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/fs/msdosfs/msdosfs_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 /* $FreeBSD$ */
    2 /*      $NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $   */
    3 
    4 /*-
    5  * SPDX-License-Identifier: BSD-4-Clause
    6  *
    7  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
    8  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
    9  * All rights reserved.
   10  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
   11  *
   12  * Redistribution and use in source and binary forms, with or without
   13  * modification, are permitted provided that the following conditions
   14  * are met:
   15  * 1. Redistributions of source code must retain the above copyright
   16  *    notice, this list of conditions and the following disclaimer.
   17  * 2. Redistributions in binary form must reproduce the above copyright
   18  *    notice, this list of conditions and the following disclaimer in the
   19  *    documentation and/or other materials provided with the distribution.
   20  * 3. All advertising materials mentioning features or use of this software
   21  *    must display the following acknowledgement:
   22  *      This product includes software developed by TooLs GmbH.
   23  * 4. The name of TooLs GmbH may not be used to endorse or promote products
   24  *    derived from this software without specific prior written permission.
   25  *
   26  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
   27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   28  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
   29  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
   30  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
   31  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
   32  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
   33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
   34  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
   35  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
   36  */
   37 /*-
   38  * Written by Paul Popelka (paulp@uts.amdahl.com)
   39  *
   40  * You can do anything you want with this software, just don't say you wrote
   41  * it, and don't remove this notice.
   42  *
   43  * This software is provided "as is".
   44  *
   45  * The author supplies this software to be publicly redistributed on the
   46  * understanding that the author is not responsible for the correct
   47  * functioning of this software in any circumstances and is not liable for
   48  * any damages caused by this software.
   49  *
   50  * October 1992
   51  */
   52 
   53 #include <sys/param.h>
   54 #include <sys/systm.h>
   55 #include <sys/bio.h>
   56 #include <sys/buf.h>
   57 #include <sys/clock.h>
   58 #include <sys/dirent.h>
   59 #include <sys/lock.h>
   60 #include <sys/lockf.h>
   61 #include <sys/malloc.h>
   62 #include <sys/mount.h>
   63 #include <sys/mutex.h>
   64 #include <sys/namei.h>
   65 #include <sys/priv.h>
   66 #include <sys/stat.h>
   67 #include <sys/sysctl.h>
   68 #include <sys/unistd.h>
   69 #include <sys/vmmeter.h>
   70 #include <sys/vnode.h>
   71 
   72 #include <vm/vm.h>
   73 #include <vm/vm_extern.h>
   74 #include <vm/vnode_pager.h>
   75 
   76 #include <fs/msdosfs/bpb.h>
   77 #include <fs/msdosfs/direntry.h>
   78 #include <fs/msdosfs/denode.h>
   79 #include <fs/msdosfs/fat.h>
   80 #include <fs/msdosfs/msdosfsmount.h>
   81 
   82 /*
   83  * Prototypes for MSDOSFS vnode operations
   84  */
   85 static vop_create_t     msdosfs_create;
   86 static vop_mknod_t      msdosfs_mknod;
   87 static vop_open_t       msdosfs_open;
   88 static vop_close_t      msdosfs_close;
   89 static vop_access_t     msdosfs_access;
   90 static vop_getattr_t    msdosfs_getattr;
   91 static vop_setattr_t    msdosfs_setattr;
   92 static vop_read_t       msdosfs_read;
   93 static vop_write_t      msdosfs_write;
   94 static vop_fsync_t      msdosfs_fsync;
   95 static vop_remove_t     msdosfs_remove;
   96 static vop_link_t       msdosfs_link;
   97 static vop_rename_t     msdosfs_rename;
   98 static vop_mkdir_t      msdosfs_mkdir;
   99 static vop_rmdir_t      msdosfs_rmdir;
  100 static vop_symlink_t    msdosfs_symlink;
  101 static vop_readdir_t    msdosfs_readdir;
  102 static vop_bmap_t       msdosfs_bmap;
  103 static vop_getpages_t   msdosfs_getpages;
  104 static vop_strategy_t   msdosfs_strategy;
  105 static vop_print_t      msdosfs_print;
  106 static vop_pathconf_t   msdosfs_pathconf;
  107 static vop_vptofh_t     msdosfs_vptofh;
  108 
  109 /*
  110  * Some general notes:
  111  *
  112  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
  113  * read/written using the vnode for the filesystem. Blocks that represent
  114  * the contents of a file are read/written using the vnode for the file
  115  * (including directories when they are read/written as files). This
  116  * presents problems for the dos filesystem because data that should be in
  117  * an inode (if dos had them) resides in the directory itself.  Since we
  118  * must update directory entries without the benefit of having the vnode
  119  * for the directory we must use the vnode for the filesystem.  This means
  120  * that when a directory is actually read/written (via read, write, or
  121  * readdir, or seek) we must use the vnode for the filesystem instead of
  122  * the vnode for the directory as would happen in ufs. This is to insure we
  123  * retrieve the correct block from the buffer cache since the hash value is
  124  * based upon the vnode address and the desired block number.
  125  */
  126 
  127 /*
  128  * Create a regular file. On entry the directory to contain the file being
  129  * created is locked.  We must release before we return. We must also free
  130  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error.
  131  */
  132 static int
  133 msdosfs_create(struct vop_create_args *ap)
  134 {
  135         struct componentname *cnp = ap->a_cnp;
  136         struct denode ndirent;
  137         struct denode *dep;
  138         struct denode *pdep = VTODE(ap->a_dvp);
  139         struct timespec ts;
  140         int error;
  141 
  142 #ifdef MSDOSFS_DEBUG
  143         printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
  144 #endif
  145 
  146         /*
  147          * If this is the root directory and there is no space left we
  148          * can't do anything.  This is because the root directory can not
  149          * change size.
  150          */
  151         if (pdep->de_StartCluster == MSDOSFSROOT
  152             && pdep->de_fndoffset >= pdep->de_FileSize) {
  153                 error = ENOSPC;
  154                 goto bad;
  155         }
  156 
  157         /*
  158          * Create a directory entry for the file, then call createde() to
  159          * have it installed. NOTE: DOS files are always executable.  We
  160          * use the absence of the owner write bit to make the file
  161          * readonly.
  162          */
  163         memset(&ndirent, 0, sizeof(ndirent));
  164         error = uniqdosname(pdep, cnp, ndirent.de_Name);
  165         if (error)
  166                 goto bad;
  167 
  168         ndirent.de_Attributes = ATTR_ARCHIVE;
  169         ndirent.de_LowerCase = 0;
  170         ndirent.de_StartCluster = 0;
  171         ndirent.de_FileSize = 0;
  172         ndirent.de_pmp = pdep->de_pmp;
  173         ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
  174         vfs_timestamp(&ts);
  175         DETIMES(&ndirent, &ts, &ts, &ts);
  176         error = createde(&ndirent, pdep, &dep, cnp);
  177         if (error)
  178                 goto bad;
  179         *ap->a_vpp = DETOV(dep);
  180         if ((cnp->cn_flags & MAKEENTRY) != 0)
  181                 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
  182         return (0);
  183 
  184 bad:
  185         return (error);
  186 }
  187 
  188 static int
  189 msdosfs_mknod(struct vop_mknod_args *ap)
  190 {
  191 
  192     return (EINVAL);
  193 }
  194 
  195 static int
  196 msdosfs_open(struct vop_open_args *ap)
  197 {
  198         struct denode *dep = VTODE(ap->a_vp);
  199         vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
  200         return 0;
  201 }
  202 
  203 static int
  204 msdosfs_close(struct vop_close_args *ap)
  205 {
  206         struct vnode *vp = ap->a_vp;
  207         struct denode *dep = VTODE(vp);
  208         struct timespec ts;
  209 
  210         VI_LOCK(vp);
  211         if (vp->v_usecount > 1) {
  212                 vfs_timestamp(&ts);
  213                 DETIMES(dep, &ts, &ts, &ts);
  214         }
  215         VI_UNLOCK(vp);
  216         return 0;
  217 }
  218 
  219 static int
  220 msdosfs_access(struct vop_access_args *ap)
  221 {
  222         struct vnode *vp = ap->a_vp;
  223         struct denode *dep = VTODE(ap->a_vp);
  224         struct msdosfsmount *pmp = dep->de_pmp;
  225         mode_t file_mode;
  226         accmode_t accmode = ap->a_accmode;
  227 
  228         file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
  229         file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
  230 
  231         /*
  232          * Disallow writing to directories and regular files if the
  233          * filesystem is read-only.
  234          */
  235         if (accmode & VWRITE) {
  236                 switch (vp->v_type) {
  237                 case VREG:
  238                 case VDIR:
  239                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  240                                 return (EROFS);
  241                         break;
  242                 default:
  243                         break;
  244                 }
  245         }
  246 
  247         return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
  248             ap->a_accmode, ap->a_cred));
  249 }
  250 
  251 static int
  252 msdosfs_getattr(struct vop_getattr_args *ap)
  253 {
  254         struct denode *dep = VTODE(ap->a_vp);
  255         struct msdosfsmount *pmp = dep->de_pmp;
  256         struct vattr *vap = ap->a_vap;
  257         mode_t mode;
  258         struct timespec ts;
  259         u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
  260         uint64_t fileid;
  261 
  262         vfs_timestamp(&ts);
  263         DETIMES(dep, &ts, &ts, &ts);
  264         vap->va_fsid = dev2udev(pmp->pm_dev);
  265         /*
  266          * The following computation of the fileid must be the same as that
  267          * used in msdosfs_readdir() to compute d_fileno. If not, pwd
  268          * doesn't work.
