267         struct vattr *vap = ap->a_vap;
 1268         struct componentname *cnp = ap->a_cnp;
 1269         struct nfsv2_sattr *sp;
 1270         u_int32_t *tl;
 1271         struct nfsnode *np = NULL;
 1272         struct vnode *newvp = NULL;
 1273         caddr_t bpos, dpos;
 1274         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1275         struct mbuf *mreq, *mrep, *md, *mb;
 1276         struct vattr vattr;
 1277         int v3 = NFS_ISV3(dvp);
 1278 
 1279         /*
 1280          * Oops, not for me..
 1281          */
 1282         if (vap->va_type == VSOCK)
 1283                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1284 
 1285         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1286                 return (error);
 1287         }
 1288         if (vap->va_vaflags & VA_EXCLUSIVE)
 1289                 fmode |= O_EXCL;
 1290 again:
 1291         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1292         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1293                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1294         mb = mreq;
 1295         bpos = mtod(mb, caddr_t);
 1296         nfsm_fhtom(dvp, v3);
 1297         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1298         if (v3) {
 1299                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1300                 if (fmode & O_EXCL) {
 1301                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1302                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1303 #ifdef INET
 1304                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1305                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1306                         else
 1307 #endif
 1308                                 *tl++ = create_verf;
 1309                         *tl = ++create_verf;
 1310                 } else {
 1311                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1312                         nfsm_v3attrbuild(vap, FALSE);
 1313                 }
 1314         } else {
 1315                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1316                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1317                 sp->sa_uid = nfs_xdrneg1;
 1318                 sp->sa_gid = nfs_xdrneg1;
 1319                 sp->sa_size = 0;
 1320                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1321                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1322         }
 1323         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1324         if (!error) {
 1325                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1326                 if (!gotvp) {
 1327                         if (newvp) {
 1328                                 vput(newvp);
 1329                                 newvp = NULL;
 1330                         }
 1331                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1332                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1333                         if (!error)
 1334                                 newvp = NFSTOV(np);
 1335                 }
 1336         }
 1337         if (v3)
 1338                 nfsm_wcc_data(dvp, wccflag);
 1339         m_freem(mrep);
 1340 nfsmout:
 1341         if (error) {
 1342                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1343                         fmode &= ~O_EXCL;
 1344                         goto again;
 1345                 }
 1346                 if (newvp)
 1347                         vput(newvp);
 1348         } else if (v3 && (fmode & O_EXCL)) {
 1349                 /*
 1350                  * We are normally called with only a partially initialized
 1351                  * VAP.  Since the NFSv3 spec says that server may use the
 1352                  * file attributes to store the verifier, the spec requires
 1353                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1354                  * in atime, but we can't really assume that all servers will
 1355                  * so we ensure that our SETATTR sets both atime and mtime.
 1356                  */
 1357                 if (vap->va_mtime.tv_sec == VNOVAL)
 1358                         vfs_timestamp(&vap->va_mtime);
 1359                 if (vap->va_atime.tv_sec == VNOVAL)
 1360                         vap->va_atime = vap->va_mtime;
 1361                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1362                 if (error)
 1363                         vput(newvp);
 1364         }
 1365         if (!error) {
 1366                 if (cnp->cn_flags & MAKEENTRY)
 1367                         cache_enter(dvp, newvp, cnp);
 1368                 *ap->a_vpp = newvp;
 1369         }
 1370         VTONFS(dvp)->n_flag |= NMODIFIED;
 1371         if (!wccflag)
 1372                 VTONFS(dvp)->n_attrstamp = 0;
 1373         return (error);
 1374 }
 1375 
 1376 /*
 1377  * nfs file remove call
 1378  * To try and make nfs semantics closer to ufs semantics, a file that has
 1379  * other processes using the vnode is renamed instead of removed and then
 1380  * removed later on the last close.
 1381  * - If v_usecount > 1
 1382  *        If a rename is not already in the works
 1383  *           call nfs_sillyrename() to set it up
 1384  *     else
 1385  *        do the remove rpc
 1386  */
 1387 static int
 1388 nfs_remove(struct vop_remove_args *ap)
 1389 {
 1390         struct vnode *vp = ap->a_vp;
 1391         struct vnode *dvp = ap->a_dvp;
 1392         struct componentname *cnp = ap->a_cnp;
 1393         struct nfsnode *np = VTONFS(vp);
 1394         int error = 0;
 1395         struct vattr vattr;
 1396 
 1397 #ifndef DIAGNOSTIC
 1398         if ((cnp->cn_flags & HASBUF) == 0)
 1399                 panic("nfs_remove: no name");
 1400         if (vrefcnt(vp) < 1)
 1401                 panic("nfs_remove: bad v_usecount");
 1402 #endif
 1403         if (vp->v_type == VDIR)
 1404                 error = EPERM;
 1405         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1406             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1407             vattr.va_nlink > 1)) {
 1408                 /*
 1409                  * Purge the name cache so that the chance of a lookup for
 1410                  * the name succeeding while the remove is in progress is
 1411                  * minimized. Without node locking it can still happen, such
 1412                  * that an I/O op returns ESTALE, but since you get this if
 1413                  * another host removes the file..
 1414                  */
 1415                 cache_purge(vp);
 1416                 /*
 1417                  * throw away biocache buffers, mainly to avoid
 1418                  * unnecessary delayed writes later.
 1419                  */
 1420                 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1421                 /* Do the rpc */
 1422                 if (error != EINTR && error != EIO)
 1423                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1424                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1425                 /*
 1426                  * Kludge City: If the first reply to the remove rpc is lost..
 1427                  *   the reply to the retransmitted request will be ENOENT
 1428                  *   since the file was in fact removed
 1429                  *   Therefore, we cheat and return success.
 1430                  */
 1431                 if (error == ENOENT)
 1432                         error = 0;
 1433         } else if (!np->n_sillyrename)
 1434                 error = nfs_sillyrename(dvp, vp, cnp);
 1435         np->n_attrstamp = 0;
 1436         return (error);
 1437 }
 1438 
 1439 /*
 1440  * nfs file remove rpc called from nfs_inactive
 1441  */
 1442 int
 1443 nfs_removeit(struct sillyrename *sp)
 1444 {
 1445 
 1446         /*
 1447          * Make sure that the directory vnode is still valid.
 1448          * XXX we should lock sp->s_dvp here.
 1449          */
 1450         if (sp->s_dvp->v_type == VBAD)
 1451                 return (0);
 1452         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1453                 NULL));
 1454 }
 1455 
 1456 /*
 1457  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1458  */
 1459 static int
 1460 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1461     struct ucred *cred, struct thread *td)
 1462 {
 1463         caddr_t bpos, dpos;
 1464         int error = 0, wccflag = NFSV3_WCCRATTR;
 1465         struct mbuf *mreq, *mrep, *md, *mb;
 1466         int v3 = NFS_ISV3(dvp);
 1467 
 1468         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1469         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1470                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1471         mb = mreq;
 1472         bpos = mtod(mb, caddr_t);
 1473         nfsm_fhtom(dvp, v3);
 1474         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1475         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1476         if (v3)
 1477                 nfsm_wcc_data(dvp, wccflag);
 1478         m_freem(mrep);
 1479 nfsmout:
 1480         VTONFS(dvp)->n_flag |= NMODIFIED;
 1481         if (!wccflag)
 1482                 VTONFS(dvp)->n_attrstamp = 0;
 1483         return (error);
 1484 }
 1485 
 1486 /*
 1487  * nfs file rename call
 1488  */
 1489 static int
 1490 nfs_rename(struct vop_rename_args *ap)
 1491 {
 1492         struct vnode *fvp = ap->a_fvp;
 1493         struct vnode *tvp = ap->a_tvp;
 1494         struct vnode *fdvp = ap->a_fdvp;
 1495         struct vnode *tdvp = ap->a_tdvp;
 1496         struct componentname *tcnp = ap->a_tcnp;
 1497         struct componentname *fcnp = ap->a_fcnp;
 1498         int error;
 1499 
 1500 #ifndef DIAGNOSTIC
 1501         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1502             (fcnp->cn_flags & HASBUF) == 0)
 1503                 panic("nfs_rename: no name");
 1504 #endif
 1505         /* Check for cross-device rename */
 1506         if ((fvp->v_mount != tdvp->v_mount) ||
 1507             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1508                 error = EXDEV;
 1509                 goto out;
 1510         }
 1511 
 1512         if (fvp == tvp) {
 1513                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1514                 error = 0;
 1515                 goto out;
 1516         }
 1517         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1518                 goto out;
 1519 
 1520         /*
 1521          * We have to flush B_DELWRI data prior to renaming
 1522          * the file.  If we don't, the delayed-write buffers
 1523          * can be flushed out later after the file has gone stale
 1524          * under NFSV3.  NFSV2 does not have this problem because
 1525          * ( as far as I can tell ) it flushes dirty buffers more
 1526          * often.
