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


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

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

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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Rick Macklem at The University of Guelph.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: releng/6.0/sys/nfsclient/nfs_vnops.c 153538 2005-12-19 10:58:57Z delphij $");
   37 
   38 /*
   39  * vnode op calls for Sun NFS version 2 and 3
   40  */
   41 
   42 #include "opt_inet.h"
   43 
   44 #include <sys/param.h>
   45 #include <sys/kernel.h>
   46 #include <sys/systm.h>
   47 #include <sys/resourcevar.h>
   48 #include <sys/proc.h>
   49 #include <sys/mount.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/malloc.h>
   53 #include <sys/mbuf.h>
   54 #include <sys/namei.h>
   55 #include <sys/socket.h>
   56 #include <sys/vnode.h>
   57 #include <sys/dirent.h>
   58 #include <sys/fcntl.h>
   59 #include <sys/lockf.h>
   60 #include <sys/stat.h>
   61 #include <sys/sysctl.h>
   62 #include <sys/signalvar.h>
   63 
   64 #include <vm/vm.h>
   65 #include <vm/vm_object.h>
   66 #include <vm/vm_extern.h>
   67 #include <vm/vm_object.h>
   68 
   69 #include <fs/fifofs/fifo.h>
   70 
   71 #include <rpc/rpcclnt.h>
   72 
   73 #include <nfs/rpcv2.h>
   74 #include <nfs/nfsproto.h>
   75 #include <nfsclient/nfs.h>
   76 #include <nfsclient/nfsnode.h>
   77 #include <nfsclient/nfsmount.h>
   78 #include <nfsclient/nfs_lock.h>
   79 #include <nfs/xdr_subs.h>
   80 #include <nfsclient/nfsm_subs.h>
   81 
   82 #include <net/if.h>
   83 #include <netinet/in.h>
   84 #include <netinet/in_var.h>
   85 
   86 /* Defs */
   87 #define TRUE    1
   88 #define FALSE   0
   89 
   90 /*
   91  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
   92  * calls are not in getblk() and brelse() so that they would not be necessary
   93  * here.
   94  */
   95 #ifndef B_VMIO
   96 #define vfs_busy_pages(bp, f)
   97 #endif
   98 
   99 static vop_read_t       nfsfifo_read;
  100 static vop_write_t      nfsfifo_write;
  101 static vop_close_t      nfsfifo_close;
  102 static int      nfs_flush(struct vnode *, int, struct thread *,
  103                     int);
  104 static int      nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
  105                     struct thread *);
  106 static vop_lookup_t     nfs_lookup;
  107 static vop_create_t     nfs_create;
  108 static vop_mknod_t      nfs_mknod;
  109 static vop_open_t       nfs_open;
  110 static vop_close_t      nfs_close;
  111 static vop_access_t     nfs_access;
  112 static vop_getattr_t    nfs_getattr;
  113 static vop_setattr_t    nfs_setattr;
  114 static vop_read_t       nfs_read;
  115 static vop_fsync_t      nfs_fsync;
  116 static vop_remove_t     nfs_remove;
  117 static vop_link_t       nfs_link;
  118 static vop_rename_t     nfs_rename;
  119 static vop_mkdir_t      nfs_mkdir;
  120 static vop_rmdir_t      nfs_rmdir;
  121 static vop_symlink_t    nfs_symlink;
  122 static vop_readdir_t    nfs_readdir;
  123 static vop_strategy_t   nfs_strategy;
  124 static  int     nfs_lookitup(struct vnode *, const char *, int,
  125                     struct ucred *, struct thread *, struct nfsnode **);
  126 static  int     nfs_sillyrename(struct vnode *, struct vnode *,
  127                     struct componentname *);
  128 static vop_access_t     nfsspec_access;
  129 static vop_readlink_t   nfs_readlink;
  130 static vop_print_t      nfs_print;
  131 static vop_advlock_t    nfs_advlock;
  132 
  133 /*
  134  * Global vfs data structures for nfs
  135  */
  136 struct vop_vector nfs_vnodeops = {
  137         .vop_default =          &default_vnodeops,
  138         .vop_access =           nfs_access,
  139         .vop_advlock =          nfs_advlock,
  140         .vop_close =            nfs_close,
  141         .vop_create =           nfs_create,
  142         .vop_fsync =            nfs_fsync,
  143         .vop_getattr =          nfs_getattr,
  144         .vop_getpages =         nfs_getpages,
  145         .vop_putpages =         nfs_putpages,
  146         .vop_inactive =         nfs_inactive,
  147         .vop_lease =            VOP_NULL,
  148         .vop_link =             nfs_link,
  149         .vop_lookup =           nfs_lookup,
  150         .vop_mkdir =            nfs_mkdir,
  151         .vop_mknod =            nfs_mknod,
  152         .vop_open =             nfs_open,
  153         .vop_print =            nfs_print,
  154         .vop_read =             nfs_read,
  155         .vop_readdir =          nfs_readdir,
  156         .vop_readlink =         nfs_readlink,
  157         .vop_reclaim =          nfs_reclaim,
  158         .vop_remove =           nfs_remove,
  159         .vop_rename =           nfs_rename,
  160         .vop_rmdir =            nfs_rmdir,
  161         .vop_setattr =          nfs_setattr,
  162         .vop_strategy =         nfs_strategy,
  163         .vop_symlink =          nfs_symlink,
  164         .vop_write =            nfs_write,
  165 };
  166 
  167 struct vop_vector nfs_fifoops = {
  168         .vop_default =          &fifo_specops,
  169         .vop_access =           nfsspec_access,
  170         .vop_close =            nfsfifo_close,
  171         .vop_fsync =            nfs_fsync,
  172         .vop_getattr =          nfs_getattr,
  173         .vop_inactive =         nfs_inactive,
  174         .vop_print =            nfs_print,
  175         .vop_read =             nfsfifo_read,
  176         .vop_reclaim =          nfs_reclaim,
  177         .vop_setattr =          nfs_setattr,
  178         .vop_write =            nfsfifo_write,
  179 };
  180 
  181 static int      nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
  182                              struct componentname *cnp, struct vattr *vap);
  183 static int      nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
  184                               struct ucred *cred, struct thread *td);
  185 static int      nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
  186                               int fnamelen, struct vnode *tdvp,
  187                               const char *tnameptr, int tnamelen,
  188                               struct ucred *cred, struct thread *td);
  189 static int      nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
  190                              struct sillyrename *sp);
  191 
  192 /*
  193  * Global variables
  194  */
  195 struct proc     *nfs_iodwant[NFS_MAXASYNCDAEMON];
  196 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
  197 int              nfs_numasync = 0;
  198 #define DIRHDSIZ        (sizeof (struct dirent) - (MAXNAMLEN + 1))
  199 
  200 SYSCTL_DECL(_vfs_nfs);
  201 
  202 static int      nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
  203 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
  204            &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
  205 
  206 static int      nfsv3_commit_on_close = 0;
  207 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
  208            &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
  209 
  210 static int      nfs_clean_pages_on_close = 1;
  211 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
  212            &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
  213 
  214 int nfs_directio_enable = 0;
  215 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
  216            &nfs_directio_enable, 0, "Enable NFS directio");
  217 
  218 /*
  219  * This sysctl allows other processes to mmap a file that has been opened
  220  * O_DIRECT by a process.  In general, having processes mmap the file while
  221  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
  222  * this by default to prevent DoS attacks - to prevent a malicious user from
  223  * opening up files O_DIRECT preventing other users from mmap'ing these
  224  * files.  "Protected" environments where stricter consistency guarantees are
  225  * required can disable this knob.  The process that opened the file O_DIRECT
  226  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
  227  * meaningful.
  228  */
  229 int nfs_directio_allow_mmap = 1;
  230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
  231            &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
  232 
  233 #if 0
  234 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
  235            &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
  236 
  237 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
  238            &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
  239 #endif
  240 
  241 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY          \
  242                          | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE     \
  243                          | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
  244 static int
  245 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
  246     struct ucred *cred)
  247 {
  248         const int v3 = 1;
  249         u_int32_t *tl;
  250         int error = 0, attrflag;
  251 
  252         struct mbuf *mreq, *mrep, *md, *mb;
  253         caddr_t bpos, dpos;
  254         u_int32_t rmode;
  255         struct nfsnode *np = VTONFS(vp);
  256 
  257         nfsstats.rpccnt[NFSPROC_ACCESS]++;
  258         mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
  259         mb = mreq;
  260         bpos = mtod(mb, caddr_t);
  261         nfsm_fhtom(vp, v3);
  262         tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  263         *tl = txdr_unsigned(wmode);
  264         nfsm_request(vp, NFSPROC_ACCESS, td, cred);
  265         nfsm_postop_attr(vp, attrflag);
  266         if (!error) {
  267                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
  268                 rmode = fxdr_unsigned(u_int32_t, *tl);
  269                 np->n_mode = rmode;
  270                 np->n_modeuid = cred->cr_uid;
  271                 np->n_modestamp = time_second;
  272         }
  273         m_freem(mrep);
  274 nfsmout:
  275         return (error);
  276 }
  277 
  278 /*
  279  * nfs access vnode op.
  280  * For nfs version 2, just return ok. File accesses may fail later.
  281  * For nfs version 3, use the access rpc to check accessibility. If file modes
  282  * are changed on the server, accesses might still fail later.
  283  */
  284 static int
  285 nfs_access(struct vop_access_args *ap)
  286 {
  287         struct vnode *vp = ap->a_vp;
  288         int error = 0;
  289         u_int32_t mode, wmode;
  290         int v3 = NFS_ISV3(vp);
  291         struct nfsnode *np = VTONFS(vp);
  292 
  293         /*
  294          * Disallow write attempts on filesystems mounted read-only;
  295          * unless the file is a socket, fifo, or a block or character
  296          * device resident on the filesystem.
  297          */
  298         if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
  299                 switch (vp->v_type) {
  300                 case VREG:
  301                 case VDIR:
  302                 case VLNK:
  303                         return (EROFS);
  304                 default:
  305                         break;
  306                 }
  307         }
  308         /*
  309          * For nfs v3, check to see if we have done this recently, and if
  310          * so return our cached result instead of making an ACCESS call.
  311          * If not, do an access rpc, otherwise you are stuck emulating
  312          * ufs_access() locally using the vattr. This may not be correct,
  313          * since the server may apply other access criteria such as
  314          * client uid-->server uid mapping that we do not know about.
  315          */
  316         if (v3) {
  317                 if (ap->a_mode & VREAD)
  318                         mode = NFSV3ACCESS_READ;
  319                 else
  320                         mode = 0;
  321                 if (vp->v_type != VDIR) {
  322                         if (ap->a_mode & VWRITE)
  323                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
  324                         if (ap->a_mode & VEXEC)
  325                                 mode |= NFSV3ACCESS_EXECUTE;
  326                 } else {
  327                         if (ap->a_mode & VWRITE)
  328                                 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
  329                                          NFSV3ACCESS_DELETE);
  330                         if (ap->a_mode & VEXEC)
  331                                 mode |= NFSV3ACCESS_LOOKUP;
  332                 }
  333                 /* XXX safety belt, only make blanket request if caching */
  334                 if (nfsaccess_cache_timeout > 0) {
  335                         wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
  336                                 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
  337                                 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
  338                 } else {
  339                         wmode = mode;
  340                 }
  341 
  342                 /*
  343                  * Does our cached result allow us to give a definite yes to
  344                  * this request?
  345                  */
  346                 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
  347                     (ap->a_cred->cr_uid == np->n_modeuid) &&
  348                     ((np->n_mode & mode) == mode)) {
  349                         nfsstats.accesscache_hits++;
  350                 } else {
  351                         /*
  352                          * Either a no, or a don't know.  Go to the wire.
  353                          */
  354                         nfsstats.accesscache_misses++;
  355                         error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
  356                         if (!error) {
  357                                 if ((np->n_mode & mode) != mode) {
  358                                         error = EACCES;
  359                                 }
  360                         }
  361                 }
  362                 return (error);
  363         } else {
  364                 if ((error = nfsspec_access(ap)) != 0)
  365                         return (error);
  366 
  367                 /*
  368                  * Attempt to prevent a mapped root from accessing a file
  369                  * which it shouldn't.  We try to read a byte from the file
  370                  * if the user is root and the file is not zero length.
  371                  * After calling nfsspec_access, we should have the correct
  372                  * file size cached.
  373                  */
  374                 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
  375                     && VTONFS(vp)->n_size > 0) {
  376                         struct iovec aiov;
  377                         struct uio auio;
  378                         char buf[1];
  379 
  380                         aiov.iov_base = buf;
  381                         aiov.iov_len = 1;
  382                         auio.uio_iov = &aiov;
  383                         auio.uio_iovcnt = 1;
  384                         auio.uio_offset = 0;
  385                         auio.uio_resid = 1;
  386                         auio.uio_segflg = UIO_SYSSPACE;
  387                         auio.uio_rw = UIO_READ;
  388                         auio.uio_td = ap->a_td;
  389 
  390                         if (vp->v_type == VREG)
  391                                 error = nfs_readrpc(vp, &auio, ap->a_cred);
  392                         else if (vp->v_type == VDIR) {
  393                                 char* bp;
  394                                 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
  395                                 aiov.iov_base = bp;
  396                                 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
  397                                 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
  398                                 free(bp, M_TEMP);
  399                         } else if (vp->v_type == VLNK)
  400                                 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
  401                         else
  402                                 error = EACCES;
  403                 }
  404                 return (error);
  405         }
  406 }
  407 
  408 /*
  409  * nfs open vnode op
  410  * Check to see if the type is ok
  411  * and that deletion is not in progress.
  412  * For paged in text files, you will need to flush the page cache
  413  * if consistency is lost.