  269          */
  270         if (dep->de_Attributes & ATTR_DIRECTORY) {
  271                 fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
  272                     dirsperblk;
  273                 if (dep->de_StartCluster == MSDOSFSROOT)
  274                         fileid = 1;
  275         } else {
  276                 fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
  277                     dirsperblk;
  278                 if (dep->de_dirclust == MSDOSFSROOT)
  279                         fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
  280                 fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
  281         }
  282         vap->va_fileid = fileid;
  283 
  284         mode = S_IRWXU|S_IRWXG|S_IRWXO;
  285         if (dep->de_Attributes & ATTR_READONLY)
  286                 mode &= ~(S_IWUSR|S_IWGRP|S_IWOTH);
  287         vap->va_mode = mode &
  288             (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
  289         vap->va_uid = pmp->pm_uid;
  290         vap->va_gid = pmp->pm_gid;
  291         vap->va_nlink = 1;
  292         vap->va_rdev = NODEV;
  293         vap->va_size = dep->de_FileSize;
  294         fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
  295         vap->va_ctime = vap->va_mtime;
  296         if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
  297                 fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
  298                 fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
  299                     0, &vap->va_birthtime);
  300         } else {
  301                 vap->va_atime = vap->va_mtime;
  302                 vap->va_birthtime.tv_sec = -1;
  303                 vap->va_birthtime.tv_nsec = 0;
  304         }
  305         vap->va_flags = 0;
  306         if (dep->de_Attributes & ATTR_ARCHIVE)
  307                 vap->va_flags |= UF_ARCHIVE;
  308         if (dep->de_Attributes & ATTR_HIDDEN)
  309                 vap->va_flags |= UF_HIDDEN;
  310         if (dep->de_Attributes & ATTR_READONLY)
  311                 vap->va_flags |= UF_READONLY;
  312         if (dep->de_Attributes & ATTR_SYSTEM)
  313                 vap->va_flags |= UF_SYSTEM;
  314         vap->va_gen = 0;
  315         vap->va_blocksize = pmp->pm_bpcluster;
  316         vap->va_bytes =
  317             (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
  318         vap->va_type = ap->a_vp->v_type;
  319         vap->va_filerev = dep->de_modrev;
  320         return (0);
  321 }
  322 
  323 static int
  324 msdosfs_setattr(struct vop_setattr_args *ap)
  325 {
  326         struct vnode *vp = ap->a_vp;
  327         struct denode *dep = VTODE(ap->a_vp);
  328         struct msdosfsmount *pmp = dep->de_pmp;
  329         struct vattr *vap = ap->a_vap;
  330         struct ucred *cred = ap->a_cred;
  331         struct thread *td = curthread;
  332         int error = 0;
  333 
  334 #ifdef MSDOSFS_DEBUG
  335         printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
  336             ap->a_vp, vap, cred);
  337 #endif
  338 
  339         /*
  340          * Check for unsettable attributes.
  341          */
  342         if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
  343             (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
  344             (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
  345             (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
  346 #ifdef MSDOSFS_DEBUG
  347                 printf("msdosfs_setattr(): returning EINVAL\n");
  348                 printf("    va_type %d, va_nlink %llx, va_fsid %llx, va_fileid %llx\n",
  349                     vap->va_type, (unsigned long long)vap->va_nlink,
  350                     (unsigned long long)vap->va_fsid,
  351                     (unsigned long long)vap->va_fileid);
  352                 printf("    va_blocksize %lx, va_rdev %llx, va_bytes %llx, va_gen %lx\n",
  353                     vap->va_blocksize, (unsigned long long)vap->va_rdev,
  354                     (unsigned long long)vap->va_bytes, vap->va_gen);
  355                 printf("    va_uid %x, va_gid %x\n",
  356                     vap->va_uid, vap->va_gid);
  357 #endif
  358                 return (EINVAL);
  359         }
  360 
  361         /*
  362          * We don't allow setting attributes on the root directory.
  363          * The special case for the root directory is because before
  364          * FAT32, the root directory didn't have an entry for itself
  365          * (and was otherwise special).  With FAT32, the root
  366          * directory is not so special, but still doesn't have an
  367          * entry for itself.
  368          */
  369         if (vp->v_vflag & VV_ROOT)
  370                 return (EINVAL);
  371 
  372         if (vap->va_flags != VNOVAL) {
  373                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
  374                         return (EROFS);
  375                 if (cred->cr_uid != pmp->pm_uid) {
  376                         error = priv_check_cred(cred, PRIV_VFS_ADMIN);
  377                         if (error)
  378                                 return (error);
  379                 }
  380                 /*
  381                  * We are very inconsistent about handling unsupported
  382                  * attributes.  We ignored the access time and the
  383                  * read and execute bits.  We were strict for the other
  384                  * attributes.
  385                  */
  386                 if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
  387                     UF_SYSTEM))
  388                         return EOPNOTSUPP;
  389                 if (vap->va_flags & UF_ARCHIVE)
  390                         dep->de_Attributes |= ATTR_ARCHIVE;
  391                 else
  392                         dep->de_Attributes &= ~ATTR_ARCHIVE;
  393                 if (vap->va_flags & UF_HIDDEN)
  394                         dep->de_Attributes |= ATTR_HIDDEN;
  395                 else
  396                         dep->de_Attributes &= ~ATTR_HIDDEN;
  397                 /* We don't allow changing the readonly bit on directories. */
  398                 if (vp->v_type != VDIR) {
  399                         if (vap->va_flags & UF_READONLY)
  400                                 dep->de_Attributes |= ATTR_READONLY;
  401                         else
  402                                 dep->de_Attributes &= ~ATTR_READONLY;
  403                 }
  404                 if (vap->va_flags & UF_SYSTEM)
  405                         dep->de_Attributes |= ATTR_SYSTEM;
  406                 else
  407                         dep->de_Attributes &= ~ATTR_SYSTEM;
  408                 dep->de_flag |= DE_MODIFIED;
  409         }
  410 
  411         if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
  412                 uid_t uid;
  413                 gid_t gid;
  414 
  415                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
  416                         return (EROFS);
  417                 uid = vap->va_uid;
  418                 if (uid == (uid_t)VNOVAL)
  419                         uid = pmp->pm_uid;
  420                 gid = vap->va_gid;
  421                 if (gid == (gid_t)VNOVAL)
  422                         gid = pmp->pm_gid;
  423                 if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
  424                     (gid != pmp->pm_gid && !groupmember(gid, cred))) {
  425                         error = priv_check_cred(cred, PRIV_VFS_CHOWN);
  426                         if (error)
  427                                 return (error);
  428                 }
  429                 if (uid != pmp->pm_uid || gid != pmp->pm_gid)
  430                         return EINVAL;
  431         }
  432 
  433         if (vap->va_size != VNOVAL) {
  434                 switch (vp->v_type) {
  435                 case VDIR:
  436                         return (EISDIR);
  437                 case VREG:
  438                         /*
  439                          * Truncation is only supported for regular files,
  440                          * Disallow it if the filesystem is read-only.
  441                          */
  442                         if (vp->v_mount->mnt_flag & MNT_RDONLY)
  443                                 return (EROFS);
  444                         break;
  445                 default:
  446                         /*
  447                          * According to POSIX, the result is unspecified
  448                          * for file types other than regular files,
  449                          * directories and shared memory objects.  We
  450                          * don't support any file types except regular
  451                          * files and directories in this file system, so
  452                          * this (default) case is unreachable and can do
  453                          * anything.  Keep falling through to detrunc()
  454                          * for now.
  455                          */
  456                         break;
  457                 }
  458                 error = vn_rlimit_trunc(vap->va_size, td);
  459                 if (error != 0)
  460                         return (error);
  461                 error = detrunc(dep, vap->va_size, 0, cred);
  462                 if (error)
  463                         return error;
  464         }
  465         if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
  466                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
  467                         return (EROFS);
  468                 error = vn_utimes_perm(vp, vap, cred, td);
  469                 if (error != 0)
  470                         return (error);
  471                 if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
  472                     vap->va_atime.tv_sec != VNOVAL) {
  473                         dep->de_flag &= ~DE_ACCESS;
  474                         timespec2fattime(&vap->va_atime, 0,
  475                             &dep->de_ADate, NULL, NULL);
  476                 }
  477                 if (vap->va_mtime.tv_sec != VNOVAL) {
  478                         dep->de_flag &= ~DE_UPDATE;
  479                         timespec2fattime(&vap->va_mtime, 0,
  480                             &dep->de_MDate, &dep->de_MTime, NULL);
  481                 }
  482                 /*
  483                  * We don't set the archive bit when modifying the time of
  484                  * a directory to emulate the Windows/DOS behavior.
  485                  */
  486                 if (vp->v_type != VDIR)
  487                         dep->de_Attributes |= ATTR_ARCHIVE;
  488                 dep->de_flag |= DE_MODIFIED;
  489         }
  490         /*
  491          * DOS files only have the ability to have their writability
  492          * attribute set, so we use the owner write bit to set the readonly
  493          * attribute.