 1527          * 
 1528          * Skip the rename operation if the fsync fails, this can happen
 1529          * due to the server's volume being full, when we pushed out data
 1530          * that was written back to our cache earlier. Not checking for
 1531          * this condition can result in potential (silent) data loss.
 1532          */
 1533         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1534         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1535         if (!error && tvp)
 1536                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1537         if (error)
 1538                 goto out;
 1539 
 1540         /*
 1541          * If the tvp exists and is in use, sillyrename it before doing the
 1542          * rename of the new file over it.
 1543          * XXX Can't sillyrename a directory.
 1544          */
 1545         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1546                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1547                 vput(tvp);
 1548                 tvp = NULL;
 1549         }
 1550 
 1551         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1552                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1553                 tcnp->cn_thread);
 1554 
 1555         if (fvp->v_type == VDIR) {
 1556                 if (tvp != NULL && tvp->v_type == VDIR)
 1557                         cache_purge(tdvp);
 1558                 cache_purge(fdvp);
 1559         }
 1560 
 1561 out:
 1562         if (tdvp == tvp)
 1563                 vrele(tdvp);
 1564         else
 1565                 vput(tdvp);
 1566         if (tvp)
 1567                 vput(tvp);
 1568         vrele(fdvp);
 1569         vrele(fvp);
 1570         /*
 1571          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1572          */
 1573         if (error == ENOENT)
 1574                 error = 0;
 1575         return (error);
 1576 }
 1577 
 1578 /*
 1579  * nfs file rename rpc called from nfs_remove() above
 1580  */
 1581 static int
 1582 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1583     struct sillyrename *sp)
 1584 {
 1585 
 1586         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1587             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1588 }
 1589 
 1590 /*
 1591  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1592  */
 1593 static int
 1594 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1595     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1596     struct thread *td)
 1597 {
 1598         caddr_t bpos, dpos;
 1599         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1600         struct mbuf *mreq, *mrep, *md, *mb;
 1601         int v3 = NFS_ISV3(fdvp);
 1602 
 1603         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1604         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1605                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1606                 nfsm_rndup(tnamelen));
 1607         mb = mreq;
 1608         bpos = mtod(mb, caddr_t);
 1609         nfsm_fhtom(fdvp, v3);
 1610         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1611         nfsm_fhtom(tdvp, v3);
 1612         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1613         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1614         if (v3) {
 1615                 nfsm_wcc_data(fdvp, fwccflag);
 1616                 nfsm_wcc_data(tdvp, twccflag);
 1617         }
 1618         m_freem(mrep);
 1619 nfsmout:
 1620         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1621         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1622         if (!fwccflag)
 1623                 VTONFS(fdvp)->n_attrstamp = 0;
 1624         if (!twccflag)
 1625                 VTONFS(tdvp)->n_attrstamp = 0;
 1626         return (error);
 1627 }
 1628 
 1629 /*
 1630  * nfs hard link create call
 1631  */
 1632 static int
 1633 nfs_link(struct vop_link_args *ap)
 1634 {
 1635         struct vnode *vp = ap->a_vp;
 1636         struct vnode *tdvp = ap->a_tdvp;
 1637         struct componentname *cnp = ap->a_cnp;
 1638         caddr_t bpos, dpos;
 1639         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1640         struct mbuf *mreq, *mrep, *md, *mb;
 1641         int v3;
 1642 
 1643         if (vp->v_mount != tdvp->v_mount) {
 1644                 return (EXDEV);
 1645         }
 1646 
 1647         /*
 1648          * Push all writes to the server, so that the attribute cache
 1649          * doesn't get "out of sync" with the server.
 1650          * XXX There should be a better way!
 1651          */
 1652         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1653 
 1654         v3 = NFS_ISV3(vp);
 1655         nfsstats.rpccnt[NFSPROC_LINK]++;
 1656         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1657                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1658         mb = mreq;
 1659         bpos = mtod(mb, caddr_t);
 1660         nfsm_fhtom(vp, v3);
 1661         nfsm_fhtom(tdvp, v3);
 1662         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1663         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1664         if (v3) {
 1665                 nfsm_postop_attr(vp, attrflag);
 1666                 nfsm_wcc_data(tdvp, wccflag);
 1667         }
 1668         m_freem(mrep);
 1669 nfsmout:
 1670         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1671         if (!attrflag)
 1672                 VTONFS(vp)->n_attrstamp = 0;
 1673         if (!wccflag)
 1674                 VTONFS(tdvp)->n_attrstamp = 0;
 1675         return (error);
 1676 }
 1677 
 1678 /*
 1679  * nfs symbolic link create call
 1680  */
 1681 static int
 1682 nfs_symlink(struct vop_symlink_args *ap)
 1683 {
 1684         struct vnode *dvp = ap->a_dvp;
 1685         struct vattr *vap = ap->a_vap;
 1686         struct componentname *cnp = ap->a_cnp;
 1687         struct nfsv2_sattr *sp;
 1688         caddr_t bpos, dpos;
 1689         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1690         struct mbuf *mreq, *mrep, *md, *mb;
 1691         struct vnode *newvp = NULL;
 1692         int v3 = NFS_ISV3(dvp);
 1693 
 1694         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1695         slen = strlen(ap->a_target);
 1696         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1697             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1698         mb = mreq;
 1699         bpos = mtod(mb, caddr_t);
 1700         nfsm_fhtom(dvp, v3);
 1701         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1702         if (v3) {
 1703                 nfsm_v3attrbuild(vap, FALSE);
 1704         }
 1705         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1706         if (!v3) {
 1707                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1708                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1709                 sp->sa_uid = nfs_xdrneg1;
 1710                 sp->sa_gid = nfs_xdrneg1;
 1711                 sp->sa_size = nfs_xdrneg1;
 1712                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1713                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1714         }
 1715 
 1716         /*
 1717          * Issue the NFS request and get the rpc response.
 1718          *
 1719          * Only NFSv3 responses returning an error of 0 actually return
 1720          * a file handle that can be converted into newvp without having
 1721          * to do an extra lookup rpc.
 1722          */
 1723         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1724         if (v3) {
 1725                 if (error == 0)
 1726                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1727                 nfsm_wcc_data(dvp, wccflag);
 1728         }
 1729 
 1730         /*
 1731          * out code jumps -> here, mrep is also freed.
 1732          */
 1733 
 1734         m_freem(mrep);
 1735 nfsmout:
 1736 
 1737         /*
 1738          * If we do not have an error and we could not extract the newvp from
 1739          * the response due to the request being NFSv2, we have to do a
 1740          * lookup in order to obtain a newvp to return.
