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/ufs/ufs/ufs_lookup.c

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

    1 /*-
    2  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 1989, 1993
    5  *      The Regents of the University of California.  All rights reserved.
    6  * (c) UNIX System Laboratories, Inc.
    7  * All or some portions of this file are derived from material licensed
    8  * to the University of California by American Telephone and Telegraph
    9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
   10  * the permission of UNIX System Laboratories, Inc.
   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. Neither the name of the University nor the names of its contributors
   21  *    may be used to endorse or promote products derived from this software
   22  *    without specific prior written permission.
   23  *
   24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   34  * SUCH DAMAGE.
   35  *
   36  *      @(#)ufs_lookup.c        8.15 (Berkeley) 6/16/95
   37  */
   38 
   39 #include <sys/cdefs.h>
   40 __FBSDID("$FreeBSD: head/sys/ufs/ufs/ufs_lookup.c 331095 2018-03-17 12:59:55Z emaste $");
   41 
   42 #include "opt_ufs.h"
   43 #include "opt_quota.h"
   44 
   45 #include <sys/param.h>
   46 #include <sys/systm.h>
   47 #include <sys/kernel.h>
   48 #include <sys/namei.h>
   49 #include <sys/bio.h>
   50 #include <sys/buf.h>
   51 #include <sys/proc.h>
   52 #include <sys/stat.h>
   53 #include <sys/mount.h>
   54 #include <sys/vnode.h>
   55 #include <sys/sysctl.h>
   56 
   57 #include <vm/vm.h>
   58 #include <vm/vm_extern.h>
   59 
   60 #include <ufs/ufs/extattr.h>
   61 #include <ufs/ufs/quota.h>
   62 #include <ufs/ufs/inode.h>
   63 #include <ufs/ufs/dir.h>
   64 #ifdef UFS_DIRHASH
   65 #include <ufs/ufs/dirhash.h>
   66 #endif
   67 #include <ufs/ufs/ufsmount.h>
   68 #include <ufs/ufs/ufs_extern.h>
   69 
   70 #ifdef DIAGNOSTIC
   71 static int      dirchk = 1;
   72 #else
   73 static int      dirchk = 0;
   74 #endif
   75 
   76 SYSCTL_INT(_debug, OID_AUTO, dircheck, CTLFLAG_RW, &dirchk, 0, "");
   77 
   78 /* true if old FS format...*/
   79 #define OFSFMT(vp)      ((vp)->v_mount->mnt_maxsymlinklen <= 0)
   80 
   81 static int
   82 ufs_delete_denied(struct vnode *vdp, struct vnode *tdp, struct ucred *cred,
   83     struct thread *td)
   84 {
   85         int error;
   86 
   87 #ifdef UFS_ACL
   88         /*
   89          * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
   90          *
   91          * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
   92          */
   93 
   94         /*
   95          * XXX: Is this check required?
   96          */
   97         error = VOP_ACCESS(vdp, VEXEC, cred, td);
   98         if (error)
   99                 return (error);
  100 
  101         error = VOP_ACCESSX(tdp, VDELETE, cred, td);
  102         if (error == 0)
  103                 return (0);
  104 
  105         error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred, td);
  106         if (error == 0)
  107                 return (0);
  108 
  109         error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred, td);
  110         if (error)
  111                 return (error);
  112 
  113 #endif /* !UFS_ACL */
  114 
  115         /*
  116          * Standard Unix access control - delete access requires VWRITE.
  117          */
  118         error = VOP_ACCESS(vdp, VWRITE, cred, td);
  119         if (error)
  120                 return (error);
  121 
  122         /*
  123          * If directory is "sticky", then user must own
  124          * the directory, or the file in it, else she
  125          * may not delete it (unless she's root). This
  126          * implements append-only directories.
  127          */
  128         if ((VTOI(vdp)->i_mode & ISVTX) &&
  129             VOP_ACCESS(vdp, VADMIN, cred, td) &&
  130             VOP_ACCESS(tdp, VADMIN, cred, td))
  131                 return (EPERM);
  132 
  133         return (0);
  134 }
  135 
  136 /*
  137  * Convert a component of a pathname into a pointer to a locked inode.
  138  * This is a very central and rather complicated routine.
  139  * If the filesystem is not maintained in a strict tree hierarchy,
  140  * this can result in a deadlock situation (see comments in code below).
  141  *
  142  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
  143  * on whether the name is to be looked up, created, renamed, or deleted.
  144  * When CREATE, RENAME, or DELETE is specified, information usable in
  145  * creating, renaming, or deleting a directory entry may be calculated.
  146  * If flag has LOCKPARENT or'ed into it and the target of the pathname
  147  * exists, lookup returns both the target and its parent directory locked.
  148  * When creating or renaming and LOCKPARENT is specified, the target may
  149  * not be ".".  When deleting and LOCKPARENT is specified, the target may
  150  * be "."., but the caller must check to ensure it does an vrele and vput
  151  * instead of two vputs.
  152  *
  153  * This routine is actually used as VOP_CACHEDLOOKUP method, and the
  154  * filesystem employs the generic vfs_cache_lookup() as VOP_LOOKUP
  155  * method.
  156  *
  157  * vfs_cache_lookup() performs the following for us:
  158  *      check that it is a directory
  159  *      check accessibility of directory
  160  *      check for modification attempts on read-only mounts
  161  *      if name found in cache
  162  *          if at end of path and deleting or creating
  163  *              drop it
  164  *           else
  165  *              return name.