  494          */
  495         if (vap->va_mode != (mode_t)VNOVAL) {
  496                 if (vp->v_mount->mnt_flag & MNT_RDONLY)
  497                         return (EROFS);
  498                 if (cred->cr_uid != pmp->pm_uid) {
  499                         error = priv_check_cred(cred, PRIV_VFS_ADMIN);
  500                         if (error)
  501                                 return (error);
  502                 }
  503                 if (vp->v_type != VDIR) {
  504                         /* We ignore the read and execute bits. */
  505                         if (vap->va_mode & S_IWUSR)
  506                                 dep->de_Attributes &= ~ATTR_READONLY;
  507                         else
  508                                 dep->de_Attributes |= ATTR_READONLY;
  509                         dep->de_Attributes |= ATTR_ARCHIVE;
  510                         dep->de_flag |= DE_MODIFIED;
  511                 }
  512         }
  513         return (deupdat(dep, 0));
  514 }
  515 
  516 static int
  517 msdosfs_read(struct vop_read_args *ap)
  518 {
  519         int error = 0;
  520         int blsize;
  521         int isadir;
  522         ssize_t orig_resid;
  523         u_int n;
  524         u_long diff;
  525         u_long on;
  526         daddr_t lbn;
  527         daddr_t rablock;
  528         int rasize;
  529         int seqcount;
  530         struct buf *bp;
  531         struct vnode *vp = ap->a_vp;
  532         struct denode *dep = VTODE(vp);
  533         struct msdosfsmount *pmp = dep->de_pmp;
  534         struct uio *uio = ap->a_uio;
  535 
  536         /*
  537          * If they didn't ask for any data, then we are done.
  538          */
  539         orig_resid = uio->uio_resid;
  540         if (orig_resid == 0)
  541                 return (0);
  542 
  543         /*
  544          * The caller is supposed to ensure that
  545          * uio->uio_offset >= 0 and uio->uio_resid >= 0.
  546          * We don't need to check for large offsets as in ffs because
  547          * dep->de_FileSize <= MSDOSFS_FILESIZE_MAX < OFF_MAX, so large
  548          * offsets cannot cause overflow even in theory.
  549          */
  550 
  551         seqcount = ap->a_ioflag >> IO_SEQSHIFT;
  552 
  553         isadir = dep->de_Attributes & ATTR_DIRECTORY;
  554         do {
  555                 if (uio->uio_offset >= dep->de_FileSize)
  556                         break;
  557                 lbn = de_cluster(pmp, uio->uio_offset);
  558                 rablock = lbn + 1;
  559                 blsize = pmp->pm_bpcluster;
  560                 on = uio->uio_offset & pmp->pm_crbomask;
  561                 /*
  562                  * If we are operating on a directory file then be sure to
  563                  * do i/o with the vnode for the filesystem instead of the
  564                  * vnode for the directory.
  565                  */
  566                 if (isadir) {
  567                         /* convert cluster # to block # */
  568                         error = pcbmap(dep, lbn, &lbn, 0, &blsize);
  569                         if (error == E2BIG) {
  570                                 error = EINVAL;
  571                                 break;
  572                         } else if (error)
  573                                 break;
  574                         error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
  575                 } else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
  576                         error = bread(vp, lbn, blsize, NOCRED, &bp);
  577                 } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
  578                         error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
  579                             NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
  580                 } else if (seqcount > 1) {
  581                         rasize = blsize;
  582                         error = breadn(vp, lbn,
  583                             blsize, &rablock, &rasize, 1, NOCRED, &bp);
  584                 } else {
  585                         error = bread(vp, lbn, blsize, NOCRED, &bp);
  586                 }
  587                 if (error) {
  588                         brelse(bp);
  589                         break;
  590                 }
  591                 diff = pmp->pm_bpcluster - on;
  592                 n = diff > uio->uio_resid ? uio->uio_resid : diff;
  593                 diff = dep->de_FileSize - uio->uio_offset;
  594                 if (diff < n)
  595                         n = diff;
  596                 diff = blsize - bp->b_resid;
  597                 if (diff < n)
  598                         n = diff;
  599                 error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio);
  600                 brelse(bp);
  601         } while (error == 0 && uio->uio_resid > 0 && n != 0);
  602         if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
  603             (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
  604                 dep->de_flag |= DE_ACCESS;
  605         return (error);
  606 }
  607 
  608 /*
  609  * Write data to a file or directory.
  610  */
  611 static int
  612 msdosfs_write(struct vop_write_args *ap)
  613 {
  614         int n;
  615         int croffset;
  616         ssize_t resid, r;
  617         u_long osize;
  618         int error = 0;
  619         u_long count;
  620         int seqcount;
  621         daddr_t bn, lastcn;
  622         struct buf *bp;
  623         int ioflag = ap->a_ioflag;
  624         struct uio *uio = ap->a_uio;
  625         struct vnode *vp = ap->a_vp;
  626         struct vnode *thisvp;
  627         struct denode *dep = VTODE(vp);
  628         struct msdosfsmount *pmp = dep->de_pmp;
  629         struct ucred *cred = ap->a_cred;
  630 
  631 #ifdef MSDOSFS_DEBUG
  632         printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
  633             vp, uio, ioflag, cred);
  634         printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
  635             dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
  636 #endif
  637 
  638         switch (vp->v_type) {
  639         case VREG:
  640                 if (ioflag & IO_APPEND)
  641                         uio->uio_offset = dep->de_FileSize;
  642                 thisvp = vp;
  643                 break;
  644         case VDIR:
  645                 return EISDIR;
  646         default:
  647                 panic("msdosfs_write(): bad file type");
  648         }
  649 
  650         /*
  651          * This is needed (unlike in ffs_write()) because we extend the
  652          * file outside of the loop but we don't want to extend the file
  653          * for writes of 0 bytes.
  654          */
  655         if (uio->uio_resid == 0)
  656                 return (0);
  657 
  658         /*
  659          * The caller is supposed to ensure that
  660          * uio->uio_offset >= 0 and uio->uio_resid >= 0.
  661          *
  662          * If they've exceeded their filesize limit, tell them about it.
  663          */
  664         error = vn_rlimit_fsizex(vp, uio, MSDOSFS_FILESIZE_MAX, &r,
  665             uio->uio_td);
  666         if (error != 0) {
  667                 vn_rlimit_fsizex_res(uio, r);
  668                 return (error);
  669         }
  670 
  671         /*
  672          * If the offset we are starting the write at is beyond the end of
  673          * the file, then they've done a seek.  Unix filesystems allow
  674          * files with holes in them, DOS doesn't so we must fill the hole
  675          * with zeroed blocks.
  676          */
  677         if (uio->uio_offset > dep->de_FileSize) {
  678                 error = deextend(dep, uio->uio_offset, cred);
  679                 if (error != 0) {
  680                         vn_rlimit_fsizex_res(uio, r);
  681                         return (error);
  682                 }
  683         }
  684 
  685         /*
  686          * Remember some values in case the write fails.
  687          */
  688         resid = uio->uio_resid;
  689         osize = dep->de_FileSize;
  690 
  691         /*
  692          * If we write beyond the end of the file, extend it to its ultimate
  693          * size ahead of the time to hopefully get a contiguous area.
  694          */
  695         if (uio->uio_offset + resid > osize) {
  696                 count = de_clcount(pmp, uio->uio_offset + resid) -
  697                         de_clcount(pmp, osize);
  698                 error = extendfile(dep, count, NULL, NULL, 0);
  699                 if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
  700                         goto errexit;
  701                 lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
  702         } else
  703                 lastcn = de_clcount(pmp, osize) - 1;
  704 
  705         seqcount = ioflag >> IO_SEQSHIFT;
  706         do {
  707                 if (de_cluster(pmp, uio->uio_offset) > lastcn) {
  708                         error = ENOSPC;
  709                         break;
  710                 }
  711 
  712                 croffset = uio->uio_offset & pmp->pm_crbomask;
  713                 n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
  714                 if (uio->uio_offset + n > dep->de_FileSize) {
  715                         dep->de_FileSize = uio->uio_offset + n;
  716                         /* The object size needs to be set before buffer is allocated */
  717                         vnode_pager_setsize(vp, dep->de_FileSize);
  718                 }
  719 
  720                 bn = de_cluster(pmp, uio->uio_offset);
  721                 if ((uio->uio_offset & pmp->pm_crbomask) == 0
  722                     && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
  723                         > de_cluster(pmp, uio->uio_offset)
  724                         || uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
  725                         /*
  726                          * If either the whole cluster gets written,
  727                          * or we write the cluster from its start beyond EOF,
  728                          * then no need to read data from disk.
  729                          */
  730                         bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
  731                         /*
  732                          * This call to vfs_bio_clrbuf() ensures that
  733                          * even if vn_io_fault_uiomove() below faults,
  734                          * garbage from the newly instantiated buffer
  735                          * is not exposed to the userspace via mmap().
  736                          */
  737                         vfs_bio_clrbuf(bp);
  738                         /*
  739                          * Do the bmap now, since pcbmap needs buffers
  740                          * for the FAT table. (see msdosfs_strategy)
  741                          */
  742                         if (bp->b_blkno == bp->b_lblkno) {
  743                                 error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
  744                                 if (error)
  745                                         bp->b_blkno = -1;
  746                                 else
  747                                         bp->b_blkno = bn;
  748                         }
  749                         if (bp->b_blkno == -1) {
  750                                 brelse(bp);
  751                                 if (!error)
  752                                         error = EIO;            /* XXX */
  753                                 break;
  754                         }
  755                 } else {
  756                         /*
  757                          * The block we need to write into exists, so read it in.
  758                          */
  759                         error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
  760                         if (error) {
  761                                 break;
  762                         }
  763                 }
  764 
  765                 /*
  766                  * Should these vnode_pager_* functions be done on dir
  767                  * files?