 1741          */
 1742         if (error == 0 && newvp == NULL) {
 1743                 struct nfsnode *np = NULL;
 1744 
 1745                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1746                     cnp->cn_cred, cnp->cn_thread, &np);
 1747                 if (!error)
 1748                         newvp = NFSTOV(np);
 1749         }
 1750         if (error) {
 1751                 if (newvp)
 1752                         vput(newvp);
 1753         } else {
 1754                 *ap->a_vpp = newvp;
 1755         }
 1756         VTONFS(dvp)->n_flag |= NMODIFIED;
 1757         if (!wccflag)
 1758                 VTONFS(dvp)->n_attrstamp = 0;
 1759         return (error);
 1760 }
 1761 
 1762 /*
 1763  * nfs make dir call
 1764  */
 1765 static int
 1766 nfs_mkdir(struct vop_mkdir_args *ap)
 1767 {
 1768         struct vnode *dvp = ap->a_dvp;
 1769         struct vattr *vap = ap->a_vap;
 1770         struct componentname *cnp = ap->a_cnp;
 1771         struct nfsv2_sattr *sp;
 1772         int len;
 1773         struct nfsnode *np = NULL;
 1774         struct vnode *newvp = NULL;
 1775         caddr_t bpos, dpos;
 1776         int error = 0, wccflag = NFSV3_WCCRATTR;
 1777         int gotvp = 0;
 1778         struct mbuf *mreq, *mrep, *md, *mb;
 1779         struct vattr vattr;
 1780         int v3 = NFS_ISV3(dvp);
 1781 
 1782         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1783                 return (error);
 1784         }
 1785         len = cnp->cn_namelen;
 1786         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1787         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1788           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1789         mb = mreq;
 1790         bpos = mtod(mb, caddr_t);
 1791         nfsm_fhtom(dvp, v3);
 1792         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1793         if (v3) {
 1794                 nfsm_v3attrbuild(vap, FALSE);
 1795         } else {
 1796                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1797                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1798                 sp->sa_uid = nfs_xdrneg1;
 1799                 sp->sa_gid = nfs_xdrneg1;
 1800                 sp->sa_size = nfs_xdrneg1;
 1801                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1802                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1803         }
 1804         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1805         if (!error)
 1806                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1807         if (v3)
 1808                 nfsm_wcc_data(dvp, wccflag);
 1809         m_freem(mrep);
 1810 nfsmout:
 1811         VTONFS(dvp)->n_flag |= NMODIFIED;
 1812         if (!wccflag)
 1813                 VTONFS(dvp)->n_attrstamp = 0;
 1814         if (error == 0 && newvp == NULL) {
 1815                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1816                         cnp->cn_thread, &np);
 1817                 if (!error) {
 1818                         newvp = NFSTOV(np);
 1819                         if (newvp->v_type != VDIR)
 1820                                 error = EEXIST;
 1821                 }
 1822         }
 1823         if (error) {
 1824                 if (newvp)
 1825                         vput(newvp);
 1826         } else
 1827                 *ap->a_vpp = newvp;
 1828         return (error);
 1829 }
 1830 
 1831 /*
 1832  * nfs remove directory call
 1833  */
 1834 static int
 1835 nfs_rmdir(struct vop_rmdir_args *ap)
 1836 {
 1837         struct vnode *vp = ap->a_vp;
 1838         struct vnode *dvp = ap->a_dvp;
 1839         struct componentname *cnp = ap->a_cnp;
 1840         caddr_t bpos, dpos;
 1841         int error = 0, wccflag = NFSV3_WCCRATTR;
 1842         struct mbuf *mreq, *mrep, *md, *mb;
 1843         int v3 = NFS_ISV3(dvp);
 1844 
 1845         if (dvp == vp)
 1846                 return (EINVAL);
 1847         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1848         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1849                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1850         mb = mreq;
 1851         bpos = mtod(mb, caddr_t);
 1852         nfsm_fhtom(dvp, v3);
 1853         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1854         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1855         if (v3)
 1856                 nfsm_wcc_data(dvp, wccflag);
 1857         m_freem(mrep);
 1858 nfsmout:
 1859         VTONFS(dvp)->n_flag |= NMODIFIED;
 1860         if (!wccflag)
 1861                 VTONFS(dvp)->n_attrstamp = 0;
 1862         cache_purge(dvp);
 1863         cache_purge(vp);
 1864         /*
 1865          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1866          */
 1867         if (error == ENOENT)
 1868                 error = 0;
 1869         return (error);
 1870 }
 1871 
 1872 /*
 1873  * nfs readdir call
 1874  */
 1875 static int
 1876 nfs_readdir(struct vop_readdir_args *ap)
 1877 {
 1878         struct vnode *vp = ap->a_vp;
 1879         struct nfsnode *np = VTONFS(vp);
 1880         struct uio *uio = ap->a_uio;
 1881         int tresid, error;
 1882         struct vattr vattr;
 1883 
 1884         if (vp->v_type != VDIR)
 1885                 return (EPERM);
 1886         /*
 1887          * First, check for hit on the EOF offset cache
 1888          */
 1889         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1890             (np->n_flag & NMODIFIED) == 0) {
 1891                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1892                     !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 1893                         nfsstats.direofcache_hits++;
 1894                         return (0);
 1895                 }
 1896         }
 1897 
 1898         /*
 1899          * Call nfs_bioread() to do the real work.
 1900          */
 1901         tresid = uio->uio_resid;
 1902         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1903 
 1904         if (!error && uio->uio_resid == tresid)
 1905                 nfsstats.direofcache_misses++;
 1906         return (error);
 1907 }
 1908 
 1909 /*
 1910  * Readdir rpc call.
 1911  * Called from below the buffer cache by nfs_doio().
 1912  */
 1913 int
 1914 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1915 {
 1916         int len, left;
 1917         struct dirent *dp = NULL;
 1918         u_int32_t *tl;
 1919         caddr_t cp;
 1920         nfsuint64 *cookiep;
 1921         caddr_t bpos, dpos;
 1922         struct mbuf *mreq, *mrep, *md, *mb;
 1923         nfsuint64 cookie;
 1924         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1925         struct nfsnode *dnp = VTONFS(vp);
 1926         u_quad_t fileno;
 1927         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1928         int attrflag;
 1929         int v3 = NFS_ISV3(vp);
 1930 
 1931 #ifndef DIAGNOSTIC
 1932         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1933                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1934                 panic("nfs readdirrpc bad uio");
 1935 #endif
 1936 
 1937         /*
 1938          * If there is no cookie, assume directory was stale.
 1939          */
 1940         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1941         if (cookiep)
 1942                 cookie = *cookiep;
 1943         else
 1944                 return (NFSERR_BAD_COOKIE);
 1945         /*
 1946          * Loop around doing readdir rpc's of size nm_readdirsize
 1947          * truncated to a multiple of DIRBLKSIZ.
 1948          * The stopping criteria is EOF or buffer full.