  414  */
  415 /* ARGSUSED */
  416 static int
  417 nfs_open(struct vop_open_args *ap)
  418 {
  419         struct vnode *vp = ap->a_vp;
  420         struct nfsnode *np = VTONFS(vp);
  421         struct vattr vattr;
  422         int error;
  423         int fmode = ap->a_mode;
  424 
  425         if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
  426                 return (EOPNOTSUPP);
  427 
  428         /*
  429          * Get a valid lease. If cached data is stale, flush it.
  430          */
  431         if (np->n_flag & NMODIFIED) {
  432                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  433                 if (error == EINTR || error == EIO)
  434                         return (error);
  435                 np->n_attrstamp = 0;
  436                 if (vp->v_type == VDIR)
  437                         np->n_direofoffset = 0;
  438                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  439                 if (error)
  440                         return (error);
  441                 np->n_mtime = vattr.va_mtime;
  442         } else {
  443                 np->n_attrstamp = 0;
  444                 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
  445                 if (error)
  446                         return (error);
  447                 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
  448                         if (vp->v_type == VDIR)
  449                                 np->n_direofoffset = 0;
  450                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  451                         if (error == EINTR || error == EIO)
  452                                 return (error);
  453                         np->n_mtime = vattr.va_mtime;
  454                 }
  455         }
  456         /*
  457          * If the object has >= 1 O_DIRECT active opens, we disable caching.
  458          */
  459         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  460                 if (np->n_directio_opens == 0) {
  461                         error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  462                         if (error)
  463                                 return (error);
  464                         np->n_flag |= NNONCACHE;
  465                 }
  466                 np->n_directio_opens++;
  467         }
  468         np->ra_expect_lbn = 0;
  469         vnode_create_vobject(vp, vattr.va_size, ap->a_td);
  470         return (0);
  471 }
  472 
  473 /*
  474  * nfs close vnode op
  475  * What an NFS client should do upon close after writing is a debatable issue.
  476  * Most NFS clients push delayed writes to the server upon close, basically for
  477  * two reasons:
  478  * 1 - So that any write errors may be reported back to the client process
  479  *     doing the close system call. By far the two most likely errors are
  480  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
  481  * 2 - To put a worst case upper bound on cache inconsistency between
  482  *     multiple clients for the file.
  483  * There is also a consistency problem for Version 2 of the protocol w.r.t.
  484  * not being able to tell if other clients are writing a file concurrently,
  485  * since there is no way of knowing if the changed modify time in the reply
  486  * is only due to the write for this client.
  487  * (NFS Version 3 provides weak cache consistency data in the reply that
  488  *  should be sufficient to detect and handle this case.)
  489  *
  490  * The current code does the following:
  491  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
  492  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
  493  *                     or commit them (this satisfies 1 and 2 except for the
  494  *                     case where the server crashes after this close but
  495  *                     before the commit RPC, which is felt to be "good
  496  *                     enough". Changing the last argument to nfs_flush() to
  497  *                     a 1 would force a commit operation, if it is felt a
  498  *                     commit is necessary now.
  499  */
  500 /* ARGSUSED */
  501 static int
  502 nfs_close(struct vop_close_args *ap)
  503 {
  504         struct vnode *vp = ap->a_vp;
  505         struct nfsnode *np = VTONFS(vp);
  506         int error = 0;
  507         int fmode = ap->a_fflag;
  508 
  509         if (vp->v_type == VREG) {
  510             /*
  511              * Examine and clean dirty pages, regardless of NMODIFIED.
  512              * This closes a major hole in close-to-open consistency.
  513              * We want to push out all dirty pages (and buffers) on
  514              * close, regardless of whether they were dirtied by
  515              * mmap'ed writes or via write().
  516              */
  517             if (nfs_clean_pages_on_close && vp->v_object) {
  518                 VM_OBJECT_LOCK(vp->v_object);
  519                 vm_object_page_clean(vp->v_object, 0, 0, 0);
  520                 VM_OBJECT_UNLOCK(vp->v_object);
  521             }
  522             if (np->n_flag & NMODIFIED) {
  523                 if (NFS_ISV3(vp)) {
  524                     /*
  525                      * Under NFSv3 we have dirty buffers to dispose of.  We
  526                      * must flush them to the NFS server.  We have the option
  527                      * of waiting all the way through the commit rpc or just
  528                      * waiting for the initial write.  The default is to only
  529                      * wait through the initial write so the data is in the
  530                      * server's cache, which is roughly similar to the state
  531                      * a standard disk subsystem leaves the file in on close().
  532                      *
  533                      * We cannot clear the NMODIFIED bit in np->n_flag due to
  534                      * potential races with other processes, and certainly
  535                      * cannot clear it if we don't commit.
  536                      */
  537                     int cm = nfsv3_commit_on_close ? 1 : 0;
  538                     error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
  539                     /* np->n_flag &= ~NMODIFIED; */
  540                 } else
  541                     error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  542             }
  543             /* 
  544              * Invalidate the attribute cache in all cases.
  545              * An open is going to fetch fresh attrs any way, other procs
  546              * on this node that have file open will be forced to do an 
  547              * otw attr fetch, but this is safe.
  548              */
  549             np->n_attrstamp = 0;
  550             if (np->n_flag & NWRITEERR) {
  551                 np->n_flag &= ~NWRITEERR;
  552                 error = np->n_error;
  553             }
  554         }
  555         if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
  556                 KASSERT((np->n_directio_opens > 0), 
  557                         ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));           
  558                 np->n_directio_opens--;
  559                 if (np->n_directio_opens == 0)
  560                         np->n_flag &= ~NNONCACHE;
  561         }
  562         return (error);
  563 }
  564 
  565 /*
  566  * nfs getattr call from vfs.
  567  */
  568 static int
  569 nfs_getattr(struct vop_getattr_args *ap)
  570 {
  571         struct vnode *vp = ap->a_vp;
  572         struct nfsnode *np = VTONFS(vp);
  573         caddr_t bpos, dpos;
  574         int error = 0;
  575         struct mbuf *mreq, *mrep, *md, *mb;
  576         int v3 = NFS_ISV3(vp);
  577 
  578         /*
  579          * Update local times for special files.
  580          */
  581         if (np->n_flag & (NACC | NUPD))
  582                 np->n_flag |= NCHG;
  583         /*
  584          * First look in the cache.
  585          */
  586         if (nfs_getattrcache(vp, ap->a_vap) == 0)
  587                 return (0);
  588 
  589         if (v3 && nfsaccess_cache_timeout > 0) {
  590                 nfsstats.accesscache_misses++;
  591                 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
  592                 if (nfs_getattrcache(vp, ap->a_vap) == 0)
  593                         return (0);
  594         }
  595 
  596         nfsstats.rpccnt[NFSPROC_GETATTR]++;
  597         mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
  598         mb = mreq;
  599         bpos = mtod(mb, caddr_t);
  600         nfsm_fhtom(vp, v3);
  601         nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
  602         if (!error) {
  603                 nfsm_loadattr(vp, ap->a_vap);
  604         }
  605         m_freem(mrep);
  606 nfsmout:
  607         return (error);
  608 }
  609 
  610 /*
  611  * nfs setattr call.
  612  */
  613 static int
  614 nfs_setattr(struct vop_setattr_args *ap)
  615 {
  616         struct vnode *vp = ap->a_vp;
  617         struct nfsnode *np = VTONFS(vp);
  618         struct vattr *vap = ap->a_vap;
  619         int error = 0;
  620         u_quad_t tsize;
  621 
  622 #ifndef nolint
  623         tsize = (u_quad_t)0;
  624 #endif
  625 
  626         /*
  627          * Setting of flags is not supported.
  628          */
  629         if (vap->va_flags != VNOVAL)
  630                 return (EOPNOTSUPP);
  631 
  632         /*
  633          * Disallow write attempts if the filesystem is mounted read-only.
  634          */
  635         if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
  636             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
  637             vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
  638             (vp->v_mount->mnt_flag & MNT_RDONLY))
  639                 return (EROFS);
  640         if (vap->va_size != VNOVAL) {
  641                 switch (vp->v_type) {
  642                 case VDIR:
  643                         return (EISDIR);
  644                 case VCHR:
  645                 case VBLK:
  646                 case VSOCK:
  647                 case VFIFO:
  648                         if (vap->va_mtime.tv_sec == VNOVAL &&
  649                             vap->va_atime.tv_sec == VNOVAL &&
  650                             vap->va_mode == (mode_t)VNOVAL &&
  651                             vap->va_uid == (uid_t)VNOVAL &&
  652                             vap->va_gid == (gid_t)VNOVAL)
  653                                 return (0);
  654                         vap->va_size = VNOVAL;
  655                         break;
  656                 default:
  657                         /*
  658                          * Disallow write attempts if the filesystem is
  659                          * mounted read-only.
  660                          */
  661                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  662                                 return (EROFS);
  663 
  664                         /*
  665                          *  We run vnode_pager_setsize() early (why?),
  666                          * we must set np->n_size now to avoid vinvalbuf
  667                          * V_SAVE races that might setsize a lower
  668                          * value.
  669                          */
  670 
  671                         tsize = np->n_size;
  672                         error = nfs_meta_setsize(vp, ap->a_cred, 
  673                                                 ap->a_td, vap->va_size);
  674 
  675                         if (np->n_flag & NMODIFIED) {
  676                             if (vap->va_size == 0)
  677                                 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
  678                             else
  679                                 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
  680                             if (error) {
  681                                 vnode_pager_setsize(vp, np->n_size);
  682                                 return (error);
  683                             }
  684                         }
  685                         /*
  686                          * np->n_size has already been set to vap->va_size
  687                          * in nfs_meta_setsize(). We must set it again since
  688                          * nfs_loadattrcache() could be called through
  689                          * nfs_meta_setsize() and could modify np->n_size.
  690                          */
  691                         np->n_vattr.va_size = np->n_size = vap->va_size;
  692                 };
  693         } else if ((vap->va_mtime.tv_sec != VNOVAL ||
  694                 vap->va_atime.tv_sec != VNOVAL) && (np->n_flag & NMODIFIED) &&
  695                 vp->v_type == VREG &&
  696                 (error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
  697                     (error == EINTR || error == EIO))
  698                 return (error);
  699         error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
  700         if (error && vap->va_size != VNOVAL) {
  701                 np->n_size = np->n_vattr.va_size = tsize;
  702                 vnode_pager_setsize(vp, np->n_size);
  703         }
  704         return (error);
  705 }
  706 
  707 /*
  708  * Do an nfs setattr rpc.
  709  */
  710 static int
  711 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
  712     struct thread *td)
  713 {
  714         struct nfsv2_sattr *sp;
  715         struct nfsnode *np = VTONFS(vp);
  716         caddr_t bpos, dpos;
  717         u_int32_t *tl;
  718         int error = 0, wccflag = NFSV3_WCCRATTR;
  719         struct mbuf *mreq, *mrep, *md, *mb;
  720         int v3 = NFS_ISV3(vp);
  721 
  722         nfsstats.rpccnt[NFSPROC_SETATTR]++;
  723         mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
  724         mb = mreq;
  725         bpos = mtod(mb, caddr_t);
  726         nfsm_fhtom(vp, v3);
  727         if (v3) {
  728                 nfsm_v3attrbuild(vap, TRUE);
  729                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
  730                 *tl = nfs_false;
  731         } else {
  732                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
  733                 if (vap->va_mode == (mode_t)VNOVAL)
  734                         sp->sa_mode = nfs_xdrneg1;
  735                 else
  736                         sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
  737                 if (vap->va_uid == (uid_t)VNOVAL)
  738                         sp->sa_uid = nfs_xdrneg1;
  739                 else
  740                         sp->sa_uid = txdr_unsigned(vap->va_uid);
  741                 if (vap->va_gid == (gid_t)VNOVAL)
  742                         sp->sa_gid = nfs_xdrneg1;
  743                 else
  744                         sp->sa_gid = txdr_unsigned(vap->va_gid);
  745                 sp->sa_size = txdr_unsigned(vap->va_size);
  746                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
  747                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
  748         }
  749         nfsm_request(vp, NFSPROC_SETATTR, td, cred);
  750         if (v3) {
  751                 np->n_modestamp = 0;
  752                 nfsm_wcc_data(vp, wccflag);
  753         } else
  754                 nfsm_loadattr(vp, NULL);
  755         m_freem(mrep);
  756 nfsmout:
  757         return (error);
  758 }
  759 
  760 /*
  761  * nfs lookup call, one step at a time...
  762  * First look in cache
  763  * If not found, unlock the directory nfsnode and do the rpc
  764  */
  765 static int
  766 nfs_lookup(struct vop_lookup_args *ap)
  767 {
  768         struct componentname *cnp = ap->a_cnp;
  769         struct vnode *dvp = ap->a_dvp;
  770         struct vnode **vpp = ap->a_vpp;
  771         int flags = cnp->cn_flags;
  772         struct vnode *newvp;
  773         struct nfsmount *nmp;
  774         caddr_t bpos, dpos;
  775         struct mbuf *mreq, *mrep, *md, *mb;
  776         long len;
  777         nfsfh_t *fhp;
  778         struct nfsnode *np;
  779         int error = 0, attrflag, fhsize;
  780         int v3 = NFS_ISV3(dvp);
  781         struct thread *td = cnp->cn_thread;
  782 
  783         *vpp = NULLVP;
  784         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
  785             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
  786                 return (EROFS);
  787         if (dvp->v_type != VDIR)
  788                 return (ENOTDIR);
  789         nmp = VFSTONFS(dvp->v_mount);
  790         np = VTONFS(dvp);
  791         if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
  792                 *vpp = NULLVP;
  793                 return (error);
  794         }
  795         if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
  796                 struct vattr vattr;
  797 
  798                 newvp = *vpp;
  799                 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
  800                  && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
  801                      nfsstats.lookupcache_hits++;
  802                      if (cnp->cn_nameiop != LOOKUP &&
  803                          (flags & ISLASTCN))
  804                              cnp->cn_flags |= SAVENAME;
  805                      return (0);
  806                 }
  807                 cache_purge(newvp);
  808                 if (dvp != newvp)
  809                         vput(newvp);
  810                 else 
  811                         vrele(newvp);
  812                 *vpp = NULLVP;
  813         }
  814         error = 0;
  815         newvp = NULLVP;
  816         nfsstats.lookupcache_misses++;
  817         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
  818         len = cnp->cn_namelen;
  819         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
  820                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
  821         mb = mreq;
  822         bpos = mtod(mb, caddr_t);
  823         nfsm_fhtom(dvp, v3);
  824         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
  825         nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
  826         if (error) {
  827                 nfsm_postop_attr(dvp, attrflag);
  828                 m_freem(mrep);
  829                 goto nfsmout;
  830         }
  831         nfsm_getfh(fhp, fhsize, v3);
  832 
  833         /*
  834          * Handle RENAME case...