  768                  */
  769 
  770                 /*
  771                  * Copy the data from user space into the buf header.
  772                  */
  773                 error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio);
  774                 if (error) {
  775                         brelse(bp);
  776                         break;
  777                 }
  778 
  779                 /* Prepare for clustered writes in some else clauses. */
  780                 if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
  781                         bp->b_flags |= B_CLUSTEROK;
  782 
  783                 /*
  784                  * If IO_SYNC, then each buffer is written synchronously.
  785                  * Otherwise, if we have a severe page deficiency then
  786                  * write the buffer asynchronously.  Otherwise, if on a
  787                  * cluster boundary then write the buffer asynchronously,
  788                  * combining it with contiguous clusters if permitted and
  789                  * possible, since we don't expect more writes into this
  790                  * buffer soon.  Otherwise, do a delayed write because we
  791                  * expect more writes into this buffer soon.
  792                  */
  793                 if (ioflag & IO_SYNC)
  794                         (void)bwrite(bp);
  795                 else if (vm_page_count_severe() || buf_dirty_count_severe())
  796                         bawrite(bp);
  797                 else if (n + croffset == pmp->pm_bpcluster) {
  798                         if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
  799                                 cluster_write(vp, &dep->de_clusterw, bp,
  800                                     dep->de_FileSize, seqcount, 0);
  801                         else
  802                                 bawrite(bp);
  803                 } else
  804                         bdwrite(bp);
  805                 dep->de_flag |= DE_UPDATE;
  806         } while (error == 0 && uio->uio_resid > 0);
  807 
  808         /*
  809          * If the write failed and they want us to, truncate the file back
  810          * to the size it was before the write was attempted.
  811          */
  812 errexit:
  813         if (error) {
  814                 if (ioflag & IO_UNIT) {
  815                         detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
  816                         uio->uio_offset -= resid - uio->uio_resid;
  817                         uio->uio_resid = resid;
  818                 } else {
  819                         detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
  820                         if (uio->uio_resid != resid)
  821                                 error = 0;
  822                 }
  823         } else if (ioflag & IO_SYNC)
  824                 error = deupdat(dep, 1);
  825         vn_rlimit_fsizex_res(uio, r);
  826         return (error);
  827 }
  828 
  829 /*
  830  * Flush the blocks of a file to disk.
  831  */
  832 static int
  833 msdosfs_fsync(struct vop_fsync_args *ap)
  834 {
  835         struct vnode *devvp;
  836         int allerror, error;
  837 
  838         vop_stdfsync(ap);
  839 
  840         /*
  841         * If the syncing request comes from fsync(2), sync the entire
  842         * FAT and any other metadata that happens to be on devvp.  We
  843         * need this mainly for the FAT.  We write the FAT sloppily, and
  844         * syncing it all now is the best we can easily do to get all
  845         * directory entries associated with the file (not just the file)
  846         * fully synced.  The other metadata includes critical metadata
  847         * for all directory entries, but only in the MNT_ASYNC case.  We
  848         * will soon sync all metadata in the file's directory entry.
  849         * Non-critical metadata for associated directory entries only
  850         * gets synced accidentally, as in most file systems.
  851         */
  852         if (ap->a_waitfor != MNT_NOWAIT) {
  853                 devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
  854                 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
  855                 allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
  856                 VOP_UNLOCK(devvp);
  857         } else
  858                 allerror = 0;
  859 
  860         error = deupdat(VTODE(ap->a_vp), ap->a_waitfor != MNT_NOWAIT);
  861         if (allerror == 0)
  862                 allerror = error;
  863         return (allerror);
  864 }
  865 
  866 static int
  867 msdosfs_remove(struct vop_remove_args *ap)
  868 {
  869         struct denode *dep = VTODE(ap->a_vp);
  870         struct denode *ddep = VTODE(ap->a_dvp);
  871         int error;
  872 
  873         if (ap->a_vp->v_type == VDIR)
  874                 error = EPERM;
  875         else
  876                 error = removede(ddep, dep);
  877 #ifdef MSDOSFS_DEBUG
  878         printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
  879 #endif
  880         return (error);
  881 }
  882 
  883 /*
  884  * DOS filesystems don't know what links are.
  885  */
  886 static int
  887 msdosfs_link(struct vop_link_args *ap)
  888 {
  889         return (EOPNOTSUPP);
  890 }
  891 
  892 /*
  893  * Renames on files require moving the denode to a new hash queue since the
  894  * denode's location is used to compute which hash queue to put the file
  895  * in. Unless it is a rename in place.  For example "mv a b".
  896  *
  897  * What follows is the basic algorithm:
  898  *
  899  * if (file move) {
  900  *      if (dest file exists) {
  901  *              remove dest file
  902  *      }
  903  *      if (dest and src in same directory) {
  904  *              rewrite name in existing directory slot
  905  *      } else {
  906  *              write new entry in dest directory
  907  *              update offset and dirclust in denode
  908  *              move denode to new hash chain
  909  *              clear old directory entry
  910  *      }
  911  * } else {
  912  *      directory move
  913  *      if (dest directory exists) {
  914  *              if (dest is not empty) {
  915  *                      return ENOTEMPTY
  916  *              }
  917  *              remove dest directory
  918  *      }
  919  *      if (dest and src in same directory) {
  920  *              rewrite name in existing entry
  921  *      } else {
  922  *              be sure dest is not a child of src directory
  923  *              write entry in dest directory
  924  *              update "." and ".." in moved directory
  925  *              clear old directory entry for moved directory
  926  *      }
  927  * }
  928  *
  929  * On entry:
  930  *      source's parent directory is unlocked
  931  *      source file or directory is unlocked
  932  *      destination's parent directory is locked
  933  *      destination file or directory is locked if it exists
  934  *
  935  * On exit:
  936  *      all denodes should be released
  937  */
  938 static int
  939 msdosfs_rename(struct vop_rename_args *ap)
  940 {
  941         struct vnode *fdvp, *fvp, *tdvp, *tvp, *vp;
  942         struct componentname *fcnp, *tcnp;
  943         struct denode *fdip, *fip, *tdip, *tip, *nip;
  944         u_char toname[12], oldname[11];
  945         u_long to_diroffset;
  946         bool checkpath_locked, doingdirectory, newparent;
  947         int error;
  948         u_long cn, pcl, blkoff;
  949         daddr_t bn, wait_scn, scn;
  950         struct msdosfsmount *pmp;
  951         struct direntry *dotdotp;
  952         struct buf *bp;
  953 
  954         tdvp = ap->a_tdvp;
  955         fvp = ap->a_fvp;
  956         fdvp = ap->a_fdvp;
  957         tvp = ap->a_tvp;
  958         tcnp = ap->a_tcnp;
  959         fcnp = ap->a_fcnp;
  960         pmp = VFSTOMSDOSFS(fdvp->v_mount);
  961 
  962         /*
  963          * Check for cross-device rename.
  964          */
  965         if (fvp->v_mount != tdvp->v_mount ||
  966             (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
  967                 error = EXDEV;
  968                 goto abortit;
  969         }
  970 
  971         /*
  972          * If source and dest are the same, do nothing.
  973          */
  974         if (tvp == fvp) {
  975                 error = 0;
  976                 goto abortit;
  977         }
  978 
  979         /*
  980          * When the target exists, both the directory
  981          * and target vnodes are passed locked.
  982          */
  983         VOP_UNLOCK(tdvp);
  984         if (tvp != NULL && tvp != tdvp)
  985                 VOP_UNLOCK(tvp);
  986 
  987         checkpath_locked = false;
  988 
  989 relock:
  990         doingdirectory = newparent = false;
  991 
  992         error = vn_lock(fdvp, LK_EXCLUSIVE);
  993         if (error != 0)
  994                 goto releout;
  995         if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
  996                 VOP_UNLOCK(fdvp);
  997                 error = vn_lock(tdvp, LK_EXCLUSIVE);
  998                 if (error != 0)
  999                         goto releout;
 1000                 VOP_UNLOCK(tdvp);
 1001                 goto relock;
 1002         }
 1003 
 1004         error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff);
 1005         if (error != 0) {
 1006                 VOP_UNLOCK(fdvp);
 1007                 VOP_UNLOCK(tdvp);
 1008                 goto releout;
 1009         }
 1010         error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT, &nip);
 1011         if (error != 0) {
 1012                 VOP_UNLOCK(fdvp);
 1013                 VOP_UNLOCK(tdvp);
 1014                 if (error != EBUSY)
 1015                         goto releout;
 1016                 error = deget(pmp, scn, blkoff, LK_EXCLUSIVE, &nip);
 1017                 if (error != 0)
 1018                         goto releout;
 1019                 vp = fvp;
 1020                 fvp = DETOV(nip);
 1021                 VOP_UNLOCK(fvp);
 1022                 vrele(vp);
 1023                 goto relock;
 1024         }
 1025         vrele(fvp);
 1026         fvp = DETOV(nip);
 1027 
 1028         error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
 1029         if (error != 0 && error != EJUSTRETURN) {
 1030                 VOP_UNLOCK(fdvp);
 1031                 VOP_UNLOCK(tdvp);
 1032                 VOP_UNLOCK(fvp);
 1033                 goto releout;
 1034         }
 1035         if (error == EJUSTRETURN && tvp != NULL) {
 1036                 vrele(tvp);
 1037                 tvp = NULL;
 1038         }
 1039         if (error == 0) {
 1040                 nip = NULL;
 1041                 error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT,
 1042                     &nip);
 1043                 if (tvp != NULL) {
 1044                         vrele(tvp);
 1045                         tvp = NULL;
 1046                 }
 1047                 if (error != 0) {
 1048                         VOP_UNLOCK(fdvp);
 1049                         VOP_UNLOCK(tdvp);
 1050                         VOP_UNLOCK(fvp);
 1051                         if (error != EBUSY)
 1052                                 goto releout;
 1053                         error = deget(pmp, scn, blkoff, LK_EXCLUSIVE,
 1054                             &nip);
 1055                         if (error != 0)
 1056                                 goto releout;
 1057                         vput(DETOV(nip));
 1058                         goto relock;
 1059                 }
 1060                 tvp = DETOV(nip);
 1061         }
 1062 
 1063         fdip = VTODE(fdvp);
 1064         fip = VTODE(fvp);
 1065         tdip = VTODE(tdvp);
 1066         tip = tvp != NULL ? VTODE(tvp) : NULL;
 1067 
 1068         /*
 1069          * Remember direntry place to use for destination
 1070          */
 1071         to_diroffset = tdip->de_fndoffset;
 1072 
 1073         /*
 1074          * Be sure we are not renaming ".", "..", or an alias of ".". This
 1075          * leads to a crippled directory tree.  It's pretty tough to do a
 1076          * "ls" or "pwd" with the "." directory entry missing, and "cd .."