  166  *      return VOP_CACHEDLOOKUP()
  167  *
  168  * Overall outline of ufs_lookup:
  169  *
  170  *      search for name in directory, to found or notfound
  171  * notfound:
  172  *      if creating, return locked directory, leaving info on available slots
  173  *      else return error
  174  * found:
  175  *      if at end of path and deleting, return information to allow delete
  176  *      if at end of path and rewriting (RENAME and LOCKPARENT), lock target
  177  *        inode and return info to allow rewrite
  178  *      if not at end, add name to cache; if at end and neither creating
  179  *        nor deleting, add name to cache
  180  */
  181 int
  182 ufs_lookup(ap)
  183         struct vop_cachedlookup_args /* {
  184                 struct vnode *a_dvp;
  185                 struct vnode **a_vpp;
  186                 struct componentname *a_cnp;
  187         } */ *ap;
  188 {
  189 
  190         return (ufs_lookup_ino(ap->a_dvp, ap->a_vpp, ap->a_cnp, NULL));
  191 }
  192 
  193 int
  194 ufs_lookup_ino(struct vnode *vdp, struct vnode **vpp, struct componentname *cnp,
  195     ino_t *dd_ino)
  196 {
  197         struct inode *dp;               /* inode for directory being searched */
  198         struct buf *bp;                 /* a buffer of directory entries */
  199         struct direct *ep;              /* the current directory entry */
  200         int entryoffsetinblock;         /* offset of ep in bp's buffer */
  201         enum {NONE, COMPACT, FOUND} slotstatus;
  202         doff_t slotoffset;              /* offset of area with free space */
  203         doff_t i_diroff;                /* cached i_diroff value. */
  204         doff_t i_offset;                /* cached i_offset value. */
  205         int slotsize;                   /* size of area at slotoffset */
  206         int slotfreespace;              /* amount of space free in slot */
  207         int slotneeded;                 /* size of the entry we're seeking */
  208         int numdirpasses;               /* strategy for directory search */
  209         doff_t endsearch;               /* offset to end directory search */
  210         doff_t prevoff;                 /* prev entry dp->i_offset */
  211         struct vnode *pdp;              /* saved dp during symlink work */
  212         struct vnode *tdp;              /* returned by VFS_VGET */
  213         doff_t enduseful;               /* pointer past last used dir slot */
  214         u_long bmask;                   /* block offset mask */
  215         int namlen, error;
  216         struct ucred *cred = cnp->cn_cred;
  217         int flags = cnp->cn_flags;
  218         int nameiop = cnp->cn_nameiop;
  219         ino_t ino, ino1;
  220         int ltype;
  221 
  222         if (vpp != NULL)
  223                 *vpp = NULL;
  224 
  225         dp = VTOI(vdp);
  226         if (dp->i_effnlink == 0)
  227                 return (ENOENT);
  228 
  229         /*
  230          * Create a vm object if vmiodirenable is enabled.
  231          * Alternatively we could call vnode_create_vobject
  232          * in VFS_VGET but we could end up creating objects
  233          * that are never used.
  234          */
  235         vnode_create_vobject(vdp, DIP(dp, i_size), cnp->cn_thread);
  236 
  237         bmask = VFSTOUFS(vdp->v_mount)->um_mountp->mnt_stat.f_iosize - 1;
  238 
  239 #ifdef DEBUG_VFS_LOCKS
  240         /*
  241          * Assert that the directory vnode is locked, and locked
  242          * exclusively for the last component lookup for modifying
  243          * operations.
  244          *
  245          * The directory-modifying operations need to save
  246          * intermediate state in the inode between namei() call and
  247          * actual directory manipulations.  See fields in the struct
  248          * inode marked as 'used during directory lookup'.  We must
  249          * ensure that upgrade in namei() does not happen, since
  250          * upgrade might need to unlock vdp.  If quotas are enabled,
  251          * getinoquota() also requires exclusive lock to modify inode.
  252          */
  253         ASSERT_VOP_LOCKED(vdp, "ufs_lookup1");
  254         if ((nameiop == CREATE || nameiop == DELETE || nameiop == RENAME) &&
  255             (flags & (LOCKPARENT | ISLASTCN)) == (LOCKPARENT | ISLASTCN))
  256                 ASSERT_VOP_ELOCKED(vdp, "ufs_lookup2");
  257 #endif
  258 
  259 restart:
  260         bp = NULL;
  261         slotoffset = -1;
  262 
  263         /*
  264          * We now have a segment name to search for, and a directory to search.
  265          *
  266          * Suppress search for slots unless creating
  267          * file and at end of pathname, in which case
  268          * we watch for a place to put the new file in
  269          * case it doesn't already exist.
  270          */
  271         ino = 0;
  272         i_diroff = dp->i_diroff;
  273         slotstatus = FOUND;
  274         slotfreespace = slotsize = slotneeded = 0;
  275         if ((nameiop == CREATE || nameiop == RENAME) &&
  276             (flags & ISLASTCN)) {
  277                 slotstatus = NONE;
  278                 slotneeded = DIRECTSIZ(cnp->cn_namelen);
  279         }
  280 
  281 #ifdef UFS_DIRHASH
  282         /*
  283          * Use dirhash for fast operations on large directories. The logic
  284          * to determine whether to hash the directory is contained within
  285          * ufsdirhash_build(); a zero return means that it decided to hash
  286          * this directory and it successfully built up the hash table.
  287          */
  288         if (ufsdirhash_build(dp) == 0) {
  289                 /* Look for a free slot if needed. */
  290                 enduseful = dp->i_size;
  291                 if (slotstatus != FOUND) {
  292                         slotoffset = ufsdirhash_findfree(dp, slotneeded,
  293                             &slotsize);
  294                         if (slotoffset >= 0) {
  295                                 slotstatus = COMPACT;
  296                                 enduseful = ufsdirhash_enduseful(dp);
  297                                 if (enduseful < 0)
  298                                         enduseful = dp->i_size;
  299                         }
  300                 }
  301                 /* Look up the component. */
  302                 numdirpasses = 1;
  303                 entryoffsetinblock = 0; /* silence compiler warning */
  304                 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
  305                     &i_offset, &bp, nameiop == DELETE ? &prevoff : NULL)) {
  306                 case 0:
  307                         ep = (struct direct *)((char *)bp->b_data +
  308                             (i_offset & bmask));
  309                         goto foundentry;
  310                 case ENOENT:
  311                         i_offset = roundup2(dp->i_size, DIRBLKSIZ);
  312                         goto notfound;
  313                 default:
  314                         /* Something failed; just do a linear search. */
  315                         break;
  316                 }
  317         }
  318 #endif /* UFS_DIRHASH */
  319         /*
  320          * If there is cached information on a previous search of
  321          * this directory, pick up where we last left off.
  322          * We cache only lookups as these are the most common
  323          * and have the greatest payoff. Caching CREATE has little
  324          * benefit as it usually must search the entire directory
  325          * to determine that the entry does not exist. Caching the
  326          * location of the last DELETE or RENAME has not reduced
  327          * profiling time and hence has been removed in the interest
  328          * of simplicity.