  835          */
  836         if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
  837                 if (NFS_CMPFH(np, fhp, fhsize)) {
  838                         m_freem(mrep);
  839                         return (EISDIR);
  840                 }
  841                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  842                 if (error) {
  843                         m_freem(mrep);
  844                         return (error);
  845                 }
  846                 newvp = NFSTOV(np);
  847                 if (v3) {
  848                         nfsm_postop_attr(newvp, attrflag);
  849                         nfsm_postop_attr(dvp, attrflag);
  850                 } else
  851                         nfsm_loadattr(newvp, NULL);
  852                 *vpp = newvp;
  853                 m_freem(mrep);
  854                 cnp->cn_flags |= SAVENAME;
  855                 return (0);
  856         }
  857 
  858         if (flags & ISDOTDOT) {
  859                 VOP_UNLOCK(dvp, 0, td);
  860                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  861                 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
  862                 if (error)
  863                         return (error);
  864                 newvp = NFSTOV(np);
  865         } else if (NFS_CMPFH(np, fhp, fhsize)) {
  866                 VREF(dvp);
  867                 newvp = dvp;
  868         } else {
  869                 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np);
  870                 if (error) {
  871                         m_freem(mrep);
  872                         return (error);
  873                 }
  874                 newvp = NFSTOV(np);
  875         }
  876         if (v3) {
  877                 nfsm_postop_attr(newvp, attrflag);
  878                 nfsm_postop_attr(dvp, attrflag);
  879         } else
  880                 nfsm_loadattr(newvp, NULL);
  881         if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  882                 cnp->cn_flags |= SAVENAME;
  883         if ((cnp->cn_flags & MAKEENTRY) &&
  884             (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
  885                 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
  886                 cache_enter(dvp, newvp, cnp);
  887         }
  888         *vpp = newvp;
  889         m_freem(mrep);
  890 nfsmout:
  891         if (error) {
  892                 if (newvp != NULLVP) {
  893                         vput(newvp);
  894                         *vpp = NULLVP;
  895                 }
  896                 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
  897                     (flags & ISLASTCN) && error == ENOENT) {
  898                         if (dvp->v_mount->mnt_flag & MNT_RDONLY)
  899                                 error = EROFS;
  900                         else
  901                                 error = EJUSTRETURN;
  902                 }
  903                 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
  904                         cnp->cn_flags |= SAVENAME;
  905         }
  906         return (error);
  907 }
  908 
  909 /*
  910  * nfs read call.
  911  * Just call nfs_bioread() to do the work.
  912  */
  913 static int
  914 nfs_read(struct vop_read_args *ap)
  915 {
  916         struct vnode *vp = ap->a_vp;
  917 
  918         switch (vp->v_type) {
  919         case VREG:
  920                 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
  921         case VDIR:
  922                 return (EISDIR);
  923         default:
  924                 return (EOPNOTSUPP);
  925         }
  926 }
  927 
  928 /*
  929  * nfs readlink call
  930  */
  931 static int
  932 nfs_readlink(struct vop_readlink_args *ap)
  933 {
  934         struct vnode *vp = ap->a_vp;
  935 
  936         if (vp->v_type != VLNK)
  937                 return (EINVAL);
  938         return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
  939 }
  940 
  941 /*
  942  * Do a readlink rpc.
  943  * Called by nfs_doio() from below the buffer cache.
  944  */
  945 int
  946 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  947 {
  948         caddr_t bpos, dpos;
  949         int error = 0, len, attrflag;
  950         struct mbuf *mreq, *mrep, *md, *mb;
  951         int v3 = NFS_ISV3(vp);
  952 
  953         nfsstats.rpccnt[NFSPROC_READLINK]++;
  954         mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
  955         mb = mreq;
  956         bpos = mtod(mb, caddr_t);
  957         nfsm_fhtom(vp, v3);
  958         nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
  959         if (v3)
  960                 nfsm_postop_attr(vp, attrflag);
  961         if (!error) {
  962                 nfsm_strsiz(len, NFS_MAXPATHLEN);
  963                 if (len == NFS_MAXPATHLEN) {
  964                         struct nfsnode *np = VTONFS(vp);
  965                         if (np->n_size && np->n_size < NFS_MAXPATHLEN)
  966                                 len = np->n_size;
  967                 }
  968                 nfsm_mtouio(uiop, len);
  969         }
  970         m_freem(mrep);
  971 nfsmout:
  972         return (error);
  973 }
  974 
  975 /*
  976  * nfs read rpc call
  977  * Ditto above
  978  */
  979 int
  980 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
  981 {
  982         u_int32_t *tl;
  983         caddr_t bpos, dpos;
  984         struct mbuf *mreq, *mrep, *md, *mb;
  985         struct nfsmount *nmp;
  986         int error = 0, len, retlen, tsiz, eof, attrflag;
  987         int v3 = NFS_ISV3(vp);
  988 
  989 #ifndef nolint
  990         eof = 0;
  991 #endif
  992         nmp = VFSTONFS(vp->v_mount);
  993         tsiz = uiop->uio_resid;
  994         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
  995                 return (EFBIG);
  996         while (tsiz > 0) {
  997                 nfsstats.rpccnt[NFSPROC_READ]++;
  998                 len = (tsiz > nmp->nm_rsize) ? nmp->nm_rsize : tsiz;
  999                 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
 1000                 mb = mreq;
 1001                 bpos = mtod(mb, caddr_t);
 1002                 nfsm_fhtom(vp, v3);
 1003                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
 1004                 if (v3) {
 1005                         txdr_hyper(uiop->uio_offset, tl);
 1006                         *(tl + 2) = txdr_unsigned(len);
 1007                 } else {
 1008                         *tl++ = txdr_unsigned(uiop->uio_offset);
 1009                         *tl++ = txdr_unsigned(len);
 1010                         *tl = 0;
 1011                 }
 1012                 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
 1013                 if (v3) {
 1014                         nfsm_postop_attr(vp, attrflag);
 1015                         if (error) {
 1016                                 m_freem(mrep);
 1017                                 goto nfsmout;
 1018                         }
 1019                         tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
 1020                         eof = fxdr_unsigned(int, *(tl + 1));
 1021                 } else
 1022                         nfsm_loadattr(vp, NULL);
 1023                 nfsm_strsiz(retlen, nmp->nm_rsize);
 1024                 nfsm_mtouio(uiop, retlen);
 1025                 m_freem(mrep);
 1026                 tsiz -= retlen;
 1027                 if (v3) {
 1028                         if (eof || retlen == 0) {
 1029                                 tsiz = 0;
 1030                         }
 1031                 } else if (retlen < len) {
 1032                         tsiz = 0;
 1033                 }
 1034         }
 1035 nfsmout:
 1036         return (error);
 1037 }
 1038 
 1039 /*
 1040  * nfs write call
 1041  */
 1042 int
 1043 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
 1044     int *iomode, int *must_commit)
 1045 {
 1046         u_int32_t *tl;
 1047         int32_t backup;
 1048         caddr_t bpos, dpos;
 1049         struct mbuf *mreq, *mrep, *md, *mb;
 1050         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1051         int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
 1052         int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
 1053 
 1054 #ifndef DIAGNOSTIC
 1055         if (uiop->uio_iovcnt != 1)
 1056                 panic("nfs: writerpc iovcnt > 1");
 1057 #endif
 1058         *must_commit = 0;
 1059         tsiz = uiop->uio_resid;
 1060         if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize)
 1061                 return (EFBIG);
 1062         while (tsiz > 0) {
 1063                 nfsstats.rpccnt[NFSPROC_WRITE]++;
 1064                 len = (tsiz > nmp->nm_wsize) ? nmp->nm_wsize : tsiz;
 1065                 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
 1066                         NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
 1067                 mb = mreq;
 1068                 bpos = mtod(mb, caddr_t);
 1069                 nfsm_fhtom(vp, v3);
 1070                 if (v3) {
 1071                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1072                         txdr_hyper(uiop->uio_offset, tl);
 1073                         tl += 2;
 1074                         *tl++ = txdr_unsigned(len);
 1075                         *tl++ = txdr_unsigned(*iomode);
 1076                         *tl = txdr_unsigned(len);
 1077                 } else {
 1078                         u_int32_t x;
 1079 
 1080                         tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
 1081                         /* Set both "begin" and "current" to non-garbage. */
 1082                         x = txdr_unsigned((u_int32_t)uiop->uio_offset);
 1083                         *tl++ = x;      /* "begin offset" */
 1084                         *tl++ = x;      /* "current offset" */
 1085                         x = txdr_unsigned(len);
 1086                         *tl++ = x;      /* total to this offset */
 1087                         *tl = x;        /* size of this write */
 1088                 }
 1089                 nfsm_uiotom(uiop, len);
 1090                 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
 1091                 if (v3) {
 1092                         wccflag = NFSV3_WCCCHK;
 1093                         nfsm_wcc_data(vp, wccflag);
 1094                         if (!error) {
 1095                                 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
 1096                                         + NFSX_V3WRITEVERF);
 1097                                 rlen = fxdr_unsigned(int, *tl++);
 1098                                 if (rlen == 0) {
 1099                                         error = NFSERR_IO;
 1100                                         m_freem(mrep);
 1101                                         break;
 1102                                 } else if (rlen < len) {
 1103                                         backup = len - rlen;
 1104                                         uiop->uio_iov->iov_base =
 1105                                             (char *)uiop->uio_iov->iov_base -
 1106                                             backup;
 1107                                         uiop->uio_iov->iov_len += backup;
 1108                                         uiop->uio_offset -= backup;
 1109                                         uiop->uio_resid += backup;
 1110                                         len = rlen;
 1111                                 }
 1112                                 commit = fxdr_unsigned(int, *tl++);
 1113 
 1114                                 /*
 1115                                  * Return the lowest committment level
 1116                                  * obtained by any of the RPCs.
 1117                                  */
 1118                                 if (committed == NFSV3WRITE_FILESYNC)
 1119                                         committed = commit;
 1120                                 else if (committed == NFSV3WRITE_DATASYNC &&
 1121                                         commit == NFSV3WRITE_UNSTABLE)
 1122                                         committed = commit;
 1123                                 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
 1124                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1125                                         NFSX_V3WRITEVERF);
 1126                                     nmp->nm_state |= NFSSTA_HASWRITEVERF;
 1127                                 } else if (bcmp((caddr_t)tl,
 1128                                     (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
 1129                                     *must_commit = 1;
 1130                                     bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 1131                                         NFSX_V3WRITEVERF);
 1132                                 }
 1133                         }
 1134                 } else
 1135                     nfsm_loadattr(vp, NULL);
 1136                 if (wccflag)
 1137                     VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
 1138                 m_freem(mrep);
 1139                 if (error)
 1140                         break;
 1141                 tsiz -= len;
 1142         }
 1143 nfsmout:
 1144         if (vp->v_mount->mnt_flag & MNT_ASYNC)
 1145                 committed = NFSV3WRITE_FILESYNC;
 1146         *iomode = committed;
 1147         if (error)
 1148                 uiop->uio_resid = tsiz;
 1149         return (error);
 1150 }
 1151 
 1152 /*
 1153  * nfs mknod rpc
 1154  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
 1155  * mode set to specify the file type and the size field for rdev.
 1156  */
 1157 static int
 1158 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
 1159     struct vattr *vap)
 1160 {
 1161         struct nfsv2_sattr *sp;
 1162         u_int32_t *tl;
 1163         struct vnode *newvp = NULL;
 1164         struct nfsnode *np = NULL;
 1165         struct vattr vattr;
 1166         caddr_t bpos, dpos;
 1167         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
 1168         struct mbuf *mreq, *mrep, *md, *mb;
 1169         u_int32_t rdev;
 1170         int v3 = NFS_ISV3(dvp);
 1171 
 1172         if (vap->va_type == VCHR || vap->va_type == VBLK)
 1173                 rdev = txdr_unsigned(vap->va_rdev);
 1174         else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
 1175                 rdev = nfs_xdrneg1;
 1176         else {
 1177                 return (EOPNOTSUPP);
 1178         }
 1179         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1180                 return (error);
 1181         }
 1182         nfsstats.rpccnt[NFSPROC_MKNOD]++;
 1183         mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
 1184                 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1185         mb = mreq;
 1186         bpos = mtod(mb, caddr_t);
 1187         nfsm_fhtom(dvp, v3);
 1188         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1189         if (v3) {
 1190                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1191                 *tl++ = vtonfsv3_type(vap->va_type);
 1192                 nfsm_v3attrbuild(vap, FALSE);
 1193                 if (vap->va_type == VCHR || vap->va_type == VBLK) {
 1194                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1195                         *tl++ = txdr_unsigned(umajor(vap->va_rdev));
 1196                         *tl = txdr_unsigned(uminor(vap->va_rdev));
 1197                 }
 1198         } else {
 1199                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1200                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1201                 sp->sa_uid = nfs_xdrneg1;
 1202                 sp->sa_gid = nfs_xdrneg1;
 1203                 sp->sa_size = rdev;
 1204                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1205                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1206         }
 1207         nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
 1208         if (!error) {
 1209                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1210                 if (!gotvp) {
 1211                         if (newvp) {
 1212                                 vput(newvp);
 1213                                 newvp = NULL;
 1214                         }
 1215                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1216                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1217                         if (!error)
 1218                                 newvp = NFSTOV(np);
 1219                 }
 1220         }
 1221         if (v3)
 1222                 nfsm_wcc_data(dvp, wccflag);
 1223         m_freem(mrep);
 1224 nfsmout:
 1225         if (error) {
 1226                 if (newvp)
 1227                         vput(newvp);
 1228         } else {
 1229                 if (cnp->cn_flags & MAKEENTRY)
 1230                         cache_enter(dvp, newvp, cnp);
 1231                 *vpp = newvp;
 1232         }
 1233         VTONFS(dvp)->n_flag |= NMODIFIED;
 1234         if (!wccflag)
 1235                 VTONFS(dvp)->n_attrstamp = 0;
 1236         return (error);
 1237 }
 1238 
 1239 /*
 1240  * nfs mknod vop
 1241  * just call nfs_mknodrpc() to do the work.