 1077          * doesn't work if the ".." entry is missing.
 1078          */
 1079         if ((fip->de_Attributes & ATTR_DIRECTORY) != 0) {
 1080                 /*
 1081                  * Avoid ".", "..", and aliases of "." for obvious reasons.
 1082                  */
 1083                 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
 1084                     fdip == fip ||
 1085                     (fcnp->cn_flags & ISDOTDOT) != 0 ||
 1086                     (tcnp->cn_flags & ISDOTDOT) != 0) {
 1087                         error = EINVAL;
 1088                         goto unlock;
 1089                 }
 1090                 doingdirectory = true;
 1091         }
 1092 
 1093         /*
 1094          * If ".." must be changed (ie the directory gets a new
 1095          * parent) then the source directory must not be in the
 1096          * directory hierarchy above the target, as this would
 1097          * orphan everything below the source directory. Also
 1098          * the user must have write permission in the source so
 1099          * as to be able to change "..". We must repeat the call
 1100          * to namei, as the parent directory is unlocked by the
 1101          * call to doscheckpath().
 1102          */
 1103         error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
 1104         if (fdip->de_StartCluster != tdip->de_StartCluster)
 1105                 newparent = true;
 1106         if (doingdirectory && newparent) {
 1107                 if (error != 0) /* write access check above */
 1108                         goto unlock;
 1109                 lockmgr(&pmp->pm_checkpath_lock, LK_EXCLUSIVE, NULL);
 1110                 checkpath_locked = true;
 1111                 error = doscheckpath(fip, tdip, &wait_scn);
 1112                 if (wait_scn != 0) {
 1113                         lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
 1114                         checkpath_locked = false;
 1115                         VOP_UNLOCK(fdvp);
 1116                         VOP_UNLOCK(tdvp);
 1117                         VOP_UNLOCK(fvp);
 1118                         if (tvp != NULL && tvp != tdvp)
 1119                                 VOP_UNLOCK(tvp);
 1120                         error = deget(pmp, wait_scn, 0, LK_EXCLUSIVE,
 1121                             &nip);
 1122                         if (error == 0) {
 1123                                 vput(DETOV(nip));
 1124                                 goto relock;
 1125                         }
 1126                 }
 1127                 if (error != 0)
 1128                         goto unlock;
 1129         }
 1130 
 1131         if (tip != NULL) {
 1132                 /*
 1133                  * Target must be empty if a directory and have no links
 1134                  * to it. Also, ensure source and target are compatible
 1135                  * (both directories, or both not directories).
 1136                  */
 1137                 if ((tip->de_Attributes & ATTR_DIRECTORY) != 0) {
 1138                         if (!dosdirempty(tip)) {
 1139                                 error = ENOTEMPTY;
 1140                                 goto unlock;
 1141                         }
 1142                         if (!doingdirectory) {
 1143                                 error = ENOTDIR;
 1144                                 goto unlock;
 1145                         }
 1146                         cache_purge(tdvp);
 1147                 } else if (doingdirectory) {
 1148                         error = EISDIR;
 1149                         goto unlock;
 1150                 }
 1151                 error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
 1152                 MPASS(error == 0);
 1153                 error = removede(tdip, tip);
 1154                 if (error != 0)
 1155                         goto unlock;
 1156                 vput(tvp);
 1157                 tvp = NULL;
 1158                 tip = NULL;
 1159         }
 1160 
 1161         /*
 1162          * Convert the filename in tcnp into a dos filename. We copy this
 1163          * into the denode and directory entry for the destination
 1164          * file/directory.
 1165          */
 1166         error = uniqdosname(tdip, tcnp, toname);
 1167         if (error != 0)
 1168                 goto unlock;
 1169 
 1170         /*
 1171          * First write a new entry in the destination
 1172          * directory and mark the entry in the source directory
 1173          * as deleted.  Then move the denode to the correct hash
 1174          * chain for its new location in the filesystem.  And, if
 1175          * we moved a directory, then update its .. entry to point
 1176          * to the new parent directory.
 1177          */
 1178         memcpy(oldname, fip->de_Name, 11);
 1179         memcpy(fip->de_Name, toname, 11);       /* update denode */
 1180         error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
 1181         MPASS(error == EJUSTRETURN);
 1182         error = createde(fip, tdip, NULL, tcnp);
 1183         if (error != 0) {
 1184                 memcpy(fip->de_Name, oldname, 11);
 1185                 goto unlock;
 1186         }
 1187 
 1188         /*
 1189          * If fip is for a directory, then its name should always
 1190          * be "." since it is for the directory entry in the
 1191          * directory itself (msdosfs_lookup() always translates
 1192          * to the "." entry so as to get a unique denode, except
 1193          * for the root directory there are different
 1194          * complications).  However, we just corrupted its name
 1195          * to pass the correct name to createde().  Undo this.
 1196          */
 1197         if ((fip->de_Attributes & ATTR_DIRECTORY) != 0)
 1198                 memcpy(fip->de_Name, oldname, 11);
 1199         fip->de_refcnt++;
 1200         error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff);
 1201         MPASS(error == 0);
 1202         error = removede(fdip, fip);
 1203         if (error != 0) {
 1204                 /* XXX should downgrade to ro here, fs is corrupt */
 1205                 goto unlock;
 1206         }
 1207         if (!doingdirectory) {
 1208                 error = pcbmap(tdip, de_cluster(pmp, to_diroffset), 0,
 1209                     &fip->de_dirclust, 0);
 1210                 if (error != 0) {
 1211                         /*
 1212                          * XXX should downgrade to ro here,
 1213                          * fs is corrupt
 1214                          */
 1215                         goto unlock;
 1216                 }
 1217                 if (fip->de_dirclust == MSDOSFSROOT)
 1218                         fip->de_diroffset = to_diroffset;
 1219                 else
 1220                         fip->de_diroffset = to_diroffset & pmp->pm_crbomask;
 1221         }
 1222         reinsert(fip);
 1223 
 1224         /*
 1225          * If we moved a directory to a new parent directory, then we must
 1226          * fixup the ".." entry in the moved directory.
 1227          */
 1228         if (doingdirectory && newparent) {
 1229                 cn = fip->de_StartCluster;
 1230                 if (cn == MSDOSFSROOT) {
 1231                         /* this should never happen */
 1232                         panic("msdosfs_rename(): updating .. in root directory?");
 1233                 } else
 1234                         bn = cntobn(pmp, cn);
 1235                 error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
 1236                     NOCRED, &bp);
 1237                 if (error != 0) {
 1238                         /* XXX should downgrade to ro here, fs is corrupt */
 1239                         goto unlock;
 1240                 }
 1241                 dotdotp = (struct direntry *)bp->b_data + 1;
 1242                 pcl = tdip->de_StartCluster;
 1243                 if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
 1244                         pcl = MSDOSFSROOT;
 1245                 putushort(dotdotp->deStartCluster, pcl);
 1246                 if (FAT32(pmp))
 1247                         putushort(dotdotp->deHighClust, pcl >> 16);
 1248                 if (DOINGASYNC(fvp))
 1249                         bdwrite(bp);
 1250                 else if ((error = bwrite(bp)) != 0) {
 1251                         /* XXX should downgrade to ro here, fs is corrupt */
 1252                         goto unlock;
 1253                 }
 1254         }
 1255 
 1256         /*
 1257          * The msdosfs lookup is case insensitive. Several aliases may
 1258          * be inserted for a single directory entry. As a consequnce,
 1259          * name cache purge done by lookup for fvp when DELETE op for
 1260          * namei is specified, might be not enough to expunge all
 1261          * namecache entries that were installed for this direntry.