  329          */
  330         if (nameiop != LOOKUP || i_diroff == 0 || i_diroff >= dp->i_size) {
  331                 entryoffsetinblock = 0;
  332                 i_offset = 0;
  333                 numdirpasses = 1;
  334         } else {
  335                 i_offset = i_diroff;
  336                 if ((entryoffsetinblock = i_offset & bmask) &&
  337                     (error = UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp)))
  338                         return (error);
  339                 numdirpasses = 2;
  340                 nchstats.ncs_2passes++;
  341         }
  342         prevoff = i_offset;
  343         endsearch = roundup2(dp->i_size, DIRBLKSIZ);
  344         enduseful = 0;
  345 
  346 searchloop:
  347         while (i_offset < endsearch) {
  348                 /*
  349                  * If necessary, get the next directory block.
  350                  */
  351                 if ((i_offset & bmask) == 0) {
  352                         if (bp != NULL)
  353                                 brelse(bp);
  354                         error =
  355                             UFS_BLKATOFF(vdp, (off_t)i_offset, NULL, &bp);
  356                         if (error)
  357                                 return (error);
  358                         entryoffsetinblock = 0;
  359                 }
  360                 /*
  361                  * If still looking for a slot, and at a DIRBLKSIZE
  362                  * boundary, have to start looking for free space again.
  363                  */
  364                 if (slotstatus == NONE &&
  365                     (entryoffsetinblock & (DIRBLKSIZ - 1)) == 0) {
  366                         slotoffset = -1;
  367                         slotfreespace = 0;
  368                 }
  369                 /*
  370                  * Get pointer to next entry.
  371                  * Full validation checks are slow, so we only check
  372                  * enough to insure forward progress through the
  373                  * directory. Complete checks can be run by patching
  374                  * "dirchk" to be true.
  375                  */
  376                 ep = (struct direct *)((char *)bp->b_data + entryoffsetinblock);
  377                 if (ep->d_reclen == 0 || ep->d_reclen >
  378                     DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
  379                     (dirchk && ufs_dirbadentry(vdp, ep, entryoffsetinblock))) {
  380                         int i;
  381 
  382                         ufs_dirbad(dp, i_offset, "mangled entry");
  383                         i = DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1));
  384                         i_offset += i;
  385                         entryoffsetinblock += i;
  386                         continue;
  387                 }
  388 
  389                 /*
  390                  * If an appropriate sized slot has not yet been found,
  391                  * check to see if one is available. Also accumulate space
  392                  * in the current block so that we can determine if
  393                  * compaction is viable.
  394                  */
  395                 if (slotstatus != FOUND) {
  396                         int size = ep->d_reclen;
  397 
  398                         if (ep->d_ino != 0)
  399                                 size -= DIRSIZ(OFSFMT(vdp), ep);
  400                         if (size > 0) {
  401                                 if (size >= slotneeded) {
  402                                         slotstatus = FOUND;
  403                                         slotoffset = i_offset;
  404                                         slotsize = ep->d_reclen;
  405                                 } else if (slotstatus == NONE) {
  406                                         slotfreespace += size;
  407                                         if (slotoffset == -1)
  408                                                 slotoffset = i_offset;
  409                                         if (slotfreespace >= slotneeded) {
  410                                                 slotstatus = COMPACT;
  411                                                 slotsize = i_offset +
  412                                                       ep->d_reclen - slotoffset;
  413                                         }
  414                                 }
  415                         }
  416                 }
  417 
  418                 /*
  419                  * Check for a name match.
  420                  */
  421                 if (ep->d_ino) {
  422 #                       if (BYTE_ORDER == LITTLE_ENDIAN)
  423                                 if (OFSFMT(vdp))
  424                                         namlen = ep->d_type;
  425                                 else
  426                                         namlen = ep->d_namlen;
  427 #                       else
  428                                 namlen = ep->d_namlen;
  429 #                       endif
  430                         if (namlen == cnp->cn_namelen &&
  431                                 (cnp->cn_nameptr[0] == ep->d_name[0]) &&
  432                             !bcmp(cnp->cn_nameptr, ep->d_name,
  433                                 (unsigned)namlen)) {
  434 #ifdef UFS_DIRHASH
  435 foundentry:
  436 #endif
  437                                 /*
  438                                  * Save directory entry's inode number and
  439                                  * reclen in ndp->ni_ufs area, and release
  440                                  * directory buffer.
  441                                  */
  442                                 if (vdp->v_mount->mnt_maxsymlinklen > 0 &&
  443                                     ep->d_type == DT_WHT) {
  444                                         slotstatus = FOUND;
  445                                         slotoffset = i_offset;
  446                                         slotsize = ep->d_reclen;
  447                                         enduseful = dp->i_size;
  448                                         cnp->cn_flags |= ISWHITEOUT;
  449                                         numdirpasses--;
  450                                         goto notfound;
  451                                 }
  452                                 ino = ep->d_ino;
  453                                 goto found;
  454                         }
  455                 }
  456                 prevoff = i_offset;
  457                 i_offset += ep->d_reclen;
  458                 entryoffsetinblock += ep->d_reclen;
  459                 if (ep->d_ino)
  460                         enduseful = i_offset;
  461         }
  462 notfound:
  463         /*
  464          * If we started in the middle of the directory and failed
  465          * to find our target, we must check the beginning as well.
  466          */
  467         if (numdirpasses == 2) {
  468                 numdirpasses--;
  469                 i_offset = 0;
  470                 endsearch = i_diroff;
  471                 goto searchloop;
  472         }
  473         if (bp != NULL)
  474                 brelse(bp);
  475         /*
  476          * If creating, and at end of pathname and current
  477          * directory has not been removed, then can consider
  478          * allowing file to be created.
  479          */
  480         if ((nameiop == CREATE || nameiop == RENAME ||
  481              (nameiop == DELETE &&
  482               (cnp->cn_flags & DOWHITEOUT) &&
  483               (cnp->cn_flags & ISWHITEOUT))) &&
  484             (flags & ISLASTCN) && dp->i_effnlink != 0) {
  485                 /*
  486                  * Access for write is interpreted as allowing
  487                  * creation of files in the directory.
  488                  *
  489                  * XXX: Fix the comment above.