 1242  */
 1243 /* ARGSUSED */
 1244 static int
 1245 nfs_mknod(struct vop_mknod_args *ap)
 1246 {
 1247 
 1248         return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
 1249 }
 1250 
 1251 static u_long create_verf;
 1252 /*
 1253  * nfs file create call
 1254  */
 1255 static int
 1256 nfs_create(struct vop_create_args *ap)
 1257 {
 1258         struct vnode *dvp = ap->a_dvp;
 1259         struct vattr *vap = ap->a_vap;
 1260         struct componentname *cnp = ap->a_cnp;
 1261         struct nfsv2_sattr *sp;
 1262         u_int32_t *tl;
 1263         struct nfsnode *np = NULL;
 1264         struct vnode *newvp = NULL;
 1265         caddr_t bpos, dpos;
 1266         int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
 1267         struct mbuf *mreq, *mrep, *md, *mb;
 1268         struct vattr vattr;
 1269         int v3 = NFS_ISV3(dvp);
 1270 
 1271         /*
 1272          * Oops, not for me..
 1273          */
 1274         if (vap->va_type == VSOCK)
 1275                 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
 1276 
 1277         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1278                 return (error);
 1279         }
 1280         if (vap->va_vaflags & VA_EXCLUSIVE)
 1281                 fmode |= O_EXCL;
 1282 again:
 1283         nfsstats.rpccnt[NFSPROC_CREATE]++;
 1284         mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
 1285                 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
 1286         mb = mreq;
 1287         bpos = mtod(mb, caddr_t);
 1288         nfsm_fhtom(dvp, v3);
 1289         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1290         if (v3) {
 1291                 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
 1292                 if (fmode & O_EXCL) {
 1293                         *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
 1294                         tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
 1295 #ifdef INET
 1296                         if (!TAILQ_EMPTY(&in_ifaddrhead))
 1297                                 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
 1298                         else
 1299 #endif
 1300                                 *tl++ = create_verf;
 1301                         *tl = ++create_verf;
 1302                 } else {
 1303                         *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
 1304                         nfsm_v3attrbuild(vap, FALSE);
 1305                 }
 1306         } else {
 1307                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1308                 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
 1309                 sp->sa_uid = nfs_xdrneg1;
 1310                 sp->sa_gid = nfs_xdrneg1;
 1311                 sp->sa_size = 0;
 1312                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1313                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1314         }
 1315         nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
 1316         if (!error) {
 1317                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1318                 if (!gotvp) {
 1319                         if (newvp) {
 1320                                 vput(newvp);
 1321                                 newvp = NULL;
 1322                         }
 1323                         error = nfs_lookitup(dvp, cnp->cn_nameptr,
 1324                             cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
 1325                         if (!error)
 1326                                 newvp = NFSTOV(np);
 1327                 }
 1328         }
 1329         if (v3)
 1330                 nfsm_wcc_data(dvp, wccflag);
 1331         m_freem(mrep);
 1332 nfsmout:
 1333         if (error) {
 1334                 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
 1335                         fmode &= ~O_EXCL;
 1336                         goto again;
 1337                 }
 1338                 if (newvp)
 1339                         vput(newvp);
 1340         } else if (v3 && (fmode & O_EXCL)) {
 1341                 /*
 1342                  * We are normally called with only a partially initialized
 1343                  * VAP.  Since the NFSv3 spec says that server may use the
 1344                  * file attributes to store the verifier, the spec requires
 1345                  * us to do a SETATTR RPC. FreeBSD servers store the verifier
 1346                  * in atime, but we can't really assume that all servers will
 1347                  * so we ensure that our SETATTR sets both atime and mtime.
 1348                  */
 1349                 if (vap->va_mtime.tv_sec == VNOVAL)
 1350                         vfs_timestamp(&vap->va_mtime);
 1351                 if (vap->va_atime.tv_sec == VNOVAL)
 1352                         vap->va_atime = vap->va_mtime;
 1353                 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
 1354         }
 1355         if (!error) {
 1356                 if (cnp->cn_flags & MAKEENTRY)
 1357                         cache_enter(dvp, newvp, cnp);
 1358                 *ap->a_vpp = newvp;
 1359         }
 1360         VTONFS(dvp)->n_flag |= NMODIFIED;
 1361         if (!wccflag)
 1362                 VTONFS(dvp)->n_attrstamp = 0;
 1363         return (error);
 1364 }
 1365 
 1366 /*
 1367  * nfs file remove call
 1368  * To try and make nfs semantics closer to ufs semantics, a file that has
 1369  * other processes using the vnode is renamed instead of removed and then
 1370  * removed later on the last close.
 1371  * - If v_usecount > 1
 1372  *        If a rename is not already in the works
 1373  *           call nfs_sillyrename() to set it up
 1374  *     else
 1375  *        do the remove rpc
 1376  */
 1377 static int
 1378 nfs_remove(struct vop_remove_args *ap)
 1379 {
 1380         struct vnode *vp = ap->a_vp;
 1381         struct vnode *dvp = ap->a_dvp;
 1382         struct componentname *cnp = ap->a_cnp;
 1383         struct nfsnode *np = VTONFS(vp);
 1384         int error = 0;
 1385         struct vattr vattr;
 1386 
 1387 #ifndef DIAGNOSTIC
 1388         if ((cnp->cn_flags & HASBUF) == 0)
 1389                 panic("nfs_remove: no name");
 1390         if (vrefcnt(vp) < 1)
 1391                 panic("nfs_remove: bad v_usecount");
 1392 #endif
 1393         if (vp->v_type == VDIR)
 1394                 error = EPERM;
 1395         else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
 1396             VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
 1397             vattr.va_nlink > 1)) {
 1398                 /*
 1399                  * Purge the name cache so that the chance of a lookup for
 1400                  * the name succeeding while the remove is in progress is
 1401                  * minimized. Without node locking it can still happen, such
 1402                  * that an I/O op returns ESTALE, but since you get this if
 1403                  * another host removes the file..
 1404                  */
 1405                 cache_purge(vp);
 1406                 /*
 1407                  * throw away biocache buffers, mainly to avoid
 1408                  * unnecessary delayed writes later.
 1409                  */
 1410                 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
 1411                 /* Do the rpc */
 1412                 if (error != EINTR && error != EIO)
 1413                         error = nfs_removerpc(dvp, cnp->cn_nameptr,
 1414                                 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
 1415                 /*
 1416                  * Kludge City: If the first reply to the remove rpc is lost..
 1417                  *   the reply to the retransmitted request will be ENOENT
 1418                  *   since the file was in fact removed
 1419                  *   Therefore, we cheat and return success.
 1420                  */
 1421                 if (error == ENOENT)
 1422                         error = 0;
 1423         } else if (!np->n_sillyrename)
 1424                 error = nfs_sillyrename(dvp, vp, cnp);
 1425         np->n_attrstamp = 0;
 1426         return (error);
 1427 }
 1428 
 1429 /*
 1430  * nfs file remove rpc called from nfs_inactive
 1431  */
 1432 int
 1433 nfs_removeit(struct sillyrename *sp)
 1434 {
 1435 
 1436         /*
 1437          * Make sure that the directory vnode is still valid.
 1438          * XXX we should lock sp->s_dvp here.
 1439          */
 1440         if (sp->s_dvp->v_type == VBAD)
 1441                 return (0);
 1442         return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 1443                 NULL));
 1444 }
 1445 
 1446 /*
 1447  * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
 1448  */
 1449 static int
 1450 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
 1451     struct ucred *cred, struct thread *td)
 1452 {
 1453         caddr_t bpos, dpos;
 1454         int error = 0, wccflag = NFSV3_WCCRATTR;
 1455         struct mbuf *mreq, *mrep, *md, *mb;
 1456         int v3 = NFS_ISV3(dvp);
 1457 
 1458         nfsstats.rpccnt[NFSPROC_REMOVE]++;
 1459         mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
 1460                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
 1461         mb = mreq;
 1462         bpos = mtod(mb, caddr_t);
 1463         nfsm_fhtom(dvp, v3);
 1464         nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
 1465         nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
 1466         if (v3)
 1467                 nfsm_wcc_data(dvp, wccflag);
 1468         m_freem(mrep);
 1469 nfsmout:
 1470         VTONFS(dvp)->n_flag |= NMODIFIED;
 1471         if (!wccflag)
 1472                 VTONFS(dvp)->n_attrstamp = 0;
 1473         return (error);
 1474 }
 1475 
 1476 /*
 1477  * nfs file rename call
 1478  */
 1479 static int
 1480 nfs_rename(struct vop_rename_args *ap)
 1481 {
 1482         struct vnode *fvp = ap->a_fvp;
 1483         struct vnode *tvp = ap->a_tvp;
 1484         struct vnode *fdvp = ap->a_fdvp;
 1485         struct vnode *tdvp = ap->a_tdvp;
 1486         struct componentname *tcnp = ap->a_tcnp;
 1487         struct componentname *fcnp = ap->a_fcnp;
 1488         int error;
 1489 
 1490 #ifndef DIAGNOSTIC
 1491         if ((tcnp->cn_flags & HASBUF) == 0 ||
 1492             (fcnp->cn_flags & HASBUF) == 0)
 1493                 panic("nfs_rename: no name");
 1494 #endif
 1495         /* Check for cross-device rename */
 1496         if ((fvp->v_mount != tdvp->v_mount) ||
 1497             (tvp && (fvp->v_mount != tvp->v_mount))) {
 1498                 error = EXDEV;
 1499                 goto out;
 1500         }
 1501 
 1502         if (fvp == tvp) {
 1503                 printf("nfs_rename: fvp == tvp (can't happen)\n");
 1504                 error = 0;
 1505                 goto out;
 1506         }
 1507         if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
 1508                 goto out;
 1509 
 1510         /*
 1511          * We have to flush B_DELWRI data prior to renaming
 1512          * the file.  If we don't, the delayed-write buffers
 1513          * can be flushed out later after the file has gone stale
 1514          * under NFSV3.  NFSV2 does not have this problem because
 1515          * ( as far as I can tell ) it flushes dirty buffers more
 1516          * often.
 1517          * 
 1518          * Skip the rename operation if the fsync fails, this can happen
 1519          * due to the server's volume being full, when we pushed out data
 1520          * that was written back to our cache earlier. Not checking for
 1521          * this condition can result in potential (silent) data loss.
 1522          */
 1523         error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
 1524         VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
 1525         if (!error && tvp)
 1526                 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
 1527         if (error)
 1528                 goto out;
 1529 
 1530         /*
 1531          * If the tvp exists and is in use, sillyrename it before doing the
 1532          * rename of the new file over it.
 1533          * XXX Can't sillyrename a directory.
 1534          */
 1535         if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
 1536                 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
 1537                 vput(tvp);
 1538                 tvp = NULL;
 1539         }
 1540 
 1541         error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
 1542                 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
 1543                 tcnp->cn_thread);
 1544 
 1545         if (fvp->v_type == VDIR) {
 1546                 if (tvp != NULL && tvp->v_type == VDIR)
 1547                         cache_purge(tdvp);
 1548                 cache_purge(fdvp);
 1549         }
 1550 
 1551 out:
 1552         if (tdvp == tvp)
 1553                 vrele(tdvp);
 1554         else
 1555                 vput(tdvp);
 1556         if (tvp)
 1557                 vput(tvp);
 1558         vrele(fdvp);
 1559         vrele(fvp);
 1560         /*
 1561          * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
 1562          */
 1563         if (error == ENOENT)
 1564                 error = 0;
 1565         return (error);
 1566 }
 1567 
 1568 /*
 1569  * nfs file rename rpc called from nfs_remove() above
 1570  */
 1571 static int
 1572 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
 1573     struct sillyrename *sp)
 1574 {
 1575 
 1576         return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
 1577             sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
 1578 }
 1579 
 1580 /*
 1581  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
 1582  */
 1583 static int
 1584 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
 1585     struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
 1586     struct thread *td)
 1587 {
 1588         caddr_t bpos, dpos;
 1589         int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
 1590         struct mbuf *mreq, *mrep, *md, *mb;
 1591         int v3 = NFS_ISV3(fdvp);
 1592 
 1593         nfsstats.rpccnt[NFSPROC_RENAME]++;
 1594         mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
 1595                 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
 1596                 nfsm_rndup(tnamelen));
 1597         mb = mreq;
 1598         bpos = mtod(mb, caddr_t);
 1599         nfsm_fhtom(fdvp, v3);
 1600         nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
 1601         nfsm_fhtom(tdvp, v3);
 1602         nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
 1603         nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
 1604         if (v3) {
 1605                 nfsm_wcc_data(fdvp, fwccflag);
 1606                 nfsm_wcc_data(tdvp, twccflag);
 1607         }
 1608         m_freem(mrep);
 1609 nfsmout:
 1610         VTONFS(fdvp)->n_flag |= NMODIFIED;
 1611         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1612         if (!fwccflag)
 1613                 VTONFS(fdvp)->n_attrstamp = 0;
 1614         if (!twccflag)
 1615                 VTONFS(tdvp)->n_attrstamp = 0;
 1616         return (error);
 1617 }
 1618 
 1619 /*
 1620  * nfs hard link create call
 1621  */
 1622 static int
 1623 nfs_link(struct vop_link_args *ap)
 1624 {
 1625         struct vnode *vp = ap->a_vp;
 1626         struct vnode *tdvp = ap->a_tdvp;
 1627         struct componentname *cnp = ap->a_cnp;
 1628         caddr_t bpos, dpos;
 1629         int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
 1630         struct mbuf *mreq, *mrep, *md, *mb;
 1631         int v3;
 1632 
 1633         if (vp->v_mount != tdvp->v_mount) {
 1634                 return (EXDEV);
 1635         }
 1636 
 1637         /*
 1638          * Push all writes to the server, so that the attribute cache
 1639          * doesn't get "out of sync" with the server.