 1262          */
 1263         cache_purge(fvp);
 1264 
 1265 unlock:
 1266         if (checkpath_locked)
 1267                 lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
 1268         vput(fdvp);
 1269         vput(fvp);
 1270         if (tvp != NULL) {
 1271                 if (tvp != tdvp)
 1272                         vput(tvp);
 1273                 else
 1274                         vrele(tvp);
 1275         }
 1276         vput(tdvp);
 1277         return (error);
 1278 releout:
 1279         MPASS(!checkpath_locked);
 1280         vrele(tdvp);
 1281         if (tvp != NULL)
 1282                 vrele(tvp);
 1283         vrele(fdvp);
 1284         vrele(fvp);
 1285         return (error);
 1286 abortit:
 1287         if (tdvp == tvp)
 1288                 vrele(tdvp);
 1289         else
 1290                 vput(tdvp);
 1291         if (tvp != NULL)
 1292                 vput(tvp);
 1293         vrele(fdvp);
 1294         vrele(fvp);
 1295         return (error);
 1296 }
 1297 
 1298 static struct {
 1299         struct direntry dot;
 1300         struct direntry dotdot;
 1301 } dosdirtemplate = {
 1302         {       ".          ",                          /* the . entry */
 1303                 ATTR_DIRECTORY,                         /* file attribute */
 1304                 0,                                      /* reserved */
 1305                 0, { 0, 0 }, { 0, 0 },                  /* create time & date */
 1306                 { 0, 0 },                               /* access date */
 1307                 { 0, 0 },                               /* high bits of start cluster */
 1308                 { 210, 4 }, { 210, 4 },                 /* modify time & date */
 1309                 { 0, 0 },                               /* startcluster */
 1310                 { 0, 0, 0, 0 }                          /* filesize */
 1311         },
 1312         {       "..         ",                          /* the .. entry */
 1313                 ATTR_DIRECTORY,                         /* file attribute */
 1314                 0,                                      /* reserved */
 1315                 0, { 0, 0 }, { 0, 0 },                  /* create time & date */
 1316                 { 0, 0 },                               /* access date */
 1317                 { 0, 0 },                               /* high bits of start cluster */
 1318                 { 210, 4 }, { 210, 4 },                 /* modify time & date */
 1319                 { 0, 0 },                               /* startcluster */
 1320                 { 0, 0, 0, 0 }                          /* filesize */
 1321         }
 1322 };
 1323 
 1324 static int
 1325 msdosfs_mkdir(struct vop_mkdir_args *ap)
 1326 {
 1327         struct componentname *cnp = ap->a_cnp;
 1328         struct denode *dep;
 1329         struct denode *pdep = VTODE(ap->a_dvp);
 1330         struct direntry *denp;
 1331         struct msdosfsmount *pmp = pdep->de_pmp;
 1332         struct buf *bp;
 1333         u_long newcluster, pcl;
 1334         int bn;
 1335         int error;
 1336         struct denode ndirent;
 1337         struct timespec ts;
 1338 
 1339         /*
 1340          * If this is the root directory and there is no space left we
 1341          * can't do anything.  This is because the root directory can not
 1342          * change size.
 1343          */
 1344         if (pdep->de_StartCluster == MSDOSFSROOT
 1345             && pdep->de_fndoffset >= pdep->de_FileSize) {
 1346                 error = ENOSPC;
 1347                 goto bad2;
 1348         }
 1349 
 1350         /*
 1351          * Allocate a cluster to hold the about to be created directory.
 1352          */
 1353         error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
 1354         if (error)
 1355                 goto bad2;
 1356 
 1357         memset(&ndirent, 0, sizeof(ndirent));
 1358         ndirent.de_pmp = pmp;
 1359         ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
 1360         vfs_timestamp(&ts);
 1361         DETIMES(&ndirent, &ts, &ts, &ts);
 1362 
 1363         /*
 1364          * Now fill the cluster with the "." and ".." entries. And write
 1365          * the cluster to disk.  This way it is there for the parent
 1366          * directory to be pointing at if there were a crash.
 1367          */
 1368         bn = cntobn(pmp, newcluster);
 1369         /* always succeeds */
 1370         bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
 1371         memset(bp->b_data, 0, pmp->pm_bpcluster);
 1372         memcpy(bp->b_data, &dosdirtemplate, sizeof dosdirtemplate);
 1373         denp = (struct direntry *)bp->b_data;
 1374         putushort(denp[0].deStartCluster, newcluster);
 1375         putushort(denp[0].deCDate, ndirent.de_CDate);
 1376         putushort(denp[0].deCTime, ndirent.de_CTime);
 1377         denp[0].deCHundredth = ndirent.de_CHun;
 1378         putushort(denp[0].deADate, ndirent.de_ADate);
 1379         putushort(denp[0].deMDate, ndirent.de_MDate);
 1380         putushort(denp[0].deMTime, ndirent.de_MTime);
 1381         pcl = pdep->de_StartCluster;
 1382         /*
 1383          * Although the root directory has a non-magic starting cluster
 1384          * number for FAT32, chkdsk and fsck_msdosfs still require
 1385          * references to it in dotdot entries to be magic.
 1386          */
 1387         if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
 1388                 pcl = MSDOSFSROOT;
 1389         putushort(denp[1].deStartCluster, pcl);
 1390         putushort(denp[1].deCDate, ndirent.de_CDate);
 1391         putushort(denp[1].deCTime, ndirent.de_CTime);
 1392         denp[1].deCHundredth = ndirent.de_CHun;
 1393         putushort(denp[1].deADate, ndirent.de_ADate);
 1394         putushort(denp[1].deMDate, ndirent.de_MDate);
 1395         putushort(denp[1].deMTime, ndirent.de_MTime);
 1396         if (FAT32(pmp)) {
 1397                 putushort(denp[0].deHighClust, newcluster >> 16);
 1398                 putushort(denp[1].deHighClust, pcl >> 16);
 1399         }
 1400 
 1401         if (DOINGASYNC(ap->a_dvp))
 1402                 bdwrite(bp);
 1403         else if ((error = bwrite(bp)) != 0)
 1404                 goto bad;
 1405 
 1406         /*
 1407          * Now build up a directory entry pointing to the newly allocated
 1408          * cluster.  This will be written to an empty slot in the parent
 1409          * directory.
 1410          */
 1411         error = uniqdosname(pdep, cnp, ndirent.de_Name);
 1412         if (error)
 1413                 goto bad;
 1414 
 1415         ndirent.de_Attributes = ATTR_DIRECTORY;
 1416         ndirent.de_LowerCase = 0;
 1417         ndirent.de_StartCluster = newcluster;
 1418         ndirent.de_FileSize = 0;
 1419         error = createde(&ndirent, pdep, &dep, cnp);
 1420         if (error)
 1421                 goto bad;
 1422         *ap->a_vpp = DETOV(dep);
 1423         return (0);
 1424 
 1425 bad:
 1426         clusterfree(pmp, newcluster);
 1427 bad2:
 1428         return (error);
 1429 }
 1430 
 1431 static int
 1432 msdosfs_rmdir(struct vop_rmdir_args *ap)
 1433 {
 1434         struct vnode *vp = ap->a_vp;
 1435         struct vnode *dvp = ap->a_dvp;
 1436         struct componentname *cnp = ap->a_cnp;
 1437         struct denode *ip, *dp;
 1438         int error;
 1439 
 1440         ip = VTODE(vp);
 1441         dp = VTODE(dvp);
 1442 
 1443         /*
 1444          * Verify the directory is empty (and valid).
 1445          * (Rmdir ".." won't be valid since
 1446          *  ".." will contain a reference to
 1447          *  the current directory and thus be
 1448          *  non-empty.)
 1449          */
 1450         error = 0;
 1451         if (!dosdirempty(ip)) {
 1452                 error = ENOTEMPTY;
 1453                 goto out;
 1454         }
 1455         /*
 1456          * Delete the entry from the directory.  For dos filesystems this
 1457          * gets rid of the directory entry on disk, the in memory copy
 1458          * still exists but the de_refcnt is <= 0.  This prevents it from
 1459          * being found by deget().  When the vput() on dep is done we give
 1460          * up access and eventually msdosfs_reclaim() will be called which
 1461          * will remove it from the denode cache.
 1462          */
 1463         error = removede(dp, ip);
 1464         if (error)
 1465                 goto out;
 1466         /*
 1467          * This is where we decrement the link count in the parent
 1468          * directory.  Since dos filesystems don't do this we just purge
 1469          * the name cache.
 1470          */
 1471         cache_purge(dvp);
 1472         /*
 1473          * Truncate the directory that is being deleted.
 1474          */
 1475         error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
 1476         cache_purge(vp);
 1477 
 1478 out:
 1479         return (error);
 1480 }
 1481 
 1482 /*
 1483  * DOS filesystems don't know what symlinks are.
 1484  */
 1485 static int
 1486 msdosfs_symlink(struct vop_symlink_args *ap)
 1487 {
 1488         return (EOPNOTSUPP);
 1489 }
 1490 
 1491 static int
 1492 msdosfs_readdir(struct vop_readdir_args *ap)
 1493 {
 1494         struct mbnambuf nb;
 1495         int error = 0;
 1496         int diff;
 1497         long n;
 1498         int blsize;
 1499         long on;
 1500         u_long cn;
 1501         u_long dirsperblk;
 1502         long bias = 0;
 1503         daddr_t bn, lbn;
 1504         struct buf *bp;
 1505         struct denode *dep = VTODE(ap->a_vp);
 1506         struct msdosfsmount *pmp = dep->de_pmp;
 1507         struct direntry *dentp;
 1508         struct dirent dirbuf;
 1509         struct uio *uio = ap->a_uio;
 1510         uint64_t *cookies = NULL;
 1511         int ncookies = 0;
 1512         off_t offset, off;
 1513         int chksum = -1;
 1514 
 1515 #ifdef MSDOSFS_DEBUG
 1516         printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
 1517             ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
 1518 #endif
 1519 
 1520         /*
 1521          * msdosfs_readdir() won't operate properly on regular files since
 1522          * it does i/o only with the filesystem vnode, and hence can
 1523          * retrieve the wrong block from the buffer cache for a plain file.