  490                  */
  491                 if (flags & WILLBEDIR)
  492                         error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
  493                 else
  494                         error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
  495                 if (error)
  496                         return (error);
  497                 /*
  498                  * Return an indication of where the new directory
  499                  * entry should be put.  If we didn't find a slot,
  500                  * then set dp->i_count to 0 indicating
  501                  * that the new slot belongs at the end of the
  502                  * directory. If we found a slot, then the new entry
  503                  * can be put in the range from dp->i_offset to
  504                  * dp->i_offset + dp->i_count.
  505                  */
  506                 if (slotstatus == NONE) {
  507                         dp->i_offset = roundup2(dp->i_size, DIRBLKSIZ);
  508                         dp->i_count = 0;
  509                         enduseful = dp->i_offset;
  510                 } else if (nameiop == DELETE) {
  511                         dp->i_offset = slotoffset;
  512                         if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
  513                                 dp->i_count = 0;
  514                         else
  515                                 dp->i_count = dp->i_offset - prevoff;
  516                 } else {
  517                         dp->i_offset = slotoffset;
  518                         dp->i_count = slotsize;
  519                         if (enduseful < slotoffset + slotsize)
  520                                 enduseful = slotoffset + slotsize;
  521                 }
  522                 dp->i_endoff = roundup2(enduseful, DIRBLKSIZ);
  523                 /*
  524                  * We return with the directory locked, so that
  525                  * the parameters we set up above will still be
  526                  * valid if we actually decide to do a direnter().
  527                  * We return ni_vp == NULL to indicate that the entry
  528                  * does not currently exist; we leave a pointer to
  529                  * the (locked) directory inode in ndp->ni_dvp.
  530                  * The pathname buffer is saved so that the name
  531                  * can be obtained later.
  532                  *
  533                  * NB - if the directory is unlocked, then this
  534                  * information cannot be used.
  535                  */
  536                 cnp->cn_flags |= SAVENAME;
  537                 return (EJUSTRETURN);
  538         }
  539         /*
  540          * Insert name into cache (as non-existent) if appropriate.
  541          */
  542         if ((cnp->cn_flags & MAKEENTRY) != 0)
  543                 cache_enter(vdp, NULL, cnp);
  544         return (ENOENT);
  545 
  546 found:
  547         if (dd_ino != NULL)
  548                 *dd_ino = ino;
  549         if (numdirpasses == 2)
  550                 nchstats.ncs_pass2++;
  551         /*
  552          * Check that directory length properly reflects presence
  553          * of this entry.
  554          */
  555         if (i_offset + DIRSIZ(OFSFMT(vdp), ep) > dp->i_size) {
  556                 ufs_dirbad(dp, i_offset, "i_size too small");
  557                 dp->i_size = i_offset + DIRSIZ(OFSFMT(vdp), ep);
  558                 DIP_SET(dp, i_size, dp->i_size);
  559                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
  560         }
  561         brelse(bp);
  562 
  563         /*
  564          * Found component in pathname.
  565          * If the final component of path name, save information
  566          * in the cache as to where the entry was found.
  567          */
  568         if ((flags & ISLASTCN) && nameiop == LOOKUP)
  569                 dp->i_diroff = rounddown2(i_offset, DIRBLKSIZ);
  570 
  571         /*
  572          * If deleting, and at end of pathname, return
  573          * parameters which can be used to remove file.
  574          */
  575         if (nameiop == DELETE && (flags & ISLASTCN)) {
  576                 if (flags & LOCKPARENT)
  577                         ASSERT_VOP_ELOCKED(vdp, __FUNCTION__);
  578                 /*
  579                  * Return pointer to current entry in dp->i_offset,
  580                  * and distance past previous entry (if there
  581                  * is a previous entry in this block) in dp->i_count.
  582                  * Save directory inode pointer in ndp->ni_dvp for dirremove().
  583                  *
  584                  * Technically we shouldn't be setting these in the
  585                  * WANTPARENT case (first lookup in rename()), but any
  586                  * lookups that will result in directory changes will
  587                  * overwrite these.
  588                  */
  589                 dp->i_offset = i_offset;
  590                 if ((dp->i_offset & (DIRBLKSIZ - 1)) == 0)
  591                         dp->i_count = 0;
  592                 else
  593                         dp->i_count = dp->i_offset - prevoff;
  594                 if (dd_ino != NULL)
  595                         return (0);
  596                 if ((error = VFS_VGET(vdp->v_mount, ino,
  597                     LK_EXCLUSIVE, &tdp)) != 0)
  598                         return (error);
  599                 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
  600                 if (error) {
  601                         vput(tdp);
  602                         return (error);
  603                 }
  604                 if (dp->i_number == ino) {
  605                         VREF(vdp);
  606                         *vpp = vdp;
  607                         vput(tdp);
  608                         return (0);
  609                 }
  610 
  611                 *vpp = tdp;
  612                 return (0);
  613         }
  614 
  615         /*
  616          * If rewriting (RENAME), return the inode and the
  617          * information required to rewrite the present directory
  618          * Must get inode of directory entry to verify it's a
  619          * regular file, or empty directory.
  620          */
  621         if (nameiop == RENAME && (flags & ISLASTCN)) {
  622                 if (flags & WILLBEDIR)
  623                         error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
  624                 else
  625                         error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
  626                 if (error)
  627                         return (error);
  628                 /*
  629                  * Careful about locking second inode.
  630                  * This can only occur if the target is ".".
  631                  */
  632                 dp->i_offset = i_offset;
  633                 if (dp->i_number == ino)
  634                         return (EISDIR);
  635                 if (dd_ino != NULL)
  636                         return (0);
  637                 if ((error = VFS_VGET(vdp->v_mount, ino,
  638                     LK_EXCLUSIVE, &tdp)) != 0)
  639                         return (error);
  640 
  641                 error = ufs_delete_denied(vdp, tdp, cred, cnp->cn_thread);
  642                 if (error) {
  643                         vput(tdp);
  644                         return (error);
  645                 }
  646 
  647 #ifdef SunOS_doesnt_do_that
  648                 /*
  649                  * The only purpose of this check is to return the correct
  650                  * error.  Assume that we want to rename directory "a"
  651                  * to a file "b", and that we have no ACL_WRITE_DATA on
  652                  * a containing directory, but we _do_ have ACL_APPEND_DATA. 