 1640          * XXX There should be a better way!
 1641          */
 1642         VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
 1643 
 1644         v3 = NFS_ISV3(vp);
 1645         nfsstats.rpccnt[NFSPROC_LINK]++;
 1646         mreq = nfsm_reqhead(vp, NFSPROC_LINK,
 1647                 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1648         mb = mreq;
 1649         bpos = mtod(mb, caddr_t);
 1650         nfsm_fhtom(vp, v3);
 1651         nfsm_fhtom(tdvp, v3);
 1652         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1653         nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
 1654         if (v3) {
 1655                 nfsm_postop_attr(vp, attrflag);
 1656                 nfsm_wcc_data(tdvp, wccflag);
 1657         }
 1658         m_freem(mrep);
 1659 nfsmout:
 1660         VTONFS(tdvp)->n_flag |= NMODIFIED;
 1661         if (!attrflag)
 1662                 VTONFS(vp)->n_attrstamp = 0;
 1663         if (!wccflag)
 1664                 VTONFS(tdvp)->n_attrstamp = 0;
 1665         /*
 1666          * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
 1667          */
 1668         if (error == EEXIST)
 1669                 error = 0;
 1670         return (error);
 1671 }
 1672 
 1673 /*
 1674  * nfs symbolic link create call
 1675  */
 1676 static int
 1677 nfs_symlink(struct vop_symlink_args *ap)
 1678 {
 1679         struct vnode *dvp = ap->a_dvp;
 1680         struct vattr *vap = ap->a_vap;
 1681         struct componentname *cnp = ap->a_cnp;
 1682         struct nfsv2_sattr *sp;
 1683         caddr_t bpos, dpos;
 1684         int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
 1685         struct mbuf *mreq, *mrep, *md, *mb;
 1686         struct vnode *newvp = NULL;
 1687         int v3 = NFS_ISV3(dvp);
 1688 
 1689         nfsstats.rpccnt[NFSPROC_SYMLINK]++;
 1690         slen = strlen(ap->a_target);
 1691         mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
 1692             nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
 1693         mb = mreq;
 1694         bpos = mtod(mb, caddr_t);
 1695         nfsm_fhtom(dvp, v3);
 1696         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1697         if (v3) {
 1698                 nfsm_v3attrbuild(vap, FALSE);
 1699         }
 1700         nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
 1701         if (!v3) {
 1702                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1703                 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
 1704                 sp->sa_uid = nfs_xdrneg1;
 1705                 sp->sa_gid = nfs_xdrneg1;
 1706                 sp->sa_size = nfs_xdrneg1;
 1707                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1708                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1709         }
 1710 
 1711         /*
 1712          * Issue the NFS request and get the rpc response.
 1713          *
 1714          * Only NFSv3 responses returning an error of 0 actually return
 1715          * a file handle that can be converted into newvp without having
 1716          * to do an extra lookup rpc.
 1717          */
 1718         nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
 1719         if (v3) {
 1720                 if (error == 0)
 1721                         nfsm_mtofh(dvp, newvp, v3, gotvp);
 1722                 nfsm_wcc_data(dvp, wccflag);
 1723         }
 1724 
 1725         /*
 1726          * out code jumps -> here, mrep is also freed.
 1727          */
 1728 
 1729         m_freem(mrep);
 1730 nfsmout:
 1731 
 1732         /*
 1733          * If we get an EEXIST error, silently convert it to no-error
 1734          * in case of an NFS retry.
 1735          */
 1736         if (error == EEXIST)
 1737                 error = 0;
 1738 
 1739         /*
 1740          * If we do not have (or no longer have) an error, and we could
 1741          * not extract the newvp from the response due to the request being
 1742          * NFSv2 or the error being EEXIST.  We have to do a lookup in order
 1743          * to obtain a newvp to return.
 1744          */
 1745         if (error == 0 && newvp == NULL) {
 1746                 struct nfsnode *np = NULL;
 1747 
 1748                 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
 1749                     cnp->cn_cred, cnp->cn_thread, &np);
 1750                 if (!error)
 1751                         newvp = NFSTOV(np);
 1752         }
 1753         if (error) {
 1754                 if (newvp)
 1755                         vput(newvp);
 1756         } else {
 1757                 *ap->a_vpp = newvp;
 1758         }
 1759         VTONFS(dvp)->n_flag |= NMODIFIED;
 1760         if (!wccflag)
 1761                 VTONFS(dvp)->n_attrstamp = 0;
 1762         return (error);
 1763 }
 1764 
 1765 /*
 1766  * nfs make dir call
 1767  */
 1768 static int
 1769 nfs_mkdir(struct vop_mkdir_args *ap)
 1770 {
 1771         struct vnode *dvp = ap->a_dvp;
 1772         struct vattr *vap = ap->a_vap;
 1773         struct componentname *cnp = ap->a_cnp;
 1774         struct nfsv2_sattr *sp;
 1775         int len;
 1776         struct nfsnode *np = NULL;
 1777         struct vnode *newvp = NULL;
 1778         caddr_t bpos, dpos;
 1779         int error = 0, wccflag = NFSV3_WCCRATTR;
 1780         int gotvp = 0;
 1781         struct mbuf *mreq, *mrep, *md, *mb;
 1782         struct vattr vattr;
 1783         int v3 = NFS_ISV3(dvp);
 1784 
 1785         if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
 1786                 return (error);
 1787         }
 1788         len = cnp->cn_namelen;
 1789         nfsstats.rpccnt[NFSPROC_MKDIR]++;
 1790         mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
 1791           NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
 1792         mb = mreq;
 1793         bpos = mtod(mb, caddr_t);
 1794         nfsm_fhtom(dvp, v3);
 1795         nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
 1796         if (v3) {
 1797                 nfsm_v3attrbuild(vap, FALSE);
 1798         } else {
 1799                 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
 1800                 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
 1801                 sp->sa_uid = nfs_xdrneg1;
 1802                 sp->sa_gid = nfs_xdrneg1;
 1803                 sp->sa_size = nfs_xdrneg1;
 1804                 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
 1805                 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
 1806         }
 1807         nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
 1808         if (!error)
 1809                 nfsm_mtofh(dvp, newvp, v3, gotvp);
 1810         if (v3)
 1811                 nfsm_wcc_data(dvp, wccflag);
 1812         m_freem(mrep);
 1813 nfsmout:
 1814         VTONFS(dvp)->n_flag |= NMODIFIED;
 1815         if (!wccflag)
 1816                 VTONFS(dvp)->n_attrstamp = 0;
 1817         /*
 1818          * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
 1819          * if we can succeed in looking up the directory.
 1820          */
 1821         if (error == EEXIST || (!error && !gotvp)) {
 1822                 if (newvp) {
 1823                         vput(newvp);
 1824                         newvp = NULL;
 1825                 }
 1826                 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
 1827                         cnp->cn_thread, &np);
 1828                 if (!error) {
 1829                         newvp = NFSTOV(np);
 1830                         if (newvp->v_type != VDIR)
 1831                                 error = EEXIST;
 1832                 }
 1833         }
 1834         if (error) {
 1835                 if (newvp)
 1836                         vput(newvp);
 1837         } else
 1838                 *ap->a_vpp = newvp;
 1839         return (error);
 1840 }
 1841 
 1842 /*
 1843  * nfs remove directory call
 1844  */
 1845 static int
 1846 nfs_rmdir(struct vop_rmdir_args *ap)
 1847 {
 1848         struct vnode *vp = ap->a_vp;
 1849         struct vnode *dvp = ap->a_dvp;
 1850         struct componentname *cnp = ap->a_cnp;
 1851         caddr_t bpos, dpos;
 1852         int error = 0, wccflag = NFSV3_WCCRATTR;
 1853         struct mbuf *mreq, *mrep, *md, *mb;
 1854         int v3 = NFS_ISV3(dvp);
 1855 
 1856         if (dvp == vp)
 1857                 return (EINVAL);
 1858         nfsstats.rpccnt[NFSPROC_RMDIR]++;
 1859         mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
 1860                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
 1861         mb = mreq;
 1862         bpos = mtod(mb, caddr_t);
 1863         nfsm_fhtom(dvp, v3);
 1864         nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
 1865         nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
 1866         if (v3)
 1867                 nfsm_wcc_data(dvp, wccflag);
 1868         m_freem(mrep);
 1869 nfsmout:
 1870         VTONFS(dvp)->n_flag |= NMODIFIED;
 1871         if (!wccflag)
 1872                 VTONFS(dvp)->n_attrstamp = 0;
 1873         cache_purge(dvp);
 1874         cache_purge(vp);
 1875         /*
 1876          * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
 1877          */
 1878         if (error == ENOENT)
 1879                 error = 0;
 1880         return (error);
 1881 }
 1882 
 1883 /*
 1884  * nfs readdir call
 1885  */
 1886 static int
 1887 nfs_readdir(struct vop_readdir_args *ap)
 1888 {
 1889         struct vnode *vp = ap->a_vp;
 1890         struct nfsnode *np = VTONFS(vp);
 1891         struct uio *uio = ap->a_uio;
 1892         int tresid, error;
 1893         struct vattr vattr;
 1894 
 1895         if (vp->v_type != VDIR)
 1896                 return (EPERM);
 1897         /*
 1898          * First, check for hit on the EOF offset cache
 1899          */
 1900         if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
 1901             (np->n_flag & NMODIFIED) == 0) {
 1902                 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0 &&
 1903                     !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
 1904                         nfsstats.direofcache_hits++;
 1905                         return (0);
 1906                 }
 1907         }
 1908 
 1909         /*
 1910          * Call nfs_bioread() to do the real work.
 1911          */
 1912         tresid = uio->uio_resid;
 1913         error = nfs_bioread(vp, uio, 0, ap->a_cred);
 1914 
 1915         if (!error && uio->uio_resid == tresid)
 1916                 nfsstats.direofcache_misses++;
 1917         return (error);
 1918 }
 1919 
 1920 /*
 1921  * Readdir rpc call.
 1922  * Called from below the buffer cache by nfs_doio().
 1923  */
 1924 int
 1925 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 1926 {
 1927         int len, left;
 1928         struct dirent *dp = NULL;
 1929         u_int32_t *tl;
 1930         caddr_t cp;
 1931         nfsuint64 *cookiep;
 1932         caddr_t bpos, dpos;
 1933         struct mbuf *mreq, *mrep, *md, *mb;
 1934         nfsuint64 cookie;
 1935         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 1936         struct nfsnode *dnp = VTONFS(vp);
 1937         u_quad_t fileno;
 1938         int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
 1939         int attrflag;
 1940         int v3 = NFS_ISV3(vp);
 1941 
 1942 #ifndef DIAGNOSTIC
 1943         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 1944                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 1945                 panic("nfs readdirrpc bad uio");
 1946 #endif
 1947 
 1948         /*
 1949          * If there is no cookie, assume directory was stale.
 1950          */
 1951         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 1952         if (cookiep)
 1953                 cookie = *cookiep;
 1954         else
 1955                 return (NFSERR_BAD_COOKIE);
 1956         /*
 1957          * Loop around doing readdir rpc's of size nm_readdirsize
 1958          * truncated to a multiple of DIRBLKSIZ.
 1959          * The stopping criteria is EOF or buffer full.