 1524          * So, fail attempts to readdir() on a plain file.
 1525          */
 1526         if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
 1527                 return (ENOTDIR);
 1528 
 1529         /*
 1530          * To be safe, initialize dirbuf
 1531          */
 1532         memset(dirbuf.d_name, 0, sizeof(dirbuf.d_name));
 1533 
 1534         /*
 1535          * If the user buffer is smaller than the size of one dos directory
 1536          * entry or the file offset is not a multiple of the size of a
 1537          * directory entry, then we fail the read.
 1538          */
 1539         off = offset = uio->uio_offset;
 1540         if (uio->uio_resid < sizeof(struct direntry) ||
 1541             (offset & (sizeof(struct direntry) - 1)))
 1542                 return (EINVAL);
 1543 
 1544         if (ap->a_ncookies) {
 1545                 ncookies = uio->uio_resid / 16;
 1546                 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP,
 1547                        M_WAITOK);
 1548                 *ap->a_cookies = cookies;
 1549                 *ap->a_ncookies = ncookies;
 1550         }
 1551 
 1552         dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
 1553 
 1554         /*
 1555          * If they are reading from the root directory then, we simulate
 1556          * the . and .. entries since these don't exist in the root
 1557          * directory.  We also set the offset bias to make up for having to
 1558          * simulate these entries. By this I mean that at file offset 64 we
 1559          * read the first entry in the root directory that lives on disk.
 1560          */
 1561         if (dep->de_StartCluster == MSDOSFSROOT
 1562             || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
 1563 #if 0
 1564                 printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
 1565                     offset);
 1566 #endif
 1567                 bias = 2 * sizeof(struct direntry);
 1568                 if (offset < bias) {
 1569                         for (n = (int)offset / sizeof(struct direntry);
 1570                              n < 2; n++) {
 1571                                 dirbuf.d_fileno = FAT32(pmp) ?
 1572                                     (uint64_t)cntobn(pmp, pmp->pm_rootdirblk) *
 1573                                     dirsperblk : 1;
 1574                                 dirbuf.d_type = DT_DIR;
 1575                                 switch (n) {
 1576                                 case 0:
 1577                                         dirbuf.d_namlen = 1;
 1578                                         dirbuf.d_name[0] = '.';
 1579                                         break;
 1580                                 case 1:
 1581                                         dirbuf.d_namlen = 2;
 1582                                         dirbuf.d_name[0] = '.';
 1583                                         dirbuf.d_name[1] = '.';
 1584                                         break;
 1585                                 }
 1586                                 dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
 1587                                 /* NOTE: d_off is the offset of the *next* entry. */
 1588                                 dirbuf.d_off = offset + sizeof(struct direntry);
 1589                                 dirent_terminate(&dirbuf);
 1590                                 if (uio->uio_resid < dirbuf.d_reclen)
 1591                                         goto out;
 1592                                 error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
 1593                                 if (error)
 1594                                         goto out;
 1595                                 offset += sizeof(struct direntry);
 1596                                 off = offset;
 1597                                 if (cookies) {
 1598                                         *cookies++ = offset;
 1599                                         if (--ncookies <= 0)
 1600                                                 goto out;
 1601                                 }
 1602                         }
 1603                 }
 1604         }
 1605 
 1606         mbnambuf_init(&nb);
 1607         off = offset;
 1608         while (uio->uio_resid > 0) {
 1609                 lbn = de_cluster(pmp, offset - bias);
 1610                 on = (offset - bias) & pmp->pm_crbomask;
 1611                 n = min(pmp->pm_bpcluster - on, uio->uio_resid);
 1612                 diff = dep->de_FileSize - (offset - bias);
 1613                 if (diff <= 0)
 1614                         break;
 1615                 n = min(n, diff);
 1616                 error = pcbmap(dep, lbn, &bn, &cn, &blsize);
 1617                 if (error)
 1618                         break;
 1619                 error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
 1620                 if (error) {
 1621                         return (error);
 1622                 }
 1623                 n = min(n, blsize - bp->b_resid);
 1624                 if (n == 0) {
 1625                         brelse(bp);
 1626                         return (EIO);
 1627                 }
 1628 
 1629                 /*
 1630                  * Convert from dos directory entries to fs-independent
 1631                  * directory entries.
 1632                  */
 1633                 for (dentp = (struct direntry *)(bp->b_data + on);
 1634                      (char *)dentp < bp->b_data + on + n;
 1635                      dentp++, offset += sizeof(struct direntry)) {
 1636 #if 0
 1637                         printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
 1638                             dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
 1639 #endif
 1640                         /*
 1641                          * If this is an unused entry, we can stop.
 1642                          */
 1643                         if (dentp->deName[0] == SLOT_EMPTY) {
 1644                                 brelse(bp);
 1645                                 goto out;
 1646                         }
 1647                         /*
 1648                          * Skip deleted entries.
 1649                          */
 1650                         if (dentp->deName[0] == SLOT_DELETED) {
 1651                                 chksum = -1;
 1652                                 mbnambuf_init(&nb);
 1653                                 continue;
 1654                         }
 1655 
 1656                         /*
 1657                          * Handle Win95 long directory entries
 1658                          */
 1659                         if (dentp->deAttributes == ATTR_WIN95) {
 1660                                 if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
 1661                                         continue;
 1662                                 chksum = win2unixfn(&nb,
 1663                                     (struct winentry *)dentp, chksum, pmp);
 1664                                 continue;
 1665                         }
 1666 
 1667                         /*
 1668                          * Skip volume labels
 1669                          */
 1670                         if (dentp->deAttributes & ATTR_VOLUME) {
 1671                                 chksum = -1;
 1672                                 mbnambuf_init(&nb);
 1673                                 continue;
 1674                         }
 1675                         /*
 1676                          * This computation of d_fileno must match
 1677                          * the computation of va_fileid in
 1678                          * msdosfs_getattr.
 1679                          */
 1680                         if (dentp->deAttributes & ATTR_DIRECTORY) {
 1681                                 cn = getushort(dentp->deStartCluster);
 1682                                 if (FAT32(pmp)) {
 1683                                         cn |= getushort(dentp->deHighClust) <<
 1684                                             16;
 1685                                         if (cn == MSDOSFSROOT)
 1686                                                 cn = pmp->pm_rootdirblk;
 1687                                 }
 1688                                 if (cn == MSDOSFSROOT && !FAT32(pmp))
 1689                                         dirbuf.d_fileno = 1;
 1690                                 else
 1691                                         dirbuf.d_fileno = cntobn(pmp, cn) *
 1692                                             dirsperblk;
 1693                                 dirbuf.d_type = DT_DIR;
 1694                         } else {
 1695                                 dirbuf.d_fileno = (uoff_t)offset /
 1696                                     sizeof(struct direntry);
 1697                                 dirbuf.d_type = DT_REG;
 1698                         }
 1699 
 1700                         if (chksum != winChksum(dentp->deName)) {
 1701                                 dirbuf.d_namlen = dos2unixfn(dentp->deName,
 1702                                     (u_char *)dirbuf.d_name,
 1703                                     dentp->deLowerCase |
 1704                                         ((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
 1705                                         (LCASE_BASE | LCASE_EXT) : 0),
 1706                                     pmp);
 1707                                 mbnambuf_init(&nb);
 1708                         } else
 1709                                 mbnambuf_flush(&nb, &dirbuf);
 1710                         chksum = -1;
 1711                         dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
 1712                         /* NOTE: d_off is the offset of the *next* entry. */
 1713                         dirbuf.d_off = offset + sizeof(struct direntry);
 1714                         dirent_terminate(&dirbuf);
 1715                         if (uio->uio_resid < dirbuf.d_reclen) {
 1716                                 brelse(bp);
 1717                                 goto out;
 1718                         }
 1719                         error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
 1720                         if (error) {
 1721                                 brelse(bp);
 1722                                 goto out;
 1723                         }
 1724                         if (cookies) {
 1725                                 *cookies++ = offset + sizeof(struct direntry);
 1726                                 if (--ncookies <= 0) {
 1727                                         brelse(bp);
 1728                                         goto out;
 1729                                 }
 1730                         }
 1731                         off = offset + sizeof(struct direntry);
 1732                 }
 1733                 brelse(bp);
 1734         }
 1735 out:
 1736         /* Subtract unused cookies */
 1737         if (ap->a_ncookies)
 1738                 *ap->a_ncookies -= ncookies;
 1739 
 1740         uio->uio_offset = off;
 1741 
 1742         /*
 1743          * Set the eofflag (NFS uses it)
 1744          */
 1745         if (ap->a_eofflag) {
 1746                 if (dep->de_FileSize - (offset - bias) <= 0)
 1747                         *ap->a_eofflag = 1;
 1748                 else
 1749                         *ap->a_eofflag = 0;
 1750         }
 1751         return (error);
 1752 }
 1753 
 1754 /*-
 1755  * a_vp   - pointer to the file's vnode
 1756  * a_bn   - logical block number within the file (cluster number for us)
 1757  * a_bop  - where to return the bufobj of the special file containing the fs
 1758  * a_bnp  - where to return the "physical" block number corresponding to a_bn
 1759  *          (relative to the special file; units are blocks of size DEV_BSIZE)
 1760  * a_runp - where to return the "run past" a_bn.  This is the count of logical
 1761  *          blocks whose physical blocks (together with a_bn's physical block)
 1762  *          are contiguous.