 1949          */
 1950         while (more_dirs && bigenough) {
 1951                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1952                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1953                         NFSX_READDIR(v3));
 1954                 mb = mreq;
 1955                 bpos = mtod(mb, caddr_t);
 1956                 nfsm_fhtom(vp, v3);
 1957                 if (v3) {
 1958                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1959                         *tl++ = cookie.nfsuquad[0];
 1960                         *tl++ = cookie.nfsuquad[1];
 1961                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 1962                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 1963                 } else {
 1964                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1965                         *tl++ = cookie.nfsuquad[0];
 1966                 }
 1967                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 1968                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 1969                 if (v3) {
 1970                         nfsm_postop_attr(vp, attrflag);
 1971                         if (!error) {
 1972                                 tl = nfsm_dissect(u_int32_t *,
 1973                                     2 * NFSX_UNSIGNED);
 1974                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 1975                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 1976                         } else {
 1977                                 m_freem(mrep);
 1978                                 goto nfsmout;
 1979                         }
 1980                 }
 1981                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 1982                 more_dirs = fxdr_unsigned(int, *tl);
 1983 
 1984                 /* loop thru the dir entries, doctoring them to 4bsd form */
 1985                 while (more_dirs && bigenough) {
 1986                         if (v3) {
 1987                                 tl = nfsm_dissect(u_int32_t *,
 1988                                     3 * NFSX_UNSIGNED);
 1989                                 fileno = fxdr_hyper(tl);
 1990                                 len = fxdr_unsigned(int, *(tl + 2));
 1991                         } else {
 1992                                 tl = nfsm_dissect(u_int32_t *,
 1993                                     2 * NFSX_UNSIGNED);
 1994                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 1995                                 len = fxdr_unsigned(int, *tl);
 1996                         }
 1997                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 1998                                 error = EBADRPC;
 1999                                 m_freem(mrep);
 2000                                 goto nfsmout;
 2001                         }
 2002                         tlen = nfsm_rndup(len);
 2003                         if (tlen == len)
 2004                                 tlen += 4;      /* To ensure null termination */
 2005                         left = DIRBLKSIZ - blksiz;
 2006                         if ((tlen + DIRHDSIZ) > left) {
 2007                                 dp->d_reclen += left;
 2008                                 uiop->uio_iov->iov_base =
 2009                                     (char *)uiop->uio_iov->iov_base + left;
 2010                                 uiop->uio_iov->iov_len -= left;
 2011                                 uiop->uio_offset += left;
 2012                                 uiop->uio_resid -= left;
 2013                                 blksiz = 0;
 2014                         }
 2015                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2016                                 bigenough = 0;
 2017                         if (bigenough) {
 2018                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2019                                 dp->d_fileno = (int)fileno;
 2020                                 dp->d_namlen = len;
 2021                                 dp->d_reclen = tlen + DIRHDSIZ;
 2022                                 dp->d_type = DT_UNKNOWN;
 2023                                 blksiz += dp->d_reclen;
 2024                                 if (blksiz == DIRBLKSIZ)
 2025                                         blksiz = 0;
 2026                                 uiop->uio_offset += DIRHDSIZ;
 2027                                 uiop->uio_resid -= DIRHDSIZ;
 2028                                 uiop->uio_iov->iov_base =
 2029                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2030                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2031                                 nfsm_mtouio(uiop, len);
 2032                                 cp = uiop->uio_iov->iov_base;
 2033                                 tlen -= len;
 2034                                 *cp = '\0';     /* null terminate */
 2035                                 uiop->uio_iov->iov_base =
 2036                                     (char *)uiop->uio_iov->iov_base + tlen;
 2037                                 uiop->uio_iov->iov_len -= tlen;
 2038                                 uiop->uio_offset += tlen;
 2039                                 uiop->uio_resid -= tlen;
 2040                         } else
 2041                                 nfsm_adv(nfsm_rndup(len));
 2042                         if (v3) {
 2043                                 tl = nfsm_dissect(u_int32_t *,
 2044                                     3 * NFSX_UNSIGNED);
 2045                         } else {
 2046                                 tl = nfsm_dissect(u_int32_t *,
 2047                                     2 * NFSX_UNSIGNED);
 2048                         }
 2049                         if (bigenough) {
 2050                                 cookie.nfsuquad[0] = *tl++;
 2051                                 if (v3)
 2052                                         cookie.nfsuquad[1] = *tl++;
 2053                         } else if (v3)
 2054                                 tl += 2;
 2055                         else
 2056                                 tl++;
 2057                         more_dirs = fxdr_unsigned(int, *tl);
 2058                 }
 2059                 /*
 2060                  * If at end of rpc data, get the eof boolean
 2061                  */
 2062                 if (!more_dirs) {
 2063                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2064                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2065                 }
 2066                 m_freem(mrep);
 2067         }
 2068         /*
 2069          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2070          * by increasing d_reclen for the last record.
 2071          */
 2072         if (blksiz > 0) {
 2073                 left = DIRBLKSIZ - blksiz;
 2074                 dp->d_reclen += left;
 2075                 uiop->uio_iov->iov_base =
 2076                     (char *)uiop->uio_iov->iov_base + left;
 2077                 uiop->uio_iov->iov_len -= left;
 2078                 uiop->uio_offset += left;
 2079                 uiop->uio_resid -= left;
 2080         }
 2081 
 2082         /*
 2083          * We are now either at the end of the directory or have filled the
 2084          * block.
 2085          */
 2086         if (bigenough)
 2087                 dnp->n_direofoffset = uiop->uio_offset;
 2088         else {
 2089                 if (uiop->uio_resid > 0)
 2090                         printf("EEK! readdirrpc resid > 0\n");
 2091                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2092                 *cookiep = cookie;
 2093         }
 2094 nfsmout:
 2095         return (error);
 2096 }
 2097 
 2098 /*
 2099  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2100  */
 2101 int
 2102 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2103 {
 2104         int len, left;
 2105         struct dirent *dp;
 2106         u_int32_t *tl;
 2107         caddr_t cp;
 2108         struct vnode *newvp;
 2109         nfsuint64 *cookiep;
 2110         caddr_t bpos, dpos, dpossav1, dpossav2;
 2111         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2112         struct nameidata nami, *ndp = &nami;
 2113         struct componentname *cnp = &ndp->ni_cnd;
 2114         nfsuint64 cookie;
 2115         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2116         struct nfsnode *dnp = VTONFS(vp), *np;
 2117         nfsfh_t *fhp;
 2118         u_quad_t fileno;
 2119         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2120         int attrflag, fhsize;
 2121 
 2122 #ifndef nolint
 2123         dp = NULL;
 2124 #endif
 2125 #ifndef DIAGNOSTIC
 2126         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2127                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2128                 panic("nfs readdirplusrpc bad uio");
 2129 #endif
 2130         ndp->ni_dvp = vp;
 2131         newvp = NULLVP;
 2132 
 2133         /*
 2134          * If there is no cookie, assume directory was stale.
 2135          */
 2136         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2137         if (cookiep)
 2138                 cookie = *cookiep;
 2139         else
 2140                 return (NFSERR_BAD_COOKIE);
 2141         /*
 2142          * Loop around doing readdir rpc's of size nm_readdirsize
 2143          * truncated to a multiple of DIRBLKSIZ.
 2144          * The stopping criteria is EOF or buffer full.
 2145          */
 2146         while (more_dirs && bigenough) {
 2147                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2148                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2149                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2150                 mb = mreq;
 2151                 bpos = mtod(mb, caddr_t);
 2152                 nfsm_fhtom(vp, 1);
 2153                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2154                 *tl++ = cookie.nfsuquad[0];
 2155                 *tl++ = cookie.nfsuquad[1];
 2156                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2157                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2158                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2159                 *tl = txdr_unsigned(nmp->nm_rsize);
 2160                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2161                 nfsm_postop_attr(vp, attrflag);
 2162                 if (error) {
 2163                         m_freem(mrep);
 2164                         goto nfsmout;
 2165                 }
 2166                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2167                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2168                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2169                 more_dirs = fxdr_unsigned(int, *tl);
 2170 
 2171                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2172                 while (more_dirs && bigenough) {
 2173                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2174                         fileno = fxdr_hyper(tl);
 2175                         len = fxdr_unsigned(int, *(tl + 2));
 2176                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2177                                 error = EBADRPC;
 2178                                 m_freem(mrep);
 2179                                 goto nfsmout;
 2180                         }
 2181                         tlen = nfsm_rndup(len);
 2182                         if (tlen == len)
 2183                                 tlen += 4;      /* To ensure null termination*/
 2184                         left = DIRBLKSIZ - blksiz;
 2185                         if ((tlen + DIRHDSIZ) > left) {
 2186                                 dp->d_reclen += left;
 2187                                 uiop->uio_iov->iov_base =
 2188                                     (char *)uiop->uio_iov->iov_base + left;
 2189                                 uiop->uio_iov->iov_len -= left;
 2190                                 uiop->uio_offset += left;
 2191                                 uiop->uio_resid -= left;
 2192                                 blksiz = 0;
 2193                         }
 2194                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2195                                 bigenough = 0;
 2196                         if (bigenough) {
 2197                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2198                                 dp->d_fileno = (int)fileno;
 2199                                 dp->d_namlen = len;
 2200                                 dp->d_reclen = tlen + DIRHDSIZ;
 2201                                 dp->d_type = DT_UNKNOWN;
 2202                                 blksiz += dp->d_reclen;
 2203                                 if (blksiz == DIRBLKSIZ)
 2204                                         blksiz = 0;
 2205                                 uiop->uio_offset += DIRHDSIZ;
 2206                                 uiop->uio_resid -= DIRHDSIZ;
 2207                                 uiop->uio_iov->iov_base =
 2208                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2209                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2210                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2211                                 cnp->cn_namelen = len;
 2212                                 nfsm_mtouio(uiop, len);
 2213                                 cp = uiop->uio_iov->iov_base;
 2214                                 tlen -= len;
 2215                                 *cp = '\0';
 2216                                 uiop->uio_iov->iov_base =
 2217                                     (char *)uiop->uio_iov->iov_base + tlen;
 2218                                 uiop->uio_iov->iov_len -= tlen;
 2219                                 uiop->uio_offset += tlen;
 2220                                 uiop->uio_resid -= tlen;
 2221                         } else
 2222                                 nfsm_adv(nfsm_rndup(len));
 2223                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2224                         if (bigenough) {
 2225                                 cookie.nfsuquad[0] = *tl++;
 2226                                 cookie.nfsuquad[1] = *tl++;
 2227                         } else
 2228                                 tl += 2;
 2229 
 2230                         /*
 2231                          * Since the attributes are before the file handle
 2232                          * (sigh), we must skip over the attributes and then
 2233                          * come back and get them.