  653                  * In that case, the VOP_ACCESS check above will return 0,
  654                  * and the operation will fail with ENOTDIR instead
  655                  * of EACCESS.
  656                  */
  657                 if (tdp->v_type == VDIR)
  658                         error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred, cnp->cn_thread);
  659                 else
  660                         error = VOP_ACCESS(vdp, VWRITE, cred, cnp->cn_thread);
  661                 if (error) {
  662                         vput(tdp);
  663                         return (error);
  664                 }
  665 #endif
  666 
  667                 *vpp = tdp;
  668                 cnp->cn_flags |= SAVENAME;
  669                 return (0);
  670         }
  671         if (dd_ino != NULL)
  672                 return (0);
  673 
  674         /*
  675          * Step through the translation in the name.  We do not `vput' the
  676          * directory because we may need it again if a symbolic link
  677          * is relative to the current directory.  Instead we save it
  678          * unlocked as "pdp".  We must get the target inode before unlocking
  679          * the directory to insure that the inode will not be removed
  680          * before we get it.  We prevent deadlock by always fetching
  681          * inodes from the root, moving down the directory tree. Thus
  682          * when following backward pointers ".." we must unlock the
  683          * parent directory before getting the requested directory.
  684          * There is a potential race condition here if both the current
  685          * and parent directories are removed before the VFS_VGET for the
  686          * inode associated with ".." returns.  We hope that this occurs
  687          * infrequently since we cannot avoid this race condition without
  688          * implementing a sophisticated deadlock detection algorithm.
  689          * Note also that this simple deadlock detection scheme will not
  690          * work if the filesystem has any hard links other than ".."
  691          * that point backwards in the directory structure.
  692          */
  693         pdp = vdp;
  694         if (flags & ISDOTDOT) {
  695                 error = vn_vget_ino(pdp, ino, cnp->cn_lkflags, &tdp);
  696                 if (error)
  697                         return (error);
  698 
  699                 /*
  700                  * Recheck that ".." entry in the vdp directory points
  701                  * to the inode we looked up before vdp lock was
  702                  * dropped.
  703                  */
  704                 error = ufs_lookup_ino(pdp, NULL, cnp, &ino1);
  705                 if (error) {
  706                         vput(tdp);
  707                         return (error);
  708                 }
  709                 if (ino1 != ino) {
  710                         vput(tdp);
  711                         goto restart;
  712                 }
  713 
  714                 *vpp = tdp;
  715         } else if (dp->i_number == ino) {
  716                 VREF(vdp);      /* we want ourself, ie "." */
  717                 /*
  718                  * When we lookup "." we still can be asked to lock it
  719                  * differently.
  720                  */
  721                 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
  722                 if (ltype != VOP_ISLOCKED(vdp)) {
  723                         if (ltype == LK_EXCLUSIVE)
  724                                 vn_lock(vdp, LK_UPGRADE | LK_RETRY);
  725                         else /* if (ltype == LK_SHARED) */
  726                                 vn_lock(vdp, LK_DOWNGRADE | LK_RETRY);
  727                         /*
  728                          * Relock for the "." case may left us with
  729                          * reclaimed vnode.
  730                          */
  731                         if (vdp->v_iflag & VI_DOOMED) {
  732                                 vrele(vdp);
  733                                 return (ENOENT);
  734                         }
  735                 }
  736                 *vpp = vdp;
  737         } else {
  738                 error = VFS_VGET(pdp->v_mount, ino, cnp->cn_lkflags, &tdp);
  739                 if (error)
  740                         return (error);
  741                 *vpp = tdp;
  742         }
  743 
  744         /*
  745          * Insert name into cache if appropriate.
  746          */
  747         if (cnp->cn_flags & MAKEENTRY)
  748                 cache_enter(vdp, *vpp, cnp);
  749         return (0);
  750 }
  751 
  752 void
  753 ufs_dirbad(ip, offset, how)
  754         struct inode *ip;
  755         doff_t offset;
  756         char *how;
  757 {
  758         struct mount *mp;
  759 
  760         mp = ITOV(ip)->v_mount;
  761         if ((mp->mnt_flag & MNT_RDONLY) == 0)
  762                 panic("ufs_dirbad: %s: bad dir ino %ju at offset %ld: %s",
  763                     mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
  764                     (long)offset, how);
  765         else
  766                 (void)printf("%s: bad dir ino %ju at offset %ld: %s\n",
  767                     mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
  768                     (long)offset, how);
  769 }
  770 
  771 /*
  772  * Do consistency checking on a directory entry:
  773  *      record length must be multiple of 4
  774  *      entry must fit in rest of its DIRBLKSIZ block
  775  *      record must be large enough to contain entry
  776  *      name is not longer than UFS_MAXNAMLEN
  777  *      name must be as long as advertised, and null terminated
  778  */
  779 int
  780 ufs_dirbadentry(dp, ep, entryoffsetinblock)
  781         struct vnode *dp;
  782         struct direct *ep;
  783         int entryoffsetinblock;
  784 {
  785         int i, namlen;
  786 
  787 #       if (BYTE_ORDER == LITTLE_ENDIAN)
  788                 if (OFSFMT(dp))
  789                         namlen = ep->d_type;
  790                 else
  791                         namlen = ep->d_namlen;
  792 #       else
  793                 namlen = ep->d_namlen;
  794 #       endif
  795         if ((ep->d_reclen & 0x3) != 0 ||
  796             ep->d_reclen > DIRBLKSIZ - (entryoffsetinblock & (DIRBLKSIZ - 1)) ||
  797             ep->d_reclen < DIRSIZ(OFSFMT(dp), ep) || namlen > UFS_MAXNAMLEN) {
  798                 /*return (1); */
  799                 printf("First bad\n");
  800                 goto bad;
  801         }
  802         if (ep->d_ino == 0)
  803                 return (0);
  804         for (i = 0; i < namlen; i++)
  805                 if (ep->d_name[i] == '\0') {
  806                         /*return (1); */
  807                         printf("Second bad\n");
  808                         goto bad;
  809                 }
  810         if (ep->d_name[i])
  811                 goto bad;
  812         return (0);
  813 bad:
  814         return (1);
  815 }
  816 
  817 /*
  818  * Construct a new directory entry after a call to namei, using the
  819  * parameters that it left in the componentname argument cnp. The
  820  * argument ip is the inode to which the new directory entry will refer.