 1960          */
 1961         while (more_dirs && bigenough) {
 1962                 nfsstats.rpccnt[NFSPROC_READDIR]++;
 1963                 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
 1964                         NFSX_READDIR(v3));
 1965                 mb = mreq;
 1966                 bpos = mtod(mb, caddr_t);
 1967                 nfsm_fhtom(vp, v3);
 1968                 if (v3) {
 1969                         tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
 1970                         *tl++ = cookie.nfsuquad[0];
 1971                         *tl++ = cookie.nfsuquad[1];
 1972                         *tl++ = dnp->n_cookieverf.nfsuquad[0];
 1973                         *tl++ = dnp->n_cookieverf.nfsuquad[1];
 1974                 } else {
 1975                         tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
 1976                         *tl++ = cookie.nfsuquad[0];
 1977                 }
 1978                 *tl = txdr_unsigned(nmp->nm_readdirsize);
 1979                 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
 1980                 if (v3) {
 1981                         nfsm_postop_attr(vp, attrflag);
 1982                         if (!error) {
 1983                                 tl = nfsm_dissect(u_int32_t *,
 1984                                     2 * NFSX_UNSIGNED);
 1985                                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 1986                                 dnp->n_cookieverf.nfsuquad[1] = *tl;
 1987                         } else {
 1988                                 m_freem(mrep);
 1989                                 goto nfsmout;
 1990                         }
 1991                 }
 1992                 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 1993                 more_dirs = fxdr_unsigned(int, *tl);
 1994 
 1995                 /* loop thru the dir entries, doctoring them to 4bsd form */
 1996                 while (more_dirs && bigenough) {
 1997                         if (v3) {
 1998                                 tl = nfsm_dissect(u_int32_t *,
 1999                                     3 * NFSX_UNSIGNED);
 2000                                 fileno = fxdr_hyper(tl);
 2001                                 len = fxdr_unsigned(int, *(tl + 2));
 2002                         } else {
 2003                                 tl = nfsm_dissect(u_int32_t *,
 2004                                     2 * NFSX_UNSIGNED);
 2005                                 fileno = fxdr_unsigned(u_quad_t, *tl++);
 2006                                 len = fxdr_unsigned(int, *tl);
 2007                         }
 2008                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2009                                 error = EBADRPC;
 2010                                 m_freem(mrep);
 2011                                 goto nfsmout;
 2012                         }
 2013                         tlen = nfsm_rndup(len);
 2014                         if (tlen == len)
 2015                                 tlen += 4;      /* To ensure null termination */
 2016                         left = DIRBLKSIZ - blksiz;
 2017                         if ((tlen + DIRHDSIZ) > left) {
 2018                                 dp->d_reclen += left;
 2019                                 uiop->uio_iov->iov_base =
 2020                                     (char *)uiop->uio_iov->iov_base + left;
 2021                                 uiop->uio_iov->iov_len -= left;
 2022                                 uiop->uio_offset += left;
 2023                                 uiop->uio_resid -= left;
 2024                                 blksiz = 0;
 2025                         }
 2026                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2027                                 bigenough = 0;
 2028                         if (bigenough) {
 2029                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2030                                 dp->d_fileno = (int)fileno;
 2031                                 dp->d_namlen = len;
 2032                                 dp->d_reclen = tlen + DIRHDSIZ;
 2033                                 dp->d_type = DT_UNKNOWN;
 2034                                 blksiz += dp->d_reclen;
 2035                                 if (blksiz == DIRBLKSIZ)
 2036                                         blksiz = 0;
 2037                                 uiop->uio_offset += DIRHDSIZ;
 2038                                 uiop->uio_resid -= DIRHDSIZ;
 2039                                 uiop->uio_iov->iov_base =
 2040                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2041                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2042                                 nfsm_mtouio(uiop, len);
 2043                                 cp = uiop->uio_iov->iov_base;
 2044                                 tlen -= len;
 2045                                 *cp = '\0';     /* null terminate */
 2046                                 uiop->uio_iov->iov_base =
 2047                                     (char *)uiop->uio_iov->iov_base + tlen;
 2048                                 uiop->uio_iov->iov_len -= tlen;
 2049                                 uiop->uio_offset += tlen;
 2050                                 uiop->uio_resid -= tlen;
 2051                         } else
 2052                                 nfsm_adv(nfsm_rndup(len));
 2053                         if (v3) {
 2054                                 tl = nfsm_dissect(u_int32_t *,
 2055                                     3 * NFSX_UNSIGNED);
 2056                         } else {
 2057                                 tl = nfsm_dissect(u_int32_t *,
 2058                                     2 * NFSX_UNSIGNED);
 2059                         }
 2060                         if (bigenough) {
 2061                                 cookie.nfsuquad[0] = *tl++;
 2062                                 if (v3)
 2063                                         cookie.nfsuquad[1] = *tl++;
 2064                         } else if (v3)
 2065                                 tl += 2;
 2066                         else
 2067                                 tl++;
 2068                         more_dirs = fxdr_unsigned(int, *tl);
 2069                 }
 2070                 /*
 2071                  * If at end of rpc data, get the eof boolean
 2072                  */
 2073                 if (!more_dirs) {
 2074                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2075                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2076                 }
 2077                 m_freem(mrep);
 2078         }
 2079         /*
 2080          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2081          * by increasing d_reclen for the last record.
 2082          */
 2083         if (blksiz > 0) {
 2084                 left = DIRBLKSIZ - blksiz;
 2085                 dp->d_reclen += left;
 2086                 uiop->uio_iov->iov_base =
 2087                     (char *)uiop->uio_iov->iov_base + left;
 2088                 uiop->uio_iov->iov_len -= left;
 2089                 uiop->uio_offset += left;
 2090                 uiop->uio_resid -= left;
 2091         }
 2092 
 2093         /*
 2094          * We are now either at the end of the directory or have filled the
 2095          * block.
 2096          */
 2097         if (bigenough)
 2098                 dnp->n_direofoffset = uiop->uio_offset;
 2099         else {
 2100                 if (uiop->uio_resid > 0)
 2101                         printf("EEK! readdirrpc resid > 0\n");
 2102                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2103                 *cookiep = cookie;
 2104         }
 2105 nfsmout:
 2106         return (error);
 2107 }
 2108 
 2109 /*
 2110  * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
 2111  */
 2112 int
 2113 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
 2114 {
 2115         int len, left;
 2116         struct dirent *dp;
 2117         u_int32_t *tl;
 2118         caddr_t cp;
 2119         struct vnode *newvp;
 2120         nfsuint64 *cookiep;
 2121         caddr_t bpos, dpos, dpossav1, dpossav2;
 2122         struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
 2123         struct nameidata nami, *ndp = &nami;
 2124         struct componentname *cnp = &ndp->ni_cnd;
 2125         nfsuint64 cookie;
 2126         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2127         struct nfsnode *dnp = VTONFS(vp), *np;
 2128         nfsfh_t *fhp;
 2129         u_quad_t fileno;
 2130         int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
 2131         int attrflag, fhsize;
 2132 
 2133 #ifndef nolint
 2134         dp = NULL;
 2135 #endif
 2136 #ifndef DIAGNOSTIC
 2137         if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
 2138                 (uiop->uio_resid & (DIRBLKSIZ - 1)))
 2139                 panic("nfs readdirplusrpc bad uio");
 2140 #endif
 2141         ndp->ni_dvp = vp;
 2142         newvp = NULLVP;
 2143 
 2144         /*
 2145          * If there is no cookie, assume directory was stale.
 2146          */
 2147         cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
 2148         if (cookiep)
 2149                 cookie = *cookiep;
 2150         else
 2151                 return (NFSERR_BAD_COOKIE);
 2152         /*
 2153          * Loop around doing readdir rpc's of size nm_readdirsize
 2154          * truncated to a multiple of DIRBLKSIZ.
 2155          * The stopping criteria is EOF or buffer full.
 2156          */
 2157         while (more_dirs && bigenough) {
 2158                 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
 2159                 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
 2160                         NFSX_FH(1) + 6 * NFSX_UNSIGNED);
 2161                 mb = mreq;
 2162                 bpos = mtod(mb, caddr_t);
 2163                 nfsm_fhtom(vp, 1);
 2164                 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
 2165                 *tl++ = cookie.nfsuquad[0];
 2166                 *tl++ = cookie.nfsuquad[1];
 2167                 *tl++ = dnp->n_cookieverf.nfsuquad[0];
 2168                 *tl++ = dnp->n_cookieverf.nfsuquad[1];
 2169                 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
 2170                 *tl = txdr_unsigned(nmp->nm_rsize);
 2171                 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
 2172                 nfsm_postop_attr(vp, attrflag);
 2173                 if (error) {
 2174                         m_freem(mrep);
 2175                         goto nfsmout;
 2176                 }
 2177                 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2178                 dnp->n_cookieverf.nfsuquad[0] = *tl++;
 2179                 dnp->n_cookieverf.nfsuquad[1] = *tl++;
 2180                 more_dirs = fxdr_unsigned(int, *tl);
 2181 
 2182                 /* loop thru the dir entries, doctoring them to 4bsd form */
 2183                 while (more_dirs && bigenough) {
 2184                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2185                         fileno = fxdr_hyper(tl);
 2186                         len = fxdr_unsigned(int, *(tl + 2));
 2187                         if (len <= 0 || len > NFS_MAXNAMLEN) {
 2188                                 error = EBADRPC;
 2189                                 m_freem(mrep);
 2190                                 goto nfsmout;
 2191                         }
 2192                         tlen = nfsm_rndup(len);
 2193                         if (tlen == len)
 2194                                 tlen += 4;      /* To ensure null termination*/
 2195                         left = DIRBLKSIZ - blksiz;
 2196                         if ((tlen + DIRHDSIZ) > left) {
 2197                                 dp->d_reclen += left;
 2198                                 uiop->uio_iov->iov_base =
 2199                                     (char *)uiop->uio_iov->iov_base + left;
 2200                                 uiop->uio_iov->iov_len -= left;
 2201                                 uiop->uio_offset += left;
 2202                                 uiop->uio_resid -= left;
 2203                                 blksiz = 0;
 2204                         }
 2205                         if ((tlen + DIRHDSIZ) > uiop->uio_resid)
 2206                                 bigenough = 0;
 2207                         if (bigenough) {
 2208                                 dp = (struct dirent *)uiop->uio_iov->iov_base;
 2209                                 dp->d_fileno = (int)fileno;
 2210                                 dp->d_namlen = len;
 2211                                 dp->d_reclen = tlen + DIRHDSIZ;
 2212                                 dp->d_type = DT_UNKNOWN;
 2213                                 blksiz += dp->d_reclen;
 2214                                 if (blksiz == DIRBLKSIZ)
 2215                                         blksiz = 0;
 2216                                 uiop->uio_offset += DIRHDSIZ;
 2217                                 uiop->uio_resid -= DIRHDSIZ;
 2218                                 uiop->uio_iov->iov_base =
 2219                                     (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
 2220                                 uiop->uio_iov->iov_len -= DIRHDSIZ;
 2221                                 cnp->cn_nameptr = uiop->uio_iov->iov_base;
 2222                                 cnp->cn_namelen = len;
 2223                                 nfsm_mtouio(uiop, len);
 2224                                 cp = uiop->uio_iov->iov_base;
 2225                                 tlen -= len;
 2226                                 *cp = '\0';
 2227                                 uiop->uio_iov->iov_base =
 2228                                     (char *)uiop->uio_iov->iov_base + tlen;
 2229                                 uiop->uio_iov->iov_len -= tlen;
 2230                                 uiop->uio_offset += tlen;
 2231                                 uiop->uio_resid -= tlen;
 2232                         } else
 2233                                 nfsm_adv(nfsm_rndup(len));
 2234                         tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
 2235                         if (bigenough) {
 2236                                 cookie.nfsuquad[0] = *tl++;
 2237                                 cookie.nfsuquad[1] = *tl++;
 2238                         } else
 2239                                 tl += 2;
 2240 
 2241                         /*
 2242                          * Since the attributes are before the file handle
 2243                          * (sigh), we must skip over the attributes and then
 2244                          * come back and get them.
 2245                          */
 2246                         attrflag = fxdr_unsigned(int, *tl);
 2247                         if (attrflag) {
 2248                             dpossav1 = dpos;
 2249                             mdsav1 = md;
 2250                             nfsm_adv(NFSX_V3FATTR);
 2251                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2252                             doit = fxdr_unsigned(int, *tl);
 2253                             /*
 2254                              * Skip loading the attrs for "..". There's a 
 2255                              * race between loading the attrs here and 
 2256                              * lookups that look for the directory currently
 2257                              * being read (in the parent). We try to acquire
 2258                              * the exclusive lock on ".." here, owning the 
 2259                              * lock on the directory being read. Lookup will
 2260                              * hold the lock on ".." and try to acquire the 
 2261                              * lock on the directory being read.
 2262                              * 
 2263                              * There are other ways of fixing this, one would
 2264                              * be to do a trylock on the ".." vnode and skip
 2265                              * loading the attrs on ".." if it happens to be 
 2266                              * locked by another process. But skipping the
 2267                              * attrload on ".." seems the easiest option.
 2268                              */
 2269                             if (strcmp(dp->d_name, "..") == 0) {
 2270                                     doit = 0;
 2271                                     /*
 2272                                      * We've already skipped over the attrs, 
 2273                                      * skip over the filehandle. And store d_type
 2274                                      * as VDIR.
 2275                                      */
 2276                                     tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2277                                     i = fxdr_unsigned(int, *tl);
 2278                                     nfsm_adv(nfsm_rndup(i));
 2279                                     dp->d_type = IFTODT(VTTOIF(VDIR));
 2280                             }       
 2281                             if (doit) {
 2282                                 nfsm_getfh(fhp, fhsize, 1);
 2283                                 if (NFS_CMPFH(dnp, fhp, fhsize)) {
 2284                                     VREF(vp);
 2285                                     newvp = vp;
 2286                                     np = dnp;
 2287                                 } else {
 2288                                     error = nfs_nget(vp->v_mount, fhp,
 2289                                         fhsize, &np);
 2290                                     if (error)
 2291                                         doit = 0;
 2292                                     else
 2293                                         newvp = NFSTOV(np);
 2294                                 }
 2295                             }
 2296                             if (doit && bigenough) {
 2297                                 dpossav2 = dpos;
 2298                                 dpos = dpossav1;
 2299                                 mdsav2 = md;
 2300                                 md = mdsav1;
 2301                                 nfsm_loadattr(newvp, NULL);
 2302                                 dpos = dpossav2;
 2303                                 md = mdsav2;
 2304                                 dp->d_type =
 2305                                     IFTODT(VTTOIF(np->n_vattr.va_type));
 2306                                 ndp->ni_vp = newvp;
 2307                                 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
 2308                             }
 2309                         } else {
 2310                             /* Just skip over the file handle */
 2311                             tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2312                             i = fxdr_unsigned(int, *tl);
 2313                             nfsm_adv(nfsm_rndup(i));
 2314                         }
 2315                         if (newvp != NULLVP) {
 2316                             if (newvp == vp)
 2317                                 vrele(newvp);
 2318                             else
 2319                                 vput(newvp);
 2320                             newvp = NULLVP;
 2321                         }
 2322                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2323                         more_dirs = fxdr_unsigned(int, *tl);
 2324                 }
 2325                 /*
 2326                  * If at end of rpc data, get the eof boolean
 2327                  */
 2328                 if (!more_dirs) {
 2329                         tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
 2330                         more_dirs = (fxdr_unsigned(int, *tl) == 0);
 2331                 }
 2332                 m_freem(mrep);
 2333         }
 2334         /*
 2335          * Fill last record, iff any, out to a multiple of DIRBLKSIZ
 2336          * by increasing d_reclen for the last record.