 2234                          */
 2235                         attrflag = fxdr_unsigned(int, *tl);
 2236                         if (attrflag) {
 2237                             dpossav1 = dpos;
 2238                             mdsav1 = md;
 2239                             nfsm_adv(NFSX_V3FATTR);
 2240                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2241                             doit = fxdr_unsigned(int, *tl);
 2242                             /*
 2243                              * Skip loading the attrs for "..". There's a 
 2244                              * race between loading the attrs here and 
 2245                              * lookups that look for the directory currently
 2246                              * being read (in the parent). We try to acquire
 2247                              * the exclusive lock on ".." here, owning the 
 2248                              * lock on the directory being read. Lookup will
 2249                              * hold the lock on ".." and try to acquire the 
 2250                              * lock on the directory being read.
 2251                              * 
 2252                              * There are other ways of fixing this, one would
 2253                              * be to do a trylock on the ".." vnode and skip
 2254                              * loading the attrs on ".." if it happens to be 
 2255                              * locked by another process. But skipping the
 2256                              * attrload on ".." seems the easiest option.
 2257                              */
 2258                             if (strcmp(dp->d_name, "..") == 0) {
 2259                                     doit = 0;
 2260                                     /*
 2261                                      * We've already skipped over the attrs, 
 2262                                      * skip over the filehandle. And store d_type
 2263                                      * as VDIR.
 2264                                      */
 2265                                     tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2266                                     i = fxdr_unsigned(int, *tl);
 2267                                     nfsm_adv(nfsm_rndup(i));
 2268                                     dp->d_type = IFTODT(VTTOIF(VDIR));
 2269                             }       
 2270                             if (doit) {
 2271                                 nfsm_getfh(fhp, fhsize, 1);
 2272                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2273                                     VREF(vp);
 2274                                     newvp = vp;
 2275                                     np = dnp;
 2276                                 } else {
 2277                                     error = nfs_nget(vp->v_mount, fhp,
 2278                                         fhsize, &np, LK_EXCLUSIVE);
 2279                                     if (error)
 2280                                         doit = 0;
 2281                                     else
 2282                                         newvp = NFSTOV(np);
 2283                                 }
 2284                             }
 2285                             if (doit && bigenough) {
 2286                                 dpossav2 = dpos;
 2287                                 dpos = dpossav1;
 2288                                 mdsav2 = md;
 2289                                 md = mdsav1;
 2290                                 nfsm_loadattr(newvp, NULL);
 2291                                 dpos = dpossav2;
 2292                                 md = mdsav2;
 2293                                 dp->d_type =
 2294                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2295                                 ndp->ni_vp = newvp;
 2296                                 /* Update n_ctime, so subsequent lookup doesn't purge entry */
 2297                                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
 2298                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2299                             }
 2300                         } else {
 2301                             /* Just skip over the file handle */
 2302                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2303                             i = fxdr_unsigned(int, *tl);
 2304                             if (i) {
 2305                                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2306                                 fhsize = fxdr_unsigned(int, *tl);
 2307                                 nfsm_adv(nfsm_rndup(fhsize));
 2308                             }
 2309                         }
 2310                         if (newvp != NULLVP) {
 2311                             if (newvp == vp)
 2312                                 vrele(newvp);
 2313                             else
 2314                                 vput(newvp);
 2315                             newvp = NULLVP;
 2316                         }
 2317                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2318                         more_dirs = fxdr_unsigned(int, *tl);
 2319                 }
 2320                 /*
 2321                  * If at end of rpc data, get the eof boolean
 2322                  */
 2323                 if (!more_dirs) {
 2324                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2325                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2326                 }
 2327                 m_freem(mrep);
 2328         }
 2329         /*
 2330          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2331          * by increasing d_reclen for the last record.
 2332          */
 2333         if (blksiz > 0) {
 2334                 left = DIRBLKSIZ - blksiz;
 2335                 dp->d_reclen += left;
 2336                 uiop->uio_iov->iov_base =
 2337                     (char *)uiop->uio_iov->iov_base + left;
 2338                 uiop->uio_iov->iov_len -= left;
 2339                 uiop->uio_offset += left;
 2340                 uiop->uio_resid -= left;
 2341         }
 2342 
 2343         /*
 2344          * We are now either at the end of the directory or have filled the
 2345          * block.
 2346          */
 2347         if (bigenough)
 2348                 dnp->n_direofoffset = uiop->uio_offset;
 2349         else {
 2350                 if (uiop->uio_resid > 0)
 2351                         printf("EEK! readdirplusrpc resid > 0\n");
 2352                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2353                 *cookiep = cookie;
 2354         }
 2355 nfsmout:
 2356         if (newvp != NULLVP) {
 2357                 if (newvp == vp)
 2358                         vrele(newvp);
 2359                 else
 2360                         vput(newvp);
 2361                 newvp = NULLVP;
 2362         }
 2363         return (error);
 2364 }
 2365 
 2366 /*
 2367  * Silly rename. To make the NFS filesystem that is stateless look a little
 2368  * more like the "ufs" a remove of an active vnode is translated to a rename
 2369  * to a funny looking filename that is removed by nfs_inactive on the
 2370  * nfsnode. There is the potential for another process on a different client
 2371  * to create the same funny name between the nfs_lookitup() fails and the
 2372  * nfs_rename() completes, but...
 2373  */
 2374 static int
 2375 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2376 {
 2377         struct sillyrename *sp;
 2378         struct nfsnode *np;
 2379         int error;
 2380         short pid;
 2381         unsigned int lticks;
 2382 
 2383         cache_purge(dvp);
 2384         np = VTONFS(vp);
 2385 #ifndef DIAGNOSTIC
 2386         if (vp->v_type == VDIR)
 2387                 panic("nfs: sillyrename dir");
 2388 #endif
 2389         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2390                 M_NFSREQ, M_WAITOK);
 2391         sp->s_cred = crhold(cnp->cn_cred);
 2392         sp->s_dvp = dvp;
 2393         sp->s_removeit = nfs_removeit;
 2394         VREF(dvp);
 2395 
 2396         /* 
 2397          * Fudge together a funny name.
 2398          * Changing the format of the funny name to accomodate more 
 2399          * sillynames per directory.
 2400          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2401          * CPU ticks since boot.