  821  */
  822 void
  823 ufs_makedirentry(ip, cnp, newdirp)
  824         struct inode *ip;
  825         struct componentname *cnp;
  826         struct direct *newdirp;
  827 {
  828 
  829 #ifdef INVARIANTS
  830         if ((cnp->cn_flags & SAVENAME) == 0)
  831                 panic("ufs_makedirentry: missing name");
  832 #endif
  833         newdirp->d_ino = ip->i_number;
  834         newdirp->d_namlen = cnp->cn_namelen;
  835         bcopy(cnp->cn_nameptr, newdirp->d_name, (unsigned)cnp->cn_namelen + 1);
  836         if (ITOV(ip)->v_mount->mnt_maxsymlinklen > 0)
  837                 newdirp->d_type = IFTODT(ip->i_mode);
  838         else {
  839                 newdirp->d_type = 0;
  840 #               if (BYTE_ORDER == LITTLE_ENDIAN)
  841                         { u_char tmp = newdirp->d_namlen;
  842                         newdirp->d_namlen = newdirp->d_type;
  843                         newdirp->d_type = tmp; }
  844 #               endif
  845         }
  846 }
  847 
  848 /*
  849  * Write a directory entry after a call to namei, using the parameters
  850  * that it left in nameidata. The argument dirp is the new directory
  851  * entry contents. Dvp is a pointer to the directory to be written,
  852  * which was left locked by namei. Remaining parameters (dp->i_offset, 
  853  * dp->i_count) indicate how the space for the new entry is to be obtained.
  854  * Non-null bp indicates that a directory is being created (for the
  855  * soft dependency code).
  856  */
  857 int
  858 ufs_direnter(dvp, tvp, dirp, cnp, newdirbp, isrename)
  859         struct vnode *dvp;
  860         struct vnode *tvp;
  861         struct direct *dirp;
  862         struct componentname *cnp;
  863         struct buf *newdirbp;
  864         int isrename;
  865 {
  866         struct ucred *cr;
  867         struct thread *td;
  868         int newentrysize;
  869         struct inode *dp;
  870         struct buf *bp;
  871         u_int dsize;
  872         struct direct *ep, *nep;
  873         u_int64_t old_isize;
  874         int error, ret, blkoff, loc, spacefree, flags, namlen;
  875         char *dirbuf;
  876 
  877         td = curthread; /* XXX */
  878         cr = td->td_ucred;
  879 
  880         dp = VTOI(dvp);
  881         newentrysize = DIRSIZ(OFSFMT(dvp), dirp);
  882 
  883         if (dp->i_count == 0) {
  884                 /*
  885                  * If dp->i_count is 0, then namei could find no
  886                  * space in the directory. Here, dp->i_offset will
  887                  * be on a directory block boundary and we will write the
  888                  * new entry into a fresh block.
  889                  */
  890                 if (dp->i_offset & (DIRBLKSIZ - 1))
  891                         panic("ufs_direnter: newblk");
  892                 flags = BA_CLRBUF;
  893                 if (!DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp))
  894                         flags |= IO_SYNC;
  895 #ifdef QUOTA
  896                 if ((error = getinoquota(dp)) != 0) {
  897                         if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
  898                                 bdwrite(newdirbp);
  899                         return (error);
  900                 }
  901 #endif
  902                 old_isize = dp->i_size;
  903                 vnode_pager_setsize(dvp, (u_long)dp->i_offset + DIRBLKSIZ);
  904                 if ((error = UFS_BALLOC(dvp, (off_t)dp->i_offset, DIRBLKSIZ,
  905                     cr, flags, &bp)) != 0) {
  906                         if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
  907                                 bdwrite(newdirbp);
  908                         vnode_pager_setsize(dvp, (u_long)old_isize);
  909                         return (error);
  910                 }
  911                 dp->i_size = dp->i_offset + DIRBLKSIZ;
  912                 DIP_SET(dp, i_size, dp->i_size);
  913                 dp->i_endoff = dp->i_size;
  914                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
  915                 dirp->d_reclen = DIRBLKSIZ;
  916                 blkoff = dp->i_offset &
  917                     (VFSTOUFS(dvp->v_mount)->um_mountp->mnt_stat.f_iosize - 1);
  918                 bcopy((caddr_t)dirp, (caddr_t)bp->b_data + blkoff,newentrysize);
  919 #ifdef UFS_DIRHASH
  920                 if (dp->i_dirhash != NULL) {
  921                         ufsdirhash_newblk(dp, dp->i_offset);
  922                         ufsdirhash_add(dp, dirp, dp->i_offset);
  923                         ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
  924                             dp->i_offset);
  925                 }
  926 #endif
  927                 if (DOINGSOFTDEP(dvp)) {
  928                         /*
  929                          * Ensure that the entire newly allocated block is a
  930                          * valid directory so that future growth within the
  931                          * block does not have to ensure that the block is
  932                          * written before the inode.
  933                          */
  934                         blkoff += DIRBLKSIZ;
  935                         while (blkoff < bp->b_bcount) {
  936                                 ((struct direct *)
  937                                    (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
  938                                 blkoff += DIRBLKSIZ;
  939                         }
  940                         if (softdep_setup_directory_add(bp, dp, dp->i_offset,
  941                             dirp->d_ino, newdirbp, 1))
  942                                 dp->i_flag |= IN_NEEDSYNC;
  943                         if (newdirbp)
  944                                 bdwrite(newdirbp);
  945                         bdwrite(bp);
  946                         if ((dp->i_flag & IN_NEEDSYNC) == 0)
  947                                 return (UFS_UPDATE(dvp, 0));
  948                         /*
  949                          * We have just allocated a directory block in an
  950                          * indirect block.  We must prevent holes in the
  951                          * directory created if directory entries are
  952                          * written out of order.  To accomplish this we
  953                          * fsync when we extend a directory into indirects.
  954                          * During rename it's not safe to drop the tvp lock
  955                          * so sync must be delayed until it is.
  956                          *
  957                          * This synchronous step could be removed if fsck and
  958                          * the kernel were taught to fill in sparse
  959                          * directories rather than panic.