 1763  * a_runb - where to return the "run before" a_bn.
 1764  */
 1765 static int
 1766 msdosfs_bmap(struct vop_bmap_args *ap)
 1767 {
 1768         struct fatcache savefc;
 1769         struct denode *dep;
 1770         struct mount *mp;
 1771         struct msdosfsmount *pmp;
 1772         struct vnode *vp;
 1773         daddr_t runbn;
 1774         u_long cn;
 1775         int bnpercn, error, maxio, maxrun, run;
 1776 
 1777         vp = ap->a_vp;
 1778         dep = VTODE(vp);
 1779         pmp = dep->de_pmp;
 1780         if (ap->a_bop != NULL)
 1781                 *ap->a_bop = &pmp->pm_devvp->v_bufobj;
 1782         if (ap->a_bnp == NULL)
 1783                 return (0);
 1784         if (ap->a_runp != NULL)
 1785                 *ap->a_runp = 0;
 1786         if (ap->a_runb != NULL)
 1787                 *ap->a_runb = 0;
 1788         cn = ap->a_bn;
 1789         if (cn != ap->a_bn)
 1790                 return (EFBIG);
 1791         error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
 1792         if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
 1793                 return (error);
 1794 
 1795         /*
 1796          * Prepare to back out updates of the fatchain cache after the one
 1797          * for the first block done by pcbmap() above.  Without the backout,
 1798          * then whenever the caller doesn't do i/o to all of the blocks that
 1799          * we find, the single useful cache entry would be too far in advance
 1800          * of the actual i/o to work for the next sequential i/o.  Then the
 1801          * FAT would be searched from the beginning.  With the backout, the
 1802          * FAT is searched starting at most a few blocks early.  This wastes
 1803          * much less time.  Time is also wasted finding more blocks than the
 1804          * caller will do i/o to.  This is necessary because the runlength
 1805          * parameters are output-only.
 1806          */
 1807         savefc = dep->de_fc[FC_LASTMAP];
 1808 
 1809         mp = vp->v_mount;
 1810         maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
 1811         bnpercn = de_cn2bn(pmp, 1);
 1812         if (ap->a_runp != NULL) {
 1813                 maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
 1814                 for (run = 1; run <= maxrun; run++) {
 1815                         if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
 1816                             runbn != *ap->a_bnp + run * bnpercn)
 1817                                 break;
 1818                 }
 1819                 *ap->a_runp = run - 1;
 1820         }
 1821         if (ap->a_runb != NULL) {
 1822                 maxrun = ulmin(maxio - 1, cn);
 1823                 for (run = 1; run < maxrun; run++) {
 1824                         if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
 1825                             runbn != *ap->a_bnp - run * bnpercn)
 1826                                 break;
 1827                 }
 1828                 *ap->a_runb = run - 1;
 1829         }
 1830         dep->de_fc[FC_LASTMAP] = savefc;
 1831         return (0);
 1832 }
 1833 
 1834 SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 1835     "msdos filesystem");
 1836 static int use_buf_pager = 1;
 1837 SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN,
 1838     &use_buf_pager, 0,
 1839     "Use buffer pager instead of bmap");
 1840 
 1841 static daddr_t
 1842 msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
 1843 {
 1844 
 1845         return (de_cluster(VTODE(vp)->de_pmp, off));
 1846 }
 1847 
 1848 static int
 1849 msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz)
 1850 {
 1851 
 1852         *sz = VTODE(vp)->de_pmp->pm_bpcluster;
 1853         return (0);
 1854 }
 1855 
 1856 static int
 1857 msdosfs_getpages(struct vop_getpages_args *ap)
 1858 {
 1859 
 1860         if (use_buf_pager)
 1861                 return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count,
 1862                     ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno,
 1863                     msdosfs_gbp_getblksz));
 1864         return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
 1865             ap->a_rbehind, ap->a_rahead, NULL, NULL));
 1866 }
 1867 
 1868 static int
 1869 msdosfs_strategy(struct vop_strategy_args *ap)
 1870 {
 1871         struct buf *bp = ap->a_bp;
 1872         struct denode *dep = VTODE(ap->a_vp);
 1873         struct bufobj *bo;
 1874         int error = 0;
 1875         daddr_t blkno;
 1876 
 1877         /*
 1878          * If we don't already know the filesystem relative block number
 1879          * then get it using pcbmap().  If pcbmap() returns the block
 1880          * number as -1 then we've got a hole in the file.  DOS filesystems
 1881          * don't allow files with holes, so we shouldn't ever see this.
 1882          */
 1883         if (bp->b_blkno == bp->b_lblkno) {
 1884                 error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
 1885                 bp->b_blkno = blkno;
 1886                 if (error) {
 1887                         bp->b_error = error;
 1888                         bp->b_ioflags |= BIO_ERROR;
 1889                         bufdone(bp);
 1890                         return (0);
 1891                 }
 1892                 if ((long)bp->b_blkno == -1)
 1893                         vfs_bio_clrbuf(bp);
 1894         }
 1895         if (bp->b_blkno == -1) {
 1896                 bufdone(bp);
 1897                 return (0);
 1898         }
 1899         /*
 1900          * Read/write the block from/to the disk that contains the desired
 1901          * file block.
 1902          */
 1903         bp->b_iooffset = dbtob(bp->b_blkno);
 1904         bo = dep->de_pmp->pm_bo;
 1905         BO_STRATEGY(bo, bp);
 1906         return (0);
 1907 }
 1908 
 1909 static int
 1910 msdosfs_print(struct vop_print_args *ap)
 1911 {
 1912         struct denode *dep = VTODE(ap->a_vp);
 1913 
 1914         printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
 1915                dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
 1916         printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
 1917         return (0);
 1918 }
 1919 
 1920 static int
 1921 msdosfs_pathconf(struct vop_pathconf_args *ap)
 1922 {
 1923         struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
 1924 
 1925         switch (ap->a_name) {
 1926         case _PC_FILESIZEBITS:
 1927                 *ap->a_retval = 32;
 1928                 return (0);
 1929         case _PC_LINK_MAX:
 1930                 *ap->a_retval = 1;
 1931                 return (0);
 1932         case _PC_NAME_MAX:
 1933                 *ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
 1934                 return (0);
 1935         case _PC_CHOWN_RESTRICTED:
 1936                 *ap->a_retval = 1;
 1937                 return (0);
 1938         case _PC_NO_TRUNC:
 1939                 *ap->a_retval = 0;
 1940                 return (0);
 1941         default:
 1942                 return (vop_stdpathconf(ap));
 1943         }
 1944         /* NOTREACHED */
 1945 }
 1946 
 1947 static int
 1948 msdosfs_vptofh(struct vop_vptofh_args *ap)
 1949 {
 1950         struct denode *dep;
 1951         struct defid *defhp;
 1952 
 1953         dep = VTODE(ap->a_vp);
 1954         defhp = (struct defid *)ap->a_fhp;
 1955         defhp->defid_len = sizeof(struct defid);
 1956         defhp->defid_dirclust = dep->de_dirclust;
 1957         defhp->defid_dirofs = dep->de_diroffset;
 1958         /* defhp->defid_gen = dep->de_gen; */
 1959         return (0);
 1960 }
 1961 
 1962 /* Global vfs data structures for msdosfs */
 1963 struct vop_vector msdosfs_vnodeops = {
 1964         .vop_default =          &default_vnodeops,
 1965 
 1966         .vop_access =           msdosfs_access,
 1967         .vop_bmap =             msdosfs_bmap,
 1968         .vop_getpages =         msdosfs_getpages,
 1969         .vop_cachedlookup =     msdosfs_lookup,
 1970         .vop_open =             msdosfs_open,
 1971         .vop_close =            msdosfs_close,
 1972         .vop_create =           msdosfs_create,
 1973         .vop_fsync =            msdosfs_fsync,
 1974         .vop_fdatasync =        vop_stdfdatasync_buf,
 1975         .vop_getattr =          msdosfs_getattr,
 1976         .vop_inactive =         msdosfs_inactive,
 1977         .vop_link =             msdosfs_link,
 1978         .vop_lookup =           vfs_cache_lookup,
 1979         .vop_mkdir =            msdosfs_mkdir,
 1980         .vop_mknod =            msdosfs_mknod,
 1981         .vop_pathconf =         msdosfs_pathconf,
 1982         .vop_print =            msdosfs_print,
 1983         .vop_read =             msdosfs_read,
 1984         .vop_readdir =          msdosfs_readdir,
 1985         .vop_reclaim =          msdosfs_reclaim,
 1986         .vop_remove =           msdosfs_remove,
 1987         .vop_rename =           msdosfs_rename,
 1988         .vop_rmdir =            msdosfs_rmdir,
 1989         .vop_setattr =          msdosfs_setattr,
 1990         .vop_strategy =         msdosfs_strategy,
 1991         .vop_symlink =          msdosfs_symlink,
 1992         .vop_write =            msdosfs_write,
 1993         .vop_vptofh =           msdosfs_vptofh,
 1994 };
 1995 VFS_VOP_VECTOR_REGISTER(msdosfs_vnodeops);

Cache object: 9b2ec790ceca45352a5ea29f2dedbbce


[ 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.