 2402          */
 2403         pid = cnp->cn_thread->td_proc->p_pid;
 2404         lticks = (unsigned int)ticks;
 2405         for ( ; ; ) {
 2406                 sp->s_namlen = sprintf(sp->s_name, 
 2407                                        ".nfs.%08x.%04x4.4", lticks, 
 2408                                        pid);
 2409                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2410                                  cnp->cn_thread, NULL))
 2411                         break;
 2412                 lticks++;
 2413         }
 2414         error = nfs_renameit(dvp, cnp, sp);
 2415         if (error)
 2416                 goto bad;
 2417         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2418                 cnp->cn_thread, &np);
 2419         np->n_sillyrename = sp;
 2420         return (0);
 2421 bad:
 2422         vrele(sp->s_dvp);
 2423         crfree(sp->s_cred);
 2424         free((caddr_t)sp, M_NFSREQ);
 2425         return (error);
 2426 }
 2427 
 2428 /*
 2429  * Look up a file name and optionally either update the file handle or
 2430  * allocate an nfsnode, depending on the value of npp.
 2431  * npp == NULL  --> just do the lookup
 2432  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2433  *                      handled too
 2434  * *npp != NULL --> update the file handle in the vnode
 2435  */
 2436 static int
 2437 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2438     struct thread *td, struct nfsnode **npp)
 2439 {
 2440         struct vnode *newvp = NULL;
 2441         struct nfsnode *np, *dnp = VTONFS(dvp);
 2442         caddr_t bpos, dpos;
 2443         int error = 0, fhlen, attrflag;
 2444         struct mbuf *mreq, *mrep, *md, *mb;
 2445         nfsfh_t *nfhp;
 2446         int v3 = NFS_ISV3(dvp);
 2447 
 2448         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2449         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2450                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2451         mb = mreq;
 2452         bpos = mtod(mb, caddr_t);
 2453         nfsm_fhtom(dvp, v3);
 2454         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2455         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2456         if (npp && !error) {
 2457                 nfsm_getfh(nfhp, fhlen, v3);
 2458                 if (*npp) {
 2459                     np = *npp;
 2460                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2461                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2462                         np->n_fhp = &np->n_fh;
 2463                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2464                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2465                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2466                     np->n_fhsize = fhlen;
 2467                     newvp = NFSTOV(np);
 2468                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2469                     VREF(dvp);
 2470                     newvp = dvp;
 2471                 } else {
 2472                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
 2473                     if (error) {
 2474                         m_freem(mrep);
 2475                         return (error);
 2476                     }
 2477                     newvp = NFSTOV(np);
 2478                 }
 2479                 if (v3) {
 2480                         nfsm_postop_attr(newvp, attrflag);
 2481                         if (!attrflag && *npp == NULL) {
 2482                                 m_freem(mrep);
 2483                                 if (newvp == dvp)
 2484                                         vrele(newvp);
 2485                                 else
 2486                                         vput(newvp);
 2487                                 return (ENOENT);
 2488                         }
 2489                 } else
 2490                         nfsm_loadattr(newvp, NULL);
 2491         }
 2492         m_freem(mrep);
 2493 nfsmout:
 2494         if (npp && *npp == NULL) {
 2495                 if (error) {
 2496                         if (newvp) {
 2497                                 if (newvp == dvp)
 2498                                         vrele(newvp);
 2499                                 else
 2500                                         vput(newvp);
 2501                         }
 2502                 } else
 2503                         *npp = np;
 2504         }
 2505         return (error);
 2506 }
 2507 
 2508 /*
 2509  * Nfs Version 3 commit rpc
 2510  */
 2511 int
 2512 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2513     struct thread *td)
 2514 {
 2515         u_int32_t *tl;
 2516         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2517         caddr_t bpos, dpos;
 2518         int error = 0, wccflag = NFSV3_WCCRATTR;
 2519         struct mbuf *mreq, *mrep, *md, *mb;
 2520 
 2521         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2522                 return (0);
 2523         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2524         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2525         mb = mreq;
 2526         bpos = mtod(mb, caddr_t);
 2527         nfsm_fhtom(vp, 1);
 2528         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2529         txdr_hyper(offset, tl);
 2530         tl += 2;
 2531         *tl = txdr_unsigned(cnt);
 2532         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2533         nfsm_wcc_data(vp, wccflag);
 2534         if (!error) {
 2535                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2536                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2537                         NFSX_V3WRITEVERF)) {
 2538                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2539                                 NFSX_V3WRITEVERF);
 2540                         error = NFSERR_STALEWRITEVERF;
 2541                 }
 2542         }
 2543         m_freem(mrep);
 2544 nfsmout:
 2545         return (error);
 2546 }
 2547 
 2548 /*
 2549  * Strategy routine.
 2550  * For async requests when nfsiod(s) are running, queue the request by
 2551  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2552  * request.
 2553  */
 2554 static int
 2555 nfs_strategy(struct vop_strategy_args *ap)
 2556 {
 2557         struct buf *bp = ap->a_bp;
 2558         struct ucred *cr;
 2559 
 2560         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2561         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2562 
 2563         if (bp->b_iocmd == BIO_READ)
 2564                 cr = bp->b_rcred;
 2565         else
 2566                 cr = bp->b_wcred;
 2567 
 2568         /*
 2569          * If the op is asynchronous and an i/o daemon is waiting
 2570          * queue the request, wake it up and wait for completion
 2571          * otherwise just do it ourselves.
 2572          */
 2573         if ((bp->b_flags & B_ASYNC) == 0 ||
 2574             nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
 2575                 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
 2576         return (0);
 2577 }
 2578 
 2579 /*
 2580  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2581  */
 2582 /* ARGSUSED */
 2583 static int
 2584 nfs_fsync(struct vop_fsync_args *ap)
 2585 {
 2586 
 2587         return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
 2588 }
 2589 
 2590 /*
 2591  * Flush all the blocks associated with a vnode.
 2592  *      Walk through the buffer pool and push any dirty pages
 2593  *      associated with the vnode.
 2594  */
 2595 static int
 2596 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
 2597     int commit)
 2598 {
 2599         struct nfsnode *np = VTONFS(vp);
 2600         struct buf *bp;
 2601         int i;
 2602         struct buf *nbp;
 2603         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2604         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2605         int passone = 1;
 2606         u_quad_t off, endoff, toff;
 2607         struct ucred* wcred = NULL;
 2608         struct buf **bvec = NULL;
 2609 #ifndef NFS_COMMITBVECSIZ
 2610 #define NFS_COMMITBVECSIZ       20
 2611 #endif
 2612         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2613         int bvecsize = 0, bveccount;
 2614 
 2615         if (nmp->nm_flag & NFSMNT_INT)
 2616                 slpflag = PCATCH;
 2617         if (!commit)
 2618                 passone = 0;
 2619         /*
 2620          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2621          * server, but has not been committed to stable storage on the server
 2622          * yet. On the first pass, the byte range is worked out and the commit
 2623          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2624          * job.
 2625          */
 2626 again:
 2627         off = (u_quad_t)-1;
 2628         endoff = 0;
 2629         bvecpos = 0;
 2630         if (NFS_ISV3(vp) && commit) {
 2631                 s = splbio();
 2632                 if (bvec != NULL && bvec != bvec_on_stack)
 2633                         free(bvec, M_TEMP);
 2634                 /*
 2635                  * Count up how many buffers waiting for a commit.
 2636                  */
 2637                 bveccount = 0;
 2638                 VI_LOCK(vp);
 2639                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2640                         if (BUF_REFCNT(bp) == 0 &&
 2641                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2642                                 == (B_DELWRI | B_NEEDCOMMIT))
 2643                                 bveccount++;
 2644                 }
 2645                 /*
 2646                  * Allocate space to remember the list of bufs to commit.  It is
 2647                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2648                  * If we can't get memory (for whatever reason), we will end up
 2649                  * committing the buffers one-by-one in the loop below.
 2650                  */
 2651                 if (bveccount > NFS_COMMITBVECSIZ) {
 2652                         /*
 2653                          * Release the vnode interlock to avoid a lock
 2654                          * order reversal.