 2337          */
 2338         if (blksiz > 0) {
 2339                 left = DIRBLKSIZ - blksiz;
 2340                 dp->d_reclen += left;
 2341                 uiop->uio_iov->iov_base =
 2342                     (char *)uiop->uio_iov->iov_base + left;
 2343                 uiop->uio_iov->iov_len -= left;
 2344                 uiop->uio_offset += left;
 2345                 uiop->uio_resid -= left;
 2346         }
 2347 
 2348         /*
 2349          * We are now either at the end of the directory or have filled the
 2350          * block.
 2351          */
 2352         if (bigenough)
 2353                 dnp->n_direofoffset = uiop->uio_offset;
 2354         else {
 2355                 if (uiop->uio_resid > 0)
 2356                         printf("EEK! readdirplusrpc resid > 0\n");
 2357                 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
 2358                 *cookiep = cookie;
 2359         }
 2360 nfsmout:
 2361         if (newvp != NULLVP) {
 2362                 if (newvp == vp)
 2363                         vrele(newvp);
 2364                 else
 2365                         vput(newvp);
 2366                 newvp = NULLVP;
 2367         }
 2368         return (error);
 2369 }
 2370 
 2371 /*
 2372  * Silly rename. To make the NFS filesystem that is stateless look a little
 2373  * more like the "ufs" a remove of an active vnode is translated to a rename
 2374  * to a funny looking filename that is removed by nfs_inactive on the
 2375  * nfsnode. There is the potential for another process on a different client
 2376  * to create the same funny name between the nfs_lookitup() fails and the
 2377  * nfs_rename() completes, but...
 2378  */
 2379 static int
 2380 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 2381 {
 2382         struct sillyrename *sp;
 2383         struct nfsnode *np;
 2384         int error;
 2385         short pid;
 2386         unsigned int lticks;
 2387 
 2388         cache_purge(dvp);
 2389         np = VTONFS(vp);
 2390 #ifndef DIAGNOSTIC
 2391         if (vp->v_type == VDIR)
 2392                 panic("nfs: sillyrename dir");
 2393 #endif
 2394         MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
 2395                 M_NFSREQ, M_WAITOK);
 2396         sp->s_cred = crhold(cnp->cn_cred);
 2397         sp->s_dvp = dvp;
 2398         sp->s_removeit = nfs_removeit;
 2399         VREF(dvp);
 2400 
 2401         /* 
 2402          * Fudge together a funny name.
 2403          * Changing the format of the funny name to accomodate more 
 2404          * sillynames per directory.
 2405          * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is 
 2406          * CPU ticks since boot.
 2407          */
 2408         pid = cnp->cn_thread->td_proc->p_pid;
 2409         lticks = (unsigned int)ticks;
 2410         for ( ; ; ) {
 2411                 sp->s_namlen = sprintf(sp->s_name, 
 2412                                        ".nfs.%08x.%04x4.4", lticks, 
 2413                                        pid);
 2414                 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2415                                  cnp->cn_thread, NULL))
 2416                         break;
 2417                 lticks++;
 2418         }
 2419         error = nfs_renameit(dvp, cnp, sp);
 2420         if (error)
 2421                 goto bad;
 2422         error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
 2423                 cnp->cn_thread, &np);
 2424         np->n_sillyrename = sp;
 2425         return (0);
 2426 bad:
 2427         vrele(sp->s_dvp);
 2428         crfree(sp->s_cred);
 2429         free((caddr_t)sp, M_NFSREQ);
 2430         return (error);
 2431 }
 2432 
 2433 /*
 2434  * Look up a file name and optionally either update the file handle or
 2435  * allocate an nfsnode, depending on the value of npp.
 2436  * npp == NULL  --> just do the lookup
 2437  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
 2438  *                      handled too
 2439  * *npp != NULL --> update the file handle in the vnode
 2440  */
 2441 static int
 2442 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
 2443     struct thread *td, struct nfsnode **npp)
 2444 {
 2445         struct vnode *newvp = NULL;
 2446         struct nfsnode *np, *dnp = VTONFS(dvp);
 2447         caddr_t bpos, dpos;
 2448         int error = 0, fhlen, attrflag;
 2449         struct mbuf *mreq, *mrep, *md, *mb;
 2450         nfsfh_t *nfhp;
 2451         int v3 = NFS_ISV3(dvp);
 2452 
 2453         nfsstats.rpccnt[NFSPROC_LOOKUP]++;
 2454         mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
 2455                 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
 2456         mb = mreq;
 2457         bpos = mtod(mb, caddr_t);
 2458         nfsm_fhtom(dvp, v3);
 2459         nfsm_strtom(name, len, NFS_MAXNAMLEN);
 2460         nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
 2461         if (npp && !error) {
 2462                 nfsm_getfh(nfhp, fhlen, v3);
 2463                 if (*npp) {
 2464                     np = *npp;
 2465                     if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
 2466                         free((caddr_t)np->n_fhp, M_NFSBIGFH);
 2467                         np->n_fhp = &np->n_fh;
 2468                     } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
 2469                         np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
 2470                     bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
 2471                     np->n_fhsize = fhlen;
 2472                     newvp = NFSTOV(np);
 2473                 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
 2474                     VREF(dvp);
 2475                     newvp = dvp;
 2476                 } else {
 2477                     error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np);
 2478                     if (error) {
 2479                         m_freem(mrep);
 2480                         return (error);
 2481                     }
 2482                     newvp = NFSTOV(np);
 2483                 }
 2484                 if (v3) {
 2485                         nfsm_postop_attr(newvp, attrflag);
 2486                         if (!attrflag && *npp == NULL) {
 2487                                 m_freem(mrep);
 2488                                 if (newvp == dvp)
 2489                                         vrele(newvp);
 2490                                 else
 2491                                         vput(newvp);
 2492                                 return (ENOENT);
 2493                         }
 2494                 } else
 2495                         nfsm_loadattr(newvp, NULL);
 2496         }
 2497         m_freem(mrep);
 2498 nfsmout:
 2499         if (npp && *npp == NULL) {
 2500                 if (error) {
 2501                         if (newvp) {
 2502                                 if (newvp == dvp)
 2503                                         vrele(newvp);
 2504                                 else
 2505                                         vput(newvp);
 2506                         }
 2507                 } else
 2508                         *npp = np;
 2509         }
 2510         return (error);
 2511 }
 2512 
 2513 /*
 2514  * Nfs Version 3 commit rpc
 2515  */
 2516 int
 2517 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
 2518     struct thread *td)
 2519 {
 2520         u_int32_t *tl;
 2521         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2522         caddr_t bpos, dpos;
 2523         int error = 0, wccflag = NFSV3_WCCRATTR;
 2524         struct mbuf *mreq, *mrep, *md, *mb;
 2525 
 2526         if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0)
 2527                 return (0);
 2528         nfsstats.rpccnt[NFSPROC_COMMIT]++;
 2529         mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
 2530         mb = mreq;
 2531         bpos = mtod(mb, caddr_t);
 2532         nfsm_fhtom(vp, 1);
 2533         tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
 2534         txdr_hyper(offset, tl);
 2535         tl += 2;
 2536         *tl = txdr_unsigned(cnt);
 2537         nfsm_request(vp, NFSPROC_COMMIT, td, cred);
 2538         nfsm_wcc_data(vp, wccflag);
 2539         if (!error) {
 2540                 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
 2541                 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
 2542                         NFSX_V3WRITEVERF)) {
 2543                         bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
 2544                                 NFSX_V3WRITEVERF);
 2545                         error = NFSERR_STALEWRITEVERF;
 2546                 }
 2547         }
 2548         m_freem(mrep);
 2549 nfsmout:
 2550         return (error);
 2551 }
 2552 
 2553 /*
 2554  * Strategy routine.
 2555  * For async requests when nfsiod(s) are running, queue the request by
 2556  * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
 2557  * request.
 2558  */
 2559 static int
 2560 nfs_strategy(struct vop_strategy_args *ap)
 2561 {
 2562         struct buf *bp = ap->a_bp;
 2563         struct ucred *cr;
 2564         struct thread *td;
 2565         int error = 0;
 2566 
 2567         KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
 2568         KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
 2569 
 2570         if (bp->b_flags & B_ASYNC)
 2571                 td = NULL;
 2572         else
 2573                 td = curthread; /* XXX */
 2574 
 2575         if (bp->b_iocmd == BIO_READ)
 2576                 cr = bp->b_rcred;
 2577         else
 2578                 cr = bp->b_wcred;
 2579 
 2580         /*
 2581          * If the op is asynchronous and an i/o daemon is waiting
 2582          * queue the request, wake it up and wait for completion
 2583          * otherwise just do it ourselves.
 2584          */
 2585         if ((bp->b_flags & B_ASYNC) == 0 ||
 2586                 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, td))
 2587                 error = nfs_doio(ap->a_vp, bp, cr, td);
 2588         return (error);
 2589 }
 2590 
 2591 /*
 2592  * fsync vnode op. Just call nfs_flush() with commit == 1.
 2593  */
 2594 /* ARGSUSED */
 2595 static int
 2596 nfs_fsync(struct vop_fsync_args *ap)
 2597 {
 2598 
 2599         return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
 2600 }
 2601 
 2602 /*
 2603  * Flush all the blocks associated with a vnode.
 2604  *      Walk through the buffer pool and push any dirty pages
 2605  *      associated with the vnode.
 2606  */
 2607 static int
 2608 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
 2609     int commit)
 2610 {
 2611         struct nfsnode *np = VTONFS(vp);
 2612         struct buf *bp;
 2613         int i;
 2614         struct buf *nbp;
 2615         struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 2616         int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
 2617         int passone = 1;
 2618         u_quad_t off, endoff, toff;
 2619         struct ucred* wcred = NULL;
 2620         struct buf **bvec = NULL;
 2621 #ifndef NFS_COMMITBVECSIZ
 2622 #define NFS_COMMITBVECSIZ       20
 2623 #endif
 2624         struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
 2625         int bvecsize = 0, bveccount;
 2626 
 2627         if (nmp->nm_flag & NFSMNT_INT)
 2628                 slpflag = PCATCH;
 2629         if (!commit)
 2630                 passone = 0;
 2631         /*
 2632          * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
 2633          * server, but nas not been committed to stable storage on the server
 2634          * yet. On the first pass, the byte range is worked out and the commit
 2635          * rpc is done. On the second pass, nfs_writebp() is called to do the
 2636          * job.
 2637          */
 2638 again:
 2639         off = (u_quad_t)-1;
 2640         endoff = 0;
 2641         bvecpos = 0;
 2642         if (NFS_ISV3(vp) && commit) {
 2643                 s = splbio();
 2644                 if (bvec != NULL && bvec != bvec_on_stack)
 2645                         free(bvec, M_TEMP);
 2646                 /*
 2647                  * Count up how many buffers waiting for a commit.
 2648                  */
 2649                 bveccount = 0;
 2650                 VI_LOCK(vp);
 2651                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2652                         if (BUF_REFCNT(bp) == 0 &&
 2653                             (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
 2654                                 == (B_DELWRI | B_NEEDCOMMIT))
 2655                                 bveccount++;
 2656                 }
 2657                 /*
 2658                  * Allocate space to remember the list of bufs to commit.  It is
 2659                  * important to use M_NOWAIT here to avoid a race with nfs_write.
 2660                  * If we can't get memory (for whatever reason), we will end up
 2661                  * committing the buffers one-by-one in the loop below.
 2662                  */
 2663                 if (bveccount > NFS_COMMITBVECSIZ) {
 2664                         /*
 2665                          * Release the vnode interlock to avoid a lock
 2666                          * order reversal.
 2667                          */
 2668                         VI_UNLOCK(vp);
 2669                         bvec = (struct buf **)
 2670                                 malloc(bveccount * sizeof(struct buf *),
 2671                                        M_TEMP, M_NOWAIT);
 2672                         VI_LOCK(vp);
 2673                         if (bvec == NULL) {
 2674                                 bvec = bvec_on_stack;
 2675                                 bvecsize = NFS_COMMITBVECSIZ;
 2676                         } else
 2677                                 bvecsize = bveccount;
 2678                 } else {
 2679                         bvec = bvec_on_stack;
 2680                         bvecsize = NFS_COMMITBVECSIZ;
 2681                 }
 2682                 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2683                         if (bvecpos >= bvecsize)
 2684                                 break;
 2685                         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2686                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2687                                 continue;
 2688                         }
 2689                         if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
 2690                             (B_DELWRI | B_NEEDCOMMIT)) {
 2691                                 BUF_UNLOCK(bp);
 2692                                 nbp = TAILQ_NEXT(bp, b_bobufs);
 2693                                 continue;
 2694                         }
 2695                         VI_UNLOCK(vp);
 2696                         bremfree(bp);
 2697                         /*
 2698                          * Work out if all buffers are using the same cred
 2699                          * so we can deal with them all with one commit.
 2700                          *
 2701                          * NOTE: we are not clearing B_DONE here, so we have
 2702                          * to do it later on in this routine if we intend to
 2703                          * initiate I/O on the bp.
 2704                          *
 2705                          * Note: to avoid loopback deadlocks, we do not
 2706                          * assign b_runningbufspace.