  960                          */
  961                         if (isrename)
  962                                 return (0);
  963                         if (tvp != NULL)
  964                                 VOP_UNLOCK(tvp, 0);
  965                         (void) VOP_FSYNC(dvp, MNT_WAIT, td);
  966                         if (tvp != NULL)
  967                                 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
  968                         return (error);
  969                 }
  970                 if (DOINGASYNC(dvp)) {
  971                         bdwrite(bp);
  972                         return (UFS_UPDATE(dvp, 0));
  973                 }
  974                 error = bwrite(bp);
  975                 ret = UFS_UPDATE(dvp, 1);
  976                 if (error == 0)
  977                         return (ret);
  978                 return (error);
  979         }
  980 
  981         /*
  982          * If dp->i_count is non-zero, then namei found space for the new
  983          * entry in the range dp->i_offset to dp->i_offset + dp->i_count
  984          * in the directory. To use this space, we may have to compact
  985          * the entries located there, by copying them together towards the
  986          * beginning of the block, leaving the free space in one usable
  987          * chunk at the end.
  988          */
  989 
  990         /*
  991          * Increase size of directory if entry eats into new space.
  992          * This should never push the size past a new multiple of
  993          * DIRBLKSIZE.
  994          *
  995          * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
  996          */
  997         if (dp->i_offset + dp->i_count > dp->i_size) {
  998                 dp->i_size = dp->i_offset + dp->i_count;
  999                 DIP_SET(dp, i_size, dp->i_size);
 1000         }
 1001         /*
 1002          * Get the block containing the space for the new directory entry.
 1003          */
 1004         error = UFS_BLKATOFF(dvp, (off_t)dp->i_offset, &dirbuf, &bp);
 1005         if (error) {
 1006                 if (DOINGSOFTDEP(dvp) && newdirbp != NULL)
 1007                         bdwrite(newdirbp);
 1008                 return (error);
 1009         }
 1010         /*
 1011          * Find space for the new entry. In the simple case, the entry at
 1012          * offset base will have the space. If it does not, then namei
 1013          * arranged that compacting the region dp->i_offset to
 1014          * dp->i_offset + dp->i_count would yield the space.
 1015          */
 1016         ep = (struct direct *)dirbuf;
 1017         dsize = ep->d_ino ? DIRSIZ(OFSFMT(dvp), ep) : 0;
 1018         spacefree = ep->d_reclen - dsize;
 1019         for (loc = ep->d_reclen; loc < dp->i_count; ) {
 1020                 nep = (struct direct *)(dirbuf + loc);
 1021 
 1022                 /* Trim the existing slot (NB: dsize may be zero). */
 1023                 ep->d_reclen = dsize;
 1024                 ep = (struct direct *)((char *)ep + dsize);
 1025 
 1026                 /* Read nep->d_reclen now as the bcopy() may clobber it. */
 1027                 loc += nep->d_reclen;
 1028                 if (nep->d_ino == 0) {
 1029                         /*
 1030                          * A mid-block unused entry. Such entries are
 1031                          * never created by the kernel, but fsck_ffs
 1032                          * can create them (and it doesn't fix them).
 1033                          *
 1034                          * Add up the free space, and initialise the
 1035                          * relocated entry since we don't bcopy it.
 1036                          */
 1037                         spacefree += nep->d_reclen;
 1038                         ep->d_ino = 0;
 1039                         dsize = 0;
 1040                         continue;
 1041                 }
 1042                 dsize = DIRSIZ(OFSFMT(dvp), nep);
 1043                 spacefree += nep->d_reclen - dsize;
 1044 #ifdef UFS_DIRHASH
 1045                 if (dp->i_dirhash != NULL)
 1046                         ufsdirhash_move(dp, nep,
 1047                             dp->i_offset + ((char *)nep - dirbuf),
 1048                             dp->i_offset + ((char *)ep - dirbuf));
 1049 #endif
 1050                 if (DOINGSOFTDEP(dvp))
 1051                         softdep_change_directoryentry_offset(bp, dp, dirbuf,
 1052                             (caddr_t)nep, (caddr_t)ep, dsize); 
 1053                 else
 1054                         bcopy((caddr_t)nep, (caddr_t)ep, dsize);
 1055         }
 1056         /*
 1057          * Here, `ep' points to a directory entry containing `dsize' in-use
 1058          * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
 1059          * then the entry is completely unused (dsize == 0). The value
 1060          * of ep->d_reclen is always indeterminate.
 1061          *
 1062          * Update the pointer fields in the previous entry (if any),
 1063          * copy in the new entry, and write out the block.
 1064          */
 1065 #       if (BYTE_ORDER == LITTLE_ENDIAN)
 1066                 if (OFSFMT(dvp))
 1067                         namlen = ep->d_type;
 1068                 else
 1069                         namlen = ep->d_namlen;
 1070 #       else
 1071                 namlen = ep->d_namlen;
 1072 #       endif
 1073         if (ep->d_ino == 0 ||
 1074             (ep->d_ino == UFS_WINO && namlen == dirp->d_namlen &&
 1075              bcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
 1076                 if (spacefree + dsize < newentrysize)
 1077                         panic("ufs_direnter: compact1");
 1078                 dirp->d_reclen = spacefree + dsize;
 1079         } else {
 1080                 if (spacefree < newentrysize)
 1081                         panic("ufs_direnter: compact2");
 1082                 dirp->d_reclen = spacefree;
 1083                 ep->d_reclen = dsize;
 1084                 ep = (struct direct *)((char *)ep + dsize);
 1085         }
 1086 #ifdef UFS_DIRHASH
 1087         if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
 1088             dirp->d_reclen == spacefree))
 1089                 ufsdirhash_add(dp, dirp, dp->i_offset + ((char *)ep - dirbuf));
 1090 #endif
 1091         bcopy((caddr_t)dirp, (caddr_t)ep, (u_int)newentrysize);
 1092 #ifdef UFS_DIRHASH
 1093         if (dp->i_dirhash != NULL)
 1094                 ufsdirhash_checkblock(dp, dirbuf -
 1095                     (dp->i_offset & (DIRBLKSIZ - 1)),
 1096                     rounddown2(dp->i_offset, DIRBLKSIZ));
 1097 #endif
 1098 
 1099         if (DOINGSOFTDEP(dvp)) {
 1100                 (void) softdep_setup_directory_add(bp, dp,
 1101                     dp->i_offset + (caddr_t)ep - dirbuf,
 1102                     dirp->d_ino, newdirbp, 0);
 1103                 if (newdirbp != NULL)
 1104                         bdwrite(newdirbp);
 1105                 bdwrite(bp);
 1106         } else {
 1107                 if (DOINGASYNC(dvp)) {
 1108                         bdwrite(bp);
 1109                         error = 0;
 1110                 } else {
 1111                         error = bwrite(bp);
 1112                 }
 1113         }
 1114         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 1115         /*
 1116          * If all went well, and the directory can be shortened, proceed
 1117          * with the truncation. Note that we have to unlock the inode for
 1118          * the entry that we just entered, as the truncation may need to
 1119          * lock other inodes which can lead to deadlock if we also hold a
 1120          * lock on the newly entered node.