 2655                          */
 2656                         VI_UNLOCK(vp);
 2657                         bvec = (struct buf **)
 2658                                 malloc(bveccount * sizeof(struct buf *),
 2659                                        M_TEMP, M_NOWAIT);
 2660                         VI_LOCK(vp);
 2661                         if (bvec == NULL) {
 2662                                 bvec = bvec_on_stack;
 2663                                 bvecsize = NFS_COMMITBVECSIZ;
 2664                         } else
 2665                                 bvecsize = bveccount;
 2666                 } else {
 2667                         bvec = bvec_on_stack;
 2668                         bvecsize = NFS_COMMITBVECSIZ;
 2669                 }
 2670                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2671                         if (bvecpos >= bvecsize)
 2672                                 break;
 2673                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2674                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2675                                 continue;
 2676                         }
 2677                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2678                             (B_DELWRI | B_NEEDCOMMIT)) {
 2679                                 BUF_UNLOCK(bp);
 2680                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2681                                 continue;
 2682                         }
 2683                         VI_UNLOCK(vp);
 2684                         bremfree(bp);
 2685                         /*
 2686                          * Work out if all buffers are using the same cred
 2687                          * so we can deal with them all with one commit.
 2688                          *
 2689                          * NOTE: we are not clearing B_DONE here, so we have
 2690                          * to do it later on in this routine if we intend to
 2691                          * initiate I/O on the bp.
 2692                          *
 2693                          * Note: to avoid loopback deadlocks, we do not
 2694                          * assign b_runningbufspace.
 2695                          */
 2696                         if (wcred == NULL)
 2697                                 wcred = bp->b_wcred;
 2698                         else if (wcred != bp->b_wcred)
 2699                                 wcred = NOCRED;
 2700                         vfs_busy_pages(bp, 1);
 2701 
 2702                         VI_LOCK(vp);
 2703                         /*
 2704                          * bp is protected by being locked, but nbp is not
 2705                          * and vfs_busy_pages() may sleep.  We have to
 2706                          * recalculate nbp.
 2707                          */
 2708                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2709 
 2710                         /*
 2711                          * A list of these buffers is kept so that the
 2712                          * second loop knows which buffers have actually
 2713                          * been committed. This is necessary, since there
 2714                          * may be a race between the commit rpc and new
 2715                          * uncommitted writes on the file.
 2716                          */
 2717                         bvec[bvecpos++] = bp;
 2718                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2719                                 bp->b_dirtyoff;
 2720                         if (toff < off)
 2721                                 off = toff;
 2722                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2723                         if (toff > endoff)
 2724                                 endoff = toff;
 2725                 }
 2726                 splx(s);
 2727                 VI_UNLOCK(vp);
 2728         }
 2729         if (bvecpos > 0) {
 2730                 /*
 2731                  * Commit data on the server, as required.
 2732                  * If all bufs are using the same wcred, then use that with
 2733                  * one call for all of them, otherwise commit each one
 2734                  * separately.
 2735                  */
 2736                 if (wcred != NOCRED)
 2737                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2738                                           wcred, td);
 2739                 else {
 2740                         retv = 0;
 2741                         for (i = 0; i < bvecpos; i++) {
 2742                                 off_t off, size;
 2743                                 bp = bvec[i];
 2744                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2745                                         bp->b_dirtyoff;
 2746                                 size = (u_quad_t)(bp->b_dirtyend
 2747                                                   - bp->b_dirtyoff);
 2748                                 retv = nfs_commit(vp, off, (int)size,
 2749                                                   bp->b_wcred, td);
 2750                                 if (retv) break;
 2751                         }
 2752                 }
 2753 
 2754                 if (retv == NFSERR_STALEWRITEVERF)
 2755                         nfs_clearcommit(vp->v_mount);
 2756 
 2757                 /*
 2758                  * Now, either mark the blocks I/O done or mark the
 2759                  * blocks dirty, depending on whether the commit
 2760                  * succeeded.
 2761                  */
 2762                 for (i = 0; i < bvecpos; i++) {
 2763                         bp = bvec[i];
 2764                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2765                         if (retv) {
 2766                                 /*
 2767                                  * Error, leave B_DELWRI intact
 2768                                  */
 2769                                 vfs_unbusy_pages(bp);
 2770                                 brelse(bp);
 2771                         } else {
 2772                                 /*
 2773                                  * Success, remove B_DELWRI ( bundirty() ).
 2774                                  *
 2775                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2776                                  * specific.  We should probably move that
 2777                                  * into bundirty(). XXX
 2778                                  */
 2779                                 s = splbio();
 2780                                 bufobj_wref(&vp->v_bufobj);
 2781                                 bp->b_flags |= B_ASYNC;
 2782                                 bundirty(bp);
 2783                                 bp->b_flags &= ~B_DONE;
 2784                                 bp->b_ioflags &= ~BIO_ERROR;
 2785                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2786                                 splx(s);
 2787                                 bufdone(bp);
 2788                         }
 2789                 }
 2790         }
 2791 
 2792         /*
 2793          * Start/do any write(s) that are required.
 2794          */
 2795 loop:
 2796         s = splbio();
 2797         VI_LOCK(vp);
 2798         TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2799                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2800                         if (waitfor != MNT_WAIT || passone)
 2801                                 continue;
 2802 
 2803                         error = BUF_TIMELOCK(bp,
 2804                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2805                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2806                         splx(s);
 2807                         if (error == 0)
 2808                                 panic("nfs_fsync: inconsistent lock");
 2809                         if (error == ENOLCK)
 2810                                 goto loop;
 2811                         if (nfs_sigintr(nmp, NULL, td)) {
 2812                                 error = EINTR;
 2813                                 goto done;
 2814                         }
 2815                         if (slpflag == PCATCH) {
 2816                                 slpflag = 0;
 2817                                 slptimeo = 2 * hz;
 2818                         }
 2819                         goto loop;
 2820                 }
 2821                 if ((bp->b_flags & B_DELWRI) == 0)
 2822                         panic("nfs_fsync: not dirty");
 2823                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2824                         BUF_UNLOCK(bp);
 2825                         continue;
 2826                 }
 2827                 VI_UNLOCK(vp);
 2828                 bremfree(bp);
 2829                 if (passone || !commit)
 2830                     bp->b_flags |= B_ASYNC;
 2831                 else
 2832                     bp->b_flags |= B_ASYNC;
 2833                 splx(s);
 2834                 bwrite(bp);
 2835                 if (nfs_sigintr(nmp, NULL, td)) {
 2836                         error = EINTR;
 2837                         goto done;
 2838                 }
 2839                 goto loop;
 2840         }
 2841         splx(s);
 2842         if (passone) {
 2843                 passone = 0;
 2844                 VI_UNLOCK(vp);
 2845                 goto again;
 2846         }
 2847         if (waitfor == MNT_WAIT) {
 2848                 while (vp->v_bufobj.bo_numoutput) {
 2849                         error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
 2850                         if (error) {
 2851                             VI_UNLOCK(vp);
 2852                             error = nfs_sigintr(nmp, NULL, td);
 2853                             if (error)
 2854                                 goto done;
 2855                             if (slpflag == PCATCH) {
 2856                                 slpflag = 0;
 2857                                 slptimeo = 2 * hz;
 2858                             }
 2859                             VI_LOCK(vp);
 2860                         }
 2861                 }
 2862                 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
 2863                         VI_UNLOCK(vp);
 2864                         goto loop;
 2865                 }
 2866                 VI_UNLOCK(vp);
 2867                 /* 
 2868                  * Wait for all the async IO requests to drain 
 2869                  */
 2870                 while (np->n_directio_asyncwr > 0) {
 2871                         np->n_flag |= NFSYNCWAIT;
 2872                         error = nfs_tsleep(td, (caddr_t)&np->n_directio_asyncwr,
 2873                                            slpflag | (PRIBIO + 1), "nfsfsync", 0);
 2874                         if (error) {
 2875                                 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
 2876                                         error = EINTR;
 2877                                         goto done;
 2878                                 }
 2879                         }
 2880                 }
 2881 
 2882         } else
 2883                 VI_UNLOCK(vp);
 2884         if (np->n_flag & NWRITEERR) {
 2885                 error = np->n_error;
 2886                 np->n_flag &= ~NWRITEERR;
 2887         }
 2888         if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
 2889             vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
 2890                 np->n_flag &= ~NMODIFIED;
 2891 done:
 2892         if (bvec != NULL && bvec != bvec_on_stack)
 2893                 free(bvec, M_TEMP);
 2894         return (error);
 2895 }
 2896 
 2897 /*
 2898  * NFS advisory byte-level locks.