 2707                          */
 2708                         if (wcred == NULL)
 2709                                 wcred = bp->b_wcred;
 2710                         else if (wcred != bp->b_wcred)
 2711                                 wcred = NOCRED;
 2712                         vfs_busy_pages(bp, 1);
 2713 
 2714                         VI_LOCK(vp);
 2715                         /*
 2716                          * bp is protected by being locked, but nbp is not
 2717                          * and vfs_busy_pages() may sleep.  We have to
 2718                          * recalculate nbp.
 2719                          */
 2720                         nbp = TAILQ_NEXT(bp, b_bobufs);
 2721 
 2722                         /*
 2723                          * A list of these buffers is kept so that the
 2724                          * second loop knows which buffers have actually
 2725                          * been committed. This is necessary, since there
 2726                          * may be a race between the commit rpc and new
 2727                          * uncommitted writes on the file.
 2728                          */
 2729                         bvec[bvecpos++] = bp;
 2730                         toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2731                                 bp->b_dirtyoff;
 2732                         if (toff < off)
 2733                                 off = toff;
 2734                         toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
 2735                         if (toff > endoff)
 2736                                 endoff = toff;
 2737                 }
 2738                 splx(s);
 2739                 VI_UNLOCK(vp);
 2740         }
 2741         if (bvecpos > 0) {
 2742                 /*
 2743                  * Commit data on the server, as required.
 2744                  * If all bufs are using the same wcred, then use that with
 2745                  * one call for all of them, otherwise commit each one
 2746                  * separately.
 2747                  */
 2748                 if (wcred != NOCRED)
 2749                         retv = nfs_commit(vp, off, (int)(endoff - off),
 2750                                           wcred, td);
 2751                 else {
 2752                         retv = 0;
 2753                         for (i = 0; i < bvecpos; i++) {
 2754                                 off_t off, size;
 2755                                 bp = bvec[i];
 2756                                 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
 2757                                         bp->b_dirtyoff;
 2758                                 size = (u_quad_t)(bp->b_dirtyend
 2759                                                   - bp->b_dirtyoff);
 2760                                 retv = nfs_commit(vp, off, (int)size,
 2761                                                   bp->b_wcred, td);
 2762                                 if (retv) break;
 2763                         }
 2764                 }
 2765 
 2766                 if (retv == NFSERR_STALEWRITEVERF)
 2767                         nfs_clearcommit(vp->v_mount);
 2768 
 2769                 /*
 2770                  * Now, either mark the blocks I/O done or mark the
 2771                  * blocks dirty, depending on whether the commit
 2772                  * succeeded.
 2773                  */
 2774                 for (i = 0; i < bvecpos; i++) {
 2775                         bp = bvec[i];
 2776                         bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
 2777                         if (retv) {
 2778                                 /*
 2779                                  * Error, leave B_DELWRI intact
 2780                                  */
 2781                                 vfs_unbusy_pages(bp);
 2782                                 brelse(bp);
 2783                         } else {
 2784                                 /*
 2785                                  * Success, remove B_DELWRI ( bundirty() ).
 2786                                  *
 2787                                  * b_dirtyoff/b_dirtyend seem to be NFS
 2788                                  * specific.  We should probably move that
 2789                                  * into bundirty(). XXX
 2790                                  */
 2791                                 s = splbio();
 2792                                 bufobj_wref(&vp->v_bufobj);
 2793                                 bp->b_flags |= B_ASYNC;
 2794                                 bundirty(bp);
 2795                                 bp->b_flags &= ~B_DONE;
 2796                                 bp->b_ioflags &= ~BIO_ERROR;
 2797                                 bp->b_dirtyoff = bp->b_dirtyend = 0;
 2798                                 splx(s);
 2799                                 bufdone(bp);
 2800                         }
 2801                 }
 2802         }
 2803 
 2804         /*
 2805          * Start/do any write(s) that are required.
 2806          */
 2807 loop:
 2808         s = splbio();
 2809         VI_LOCK(vp);
 2810         TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
 2811                 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
 2812                         if (waitfor != MNT_WAIT || passone)
 2813                                 continue;
 2814 
 2815                         error = BUF_TIMELOCK(bp,
 2816                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 2817                             VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
 2818                         splx(s);
 2819                         if (error == 0)
 2820                                 panic("nfs_fsync: inconsistent lock");
 2821                         if (error == ENOLCK)
 2822                                 goto loop;
 2823                         if (nfs_sigintr(nmp, NULL, td)) {
 2824                                 error = EINTR;
 2825                                 goto done;
 2826                         }
 2827                         if (slpflag == PCATCH) {
 2828                                 slpflag = 0;
 2829                                 slptimeo = 2 * hz;
 2830                         }
 2831                         goto loop;
 2832                 }
 2833                 if ((bp->b_flags & B_DELWRI) == 0)
 2834                         panic("nfs_fsync: not dirty");
 2835                 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
 2836                         BUF_UNLOCK(bp);
 2837                         continue;
 2838                 }
 2839                 VI_UNLOCK(vp);
 2840                 bremfree(bp);
 2841                 if (passone || !commit)
 2842                     bp->b_flags |= B_ASYNC;
 2843                 else
 2844                     bp->b_flags |= B_ASYNC;
 2845                 splx(s);
 2846                 bwrite(bp);
 2847                 if (nfs_sigintr(nmp, NULL, td)) {
 2848                         error = EINTR;
 2849                         goto done;
 2850                 }
 2851                 goto loop;
 2852         }
 2853         splx(s);
 2854         if (passone) {
 2855                 passone = 0;
 2856                 VI_UNLOCK(vp);
 2857                 goto again;
 2858         }
 2859         if (waitfor == MNT_WAIT) {
 2860                 while (vp->v_bufobj.bo_numoutput) {
 2861                         error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
 2862                         if (error) {
 2863                             VI_UNLOCK(vp);
 2864                             error = nfs_sigintr(nmp, NULL, td);
 2865                             if (error)
 2866                                 goto done;
 2867                             if (slpflag == PCATCH) {
 2868                                 slpflag = 0;
 2869                                 slptimeo = 2 * hz;
 2870                             }
 2871                             VI_LOCK(vp);
 2872                         }
 2873                 }
 2874                 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
 2875                         VI_UNLOCK(vp);
 2876                         goto loop;
 2877                 }
 2878         }
 2879         VI_UNLOCK(vp);
 2880         if (np->n_flag & NWRITEERR) {
 2881                 error = np->n_error;
 2882                 np->n_flag &= ~NWRITEERR;
 2883         }
 2884         if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0)
 2885                 np->n_flag &= ~NMODIFIED;
 2886 done:
 2887         if (bvec != NULL && bvec != bvec_on_stack)
 2888                 free(bvec, M_TEMP);
 2889         return (error);
 2890 }
 2891 
 2892 /*
 2893  * NFS advisory byte-level locks.
 2894  */
 2895 static int
 2896 nfs_advlock(struct vop_advlock_args *ap)
 2897 {
 2898 
 2899         if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
 2900                 struct nfsnode *np = VTONFS(ap->a_vp);
 2901 
 2902                 return (lf_advlock(ap, &(np->n_lockf), np->n_size));
 2903         }
 2904         return (nfs_dolock(ap));
 2905 }
 2906 
 2907 /*
 2908  * Print out the contents of an nfsnode.
 2909  */
 2910 static int
 2911 nfs_print(struct vop_print_args *ap)
 2912 {
 2913         struct vnode *vp = ap->a_vp;
 2914         struct nfsnode *np = VTONFS(vp);
 2915 
 2916         printf("\tfileid %ld fsid 0x%x",
 2917            np->n_vattr.va_fileid, np->n_vattr.va_fsid);
 2918         if (vp->v_type == VFIFO)
 2919                 fifo_printinfo(vp);
 2920         printf("\n");
 2921         return (0);
 2922 }
 2923 
 2924 /*
 2925  * This is the "real" nfs::bwrite(struct buf*).
 2926  * We set B_CACHE if this is a VMIO buffer.
 2927  */
 2928 int
 2929 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
 2930 {
 2931         int s;
 2932         int oldflags = bp->b_flags;
 2933 #if 0
 2934         int retv = 1;
 2935         off_t off;
 2936 #endif
 2937 
 2938         if (BUF_REFCNT(bp) == 0)
 2939                 panic("bwrite: buffer is not locked???");
 2940 
 2941         if (bp->b_flags & B_INVAL) {
 2942                 brelse(bp);
 2943                 return(0);
 2944         }
 2945 
 2946         bp->b_flags |= B_CACHE;
 2947 
 2948         /*
 2949          * Undirty the bp.  We will redirty it later if the I/O fails.
 2950          */
 2951 
 2952         s = splbio();
 2953         bundirty(bp);
 2954         bp->b_flags &= ~B_DONE;
 2955         bp->b_ioflags &= ~BIO_ERROR;
 2956         bp->b_iocmd = BIO_WRITE;
 2957 
 2958         bufobj_wref(bp->b_bufobj);
 2959         curthread->td_proc->p_stats->p_ru.ru_oublock++;
 2960         splx(s);
 2961 
 2962         /*
 2963          * Note: to avoid loopback deadlocks, we do not
 2964          * assign b_runningbufspace.
 2965          */
 2966         vfs_busy_pages(bp, 1);
 2967 
 2968         BUF_KERNPROC(bp);
 2969         bp->b_iooffset = dbtob(bp->b_blkno);
 2970         bstrategy(bp);
 2971 
 2972         if( (oldflags & B_ASYNC) == 0) {
 2973                 int rtval = bufwait(bp);
 2974 
 2975                 if (oldflags & B_DELWRI) {
 2976                         s = splbio();
 2977                         reassignbuf(bp);
 2978                         splx(s);
 2979                 }
 2980 
 2981                 brelse(bp);
 2982                 return (rtval);
 2983         }
 2984 
 2985         return (0);
 2986 }
 2987 
 2988 /*
 2989  * nfs special file access vnode op.
 2990  * Essentially just get vattr and then imitate iaccess() since the device is
 2991  * local to the client.
 2992  */
 2993 static int
 2994 nfsspec_access(struct vop_access_args *ap)
 2995 {
 2996         struct vattr *vap;
 2997         struct ucred *cred = ap->a_cred;
 2998         struct vnode *vp = ap->a_vp;
 2999         mode_t mode = ap->a_mode;
 3000         struct vattr vattr;
 3001         int error;
 3002 
 3003         /*
 3004          * Disallow write attempts on filesystems mounted read-only;
 3005          * unless the file is a socket, fifo, or a block or character
 3006          * device resident on the filesystem.
 3007          */
 3008         if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
 3009                 switch (vp->v_type) {
 3010                 case VREG:
 3011                 case VDIR:
 3012                 case VLNK:
 3013                         return (EROFS);
 3014                 default:
 3015                         break;
 3016                 }
 3017         }
 3018         vap = &vattr;
 3019         error = VOP_GETATTR(vp, vap, cred, ap->a_td);
 3020         if (error)
 3021                 return (error);
 3022         return (vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
 3023             mode, cred, NULL));
 3024 }
 3025 
 3026 /*
 3027  * Read wrapper for fifos.
 3028  */
 3029 static int
 3030 nfsfifo_read(struct vop_read_args *ap)
 3031 {
 3032         struct nfsnode *np = VTONFS(ap->a_vp);
 3033 
 3034         /*
 3035          * Set access flag.
 3036          */
 3037         np->n_flag |= NACC;
 3038         getnanotime(&np->n_atim);
 3039         return (fifo_specops.vop_read(ap));
 3040 }
 3041 
 3042 /*
 3043  * Write wrapper for fifos.
 3044  */
 3045 static int
 3046 nfsfifo_write(struct vop_write_args *ap)
 3047 {
 3048         struct nfsnode *np = VTONFS(ap->a_vp);
 3049 
 3050         /*
 3051          * Set update flag.
 3052          */
 3053         np->n_flag |= NUPD;
 3054         getnanotime(&np->n_mtim);
 3055         return (fifo_specops.vop_write(ap));
 3056 }
 3057 
 3058 /*
 3059  * Close wrapper for fifos.
 3060  *
 3061  * Update the times on the nfsnode then do fifo close.
 3062  */
 3063 static int
 3064 nfsfifo_close(struct vop_close_args *ap)
 3065 {
 3066         struct vnode *vp = ap->a_vp;
 3067         struct nfsnode *np = VTONFS(vp);
 3068         struct vattr vattr;
 3069         struct timespec ts;
 3070 
 3071         if (np->n_flag & (NACC | NUPD)) {
 3072                 getnanotime(&ts);
 3073                 if (np->n_flag & NACC)
 3074                         np->n_atim = ts;
 3075                 if (np->n_flag & NUPD)
 3076                         np->n_mtim = ts;
 3077                 np->n_flag |= NCHG;
 3078                 if (vrefcnt(vp) == 1 &&
 3079                     (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
 3080                         VATTR_NULL(&vattr);
 3081                         if (np->n_flag & NACC)
 3082                                 vattr.va_atime = np->n_atim;
 3083                         if (np->n_flag & NUPD)
 3084                                 vattr.va_mtime = np->n_mtim;
 3085                         (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
 3086                 }
 3087         }
 3088         return (fifo_specops.vop_close(ap));
 3089 }
 3090 
 3091 /*
 3092  * Just call nfs_writebp() with the force argument set to 1.
 3093  *
 3094  * NOTE: B_DONE may or may not be set in a_bp on call.
 3095  */
 3096 static int
 3097 nfs_bwrite(struct buf *bp)
 3098 {
 3099 
 3100         return (nfs_writebp(bp, 1, curthread));
 3101 }
 3102 
 3103 struct buf_ops buf_ops_nfs = {
 3104         .bop_name       =       "buf_ops_nfs",
 3105         .bop_write      =       nfs_bwrite,
 3106         .bop_strategy   =       bufstrategy,
 3107         .bop_sync       =       bufsync,
 3108 };

Cache object: aff71cb96dbe85e0277e586b1dba4c7d


[ 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.