 2899  */
 2900 static int
 2901 nfs_advlock(struct vop_advlock_args *ap)
 2902 {
 2903 
 2904         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2905                 struct nfsnode *np = VTONFS(ap->a_vp);
 2906 
 2907                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2908         }
 2909         return (nfs_dolock(ap));
 2910 }
 2911 
 2912 /*
 2913  * NFS advisory byte-level locks.
 2914  */
 2915 static int
 2916 nfs_advlockasync(struct vop_advlockasync_args *ap)
 2917 {
 2918         int error;
 2919         
 2920         mtx_lock(&Giant);
 2921         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2922                 struct nfsnode *np = VTONFS(ap->a_vp);
 2923 
 2924                 error = lf_advlockasync(ap, &(np->n_lockf), np->n_size);
 2925                 goto out;
 2926         }
 2927         error = EOPNOTSUPP;
 2928 out:    
 2929         mtx_unlock(&Giant);
 2930         return (error);
 2931 }
 2932 
 2933 /*
 2934  * Print out the contents of an nfsnode.
 2935  */
 2936 static int
 2937 nfs_print(struct vop_print_args *ap)
 2938 {
 2939         struct vnode *vp = ap->a_vp;
 2940         struct nfsnode *np = VTONFS(vp);
 2941 
 2942         printf("\tfileid %ld fsid 0x%x",
 2943            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2944         if (vp->v_type == VFIFO)
 2945                 fifo_printinfo(vp);
 2946         printf("\n");
 2947         return (0);
 2948 }
 2949 
 2950 /*
 2951  * This is the "real" nfs::bwrite(struct buf*).
 2952  * We set B_CACHE if this is a VMIO buffer.
 2953  */
 2954 int
 2955 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
 2956 {
 2957         int s;
 2958         int oldflags = bp->b_flags;
 2959 #if 0
 2960         int retv = 1;
 2961         off_t off;
 2962 #endif
 2963 
 2964         if (BUF_REFCNT(bp) == 0)
 2965                 panic("bwrite: buffer is not locked???");
 2966 
 2967         if (bp->b_flags & B_INVAL) {
 2968                 brelse(bp);
 2969                 return(0);
 2970         }
 2971 
 2972         bp->b_flags |= B_CACHE;
 2973 
 2974         /*
 2975          * Undirty the bp.  We will redirty it later if the I/O fails.
 2976          */
 2977 
 2978         s = splbio();
 2979         bundirty(bp);
 2980         bp->b_flags &= ~B_DONE;
 2981         bp->b_ioflags &= ~BIO_ERROR;
 2982         bp->b_iocmd = BIO_WRITE;
 2983 
 2984         bufobj_wref(bp->b_bufobj);
 2985         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2986         splx(s);
 2987 
 2988         /*
 2989          * Note: to avoid loopback deadlocks, we do not
 2990          * assign b_runningbufspace.
 2991          */
 2992         vfs_busy_pages(bp, 1);
 2993 
 2994         BUF_KERNPROC(bp);
 2995         bp->b_iooffset = dbtob(bp->b_blkno);
 2996         bstrategy(bp);
 2997 
 2998         if( (oldflags & B_ASYNC) == 0) {
 2999                 int rtval = bufwait(bp);
 3000 
 3001                 if (oldflags & B_DELWRI) {
 3002                         s = splbio();
 3003                         reassignbuf(bp);
 3004                         splx(s);
 3005                 }
 3006 
 3007                 brelse(bp);
 3008                 return (rtval);
 3009         }
 3010 
 3011         return (0);
 3012 }
 3013 
 3014 /*
 3015  * nfs special file access vnode op.
 3016  * Essentially just get vattr and then imitate iaccess() since the device is
 3017  * local to the client.
 3018  */
 3019 static int
 3020 nfsspec_access(struct vop_access_args *ap)
 3021 {
 3022         struct vattr *vap;
 3023         struct ucred *cred = ap->a_cred;
 3024         struct vnode *vp = ap->a_vp;
 3025         mode_t mode = ap->a_mode;
 3026         struct vattr vattr;
 3027         int error;
 3028 
 3029         /*
 3030          * Disallow write attempts on filesystems mounted read-only;
 3031          * unless the file is a socket, fifo, or a block or character
 3032          * device resident on the filesystem.
 3033          */
 3034         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3035                 switch (vp->v_type) {
 3036                 case VREG:
 3037                 case VDIR:
 3038                 case VLNK:
 3039                         return (EROFS);
 3040                 default:
 3041                         break;
 3042                 }
 3043         }
 3044         vap = &vattr;
 3045         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3046         if (error)
 3047                 return (error);
 3048         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3049             mode, cred, NULL));
 3050 }
 3051 
 3052 /*
 3053  * Read wrapper for fifos.
 3054  */
 3055 static int
 3056 nfsfifo_read(struct vop_read_args *ap)
 3057 {
 3058         struct nfsnode *np = VTONFS(ap->a_vp);
 3059 
 3060         /*
 3061          * Set access flag.
 3062          */
 3063         np->n_flag |= NACC;
 3064         getnanotime(&np->n_atim);
 3065         return (fifo_specops.vop_read(ap));
 3066 }
 3067 
 3068 /*
 3069  * Write wrapper for fifos.
 3070  */
 3071 static int
 3072 nfsfifo_write(struct vop_write_args *ap)
 3073 {
 3074         struct nfsnode *np = VTONFS(ap->a_vp);
 3075 
 3076         /*
 3077          * Set update flag.
 3078          */
 3079         np->n_flag |= NUPD;
 3080         getnanotime(&np->n_mtim);
 3081         return (fifo_specops.vop_write(ap));
 3082 }
 3083 
 3084 /*
 3085  * Close wrapper for fifos.
 3086  *
 3087  * Update the times on the nfsnode then do fifo close.
 3088  */
 3089 static int
 3090 nfsfifo_close(struct vop_close_args *ap)
 3091 {
 3092         struct vnode *vp = ap->a_vp;
 3093         struct nfsnode *np = VTONFS(vp);
 3094         struct vattr vattr;
 3095         struct timespec ts;
 3096 
 3097         if (np->n_flag & (NACC | NUPD)) {
 3098                 getnanotime(&ts);
 3099                 if (np->n_flag & NACC)
 3100                         np->n_atim = ts;
 3101                 if (np->n_flag & NUPD)
 3102                         np->n_mtim = ts;
 3103                 np->n_flag |= NCHG;
 3104                 if (vrefcnt(vp) == 1 &&
 3105                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3106                         VATTR_NULL(&vattr);
 3107                         if (np->n_flag & NACC)
 3108                                 vattr.va_atime = np->n_atim;
 3109                         if (np->n_flag & NUPD)
 3110                                 vattr.va_mtime = np->n_mtim;
 3111                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3112                 }
 3113         }
 3114         return (fifo_specops.vop_close(ap));
 3115 }
 3116 
 3117 /*
 3118  * Just call nfs_writebp() with the force argument set to 1.
 3119  *
 3120  * NOTE: B_DONE may or may not be set in a_bp on call.
 3121  */
 3122 static int
 3123 nfs_bwrite(struct buf *bp)
 3124 {
 3125 
 3126         return (nfs_writebp(bp, 1, curthread));
 3127 }
 3128 
 3129 struct buf_ops buf_ops_nfs = {
 3130         .bop_name       =       "buf_ops_nfs",
 3131         .bop_write      =       nfs_bwrite,
 3132         .bop_strategy   =       bufstrategy,
 3133         .bop_sync       =       bufsync,
 3134         .bop_bdflush    =       bufbdflush,
 3135 };

Cache object: ac516d346f9b45bd261a1d29f890f97f


[ 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.