 1121          */
 1122         if (isrename == 0 && error == 0 &&
 1123             dp->i_endoff && dp->i_endoff < dp->i_size) {
 1124                 if (tvp != NULL)
 1125                         VOP_UNLOCK(tvp, 0);
 1126                 error = UFS_TRUNCATE(dvp, (off_t)dp->i_endoff,
 1127                     IO_NORMAL | (DOINGASYNC(dvp) ? 0 : IO_SYNC), cr);
 1128                 if (error != 0)
 1129                         vn_printf(dvp,
 1130                             "ufs_direnter: failed to truncate, error %d\n",
 1131                             error);
 1132 #ifdef UFS_DIRHASH
 1133                 if (error == 0 && dp->i_dirhash != NULL)
 1134                         ufsdirhash_dirtrunc(dp, dp->i_endoff);
 1135 #endif
 1136                 error = 0;
 1137                 if (tvp != NULL)
 1138                         vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
 1139         }
 1140         return (error);
 1141 }
 1142 
 1143 /*
 1144  * Remove a directory entry after a call to namei, using
 1145  * the parameters which it left in nameidata. The entry
 1146  * dp->i_offset contains the offset into the directory of the
 1147  * entry to be eliminated.  The dp->i_count field contains the
 1148  * size of the previous record in the directory.  If this
 1149  * is 0, the first entry is being deleted, so we need only
 1150  * zero the inode number to mark the entry as free.  If the
 1151  * entry is not the first in the directory, we must reclaim
 1152  * the space of the now empty record by adding the record size
 1153  * to the size of the previous entry.
 1154  */
 1155 int
 1156 ufs_dirremove(dvp, ip, flags, isrmdir)
 1157         struct vnode *dvp;
 1158         struct inode *ip;
 1159         int flags;
 1160         int isrmdir;
 1161 {
 1162         struct inode *dp;
 1163         struct direct *ep, *rep;
 1164         struct buf *bp;
 1165         int error;
 1166 
 1167         dp = VTOI(dvp);
 1168 
 1169         /*
 1170          * Adjust the link count early so softdep can block if necessary.
 1171          */
 1172         if (ip) {
 1173                 ip->i_effnlink--;
 1174                 if (DOINGSOFTDEP(dvp)) {
 1175                         softdep_setup_unlink(dp, ip);
 1176                 } else {
 1177                         ip->i_nlink--;
 1178                         DIP_SET(ip, i_nlink, ip->i_nlink);
 1179                         ip->i_flag |= IN_CHANGE;
 1180                 }
 1181         }
 1182         if (flags & DOWHITEOUT) {
 1183                 /*
 1184                  * Whiteout entry: set d_ino to UFS_WINO.
 1185                  */
 1186                 if ((error =
 1187                     UFS_BLKATOFF(dvp, (off_t)dp->i_offset, (char **)&ep, &bp)) != 0)
 1188                         return (error);
 1189                 ep->d_ino = UFS_WINO;
 1190                 ep->d_type = DT_WHT;
 1191                 goto out;
 1192         }
 1193 
 1194         if ((error = UFS_BLKATOFF(dvp,
 1195             (off_t)(dp->i_offset - dp->i_count), (char **)&ep, &bp)) != 0)
 1196                 return (error);
 1197 
 1198         /* Set 'rep' to the entry being removed. */
 1199         if (dp->i_count == 0)
 1200                 rep = ep;
 1201         else
 1202                 rep = (struct direct *)((char *)ep + ep->d_reclen);
 1203 #ifdef UFS_DIRHASH
 1204         /*
 1205          * Remove the dirhash entry. This is complicated by the fact
 1206          * that `ep' is the previous entry when dp->i_count != 0.
 1207          */
 1208         if (dp->i_dirhash != NULL)
 1209                 ufsdirhash_remove(dp, rep, dp->i_offset);
 1210 #endif
 1211         if (ip && rep->d_ino != ip->i_number)
 1212                 panic("ufs_dirremove: ip %ju does not match dirent ino %ju\n",
 1213                     (uintmax_t)ip->i_number, (uintmax_t)rep->d_ino);
 1214         if (dp->i_count == 0) {
 1215                 /*
 1216                  * First entry in block: set d_ino to zero.
 1217                  */
 1218                 ep->d_ino = 0;
 1219         } else {
 1220                 /*
 1221                  * Collapse new free space into previous entry.
 1222                  */
 1223                 ep->d_reclen += rep->d_reclen;
 1224         }
 1225 #ifdef UFS_DIRHASH
 1226         if (dp->i_dirhash != NULL)
 1227                 ufsdirhash_checkblock(dp, (char *)ep -
 1228                     ((dp->i_offset - dp->i_count) & (DIRBLKSIZ - 1)),
 1229                     rounddown2(dp->i_offset, DIRBLKSIZ));
 1230 #endif
 1231 out:
 1232         error = 0;
 1233         if (DOINGSOFTDEP(dvp)) {
 1234                 if (ip)
 1235                         softdep_setup_remove(bp, dp, ip, isrmdir);
 1236                 if (softdep_slowdown(dvp))
 1237                         error = bwrite(bp);
 1238                 else
 1239                         bdwrite(bp);
 1240         } else {
 1241                 if (flags & DOWHITEOUT)
 1242                         error = bwrite(bp);
 1243                 else if (DOINGASYNC(dvp))
 1244                         bdwrite(bp);
 1245                 else
 1246                         error = bwrite(bp);
 1247         }
 1248         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 1249         /*
 1250          * If the last named reference to a snapshot goes away,
 1251          * drop its snapshot reference so that it will be reclaimed
 1252          * when last open reference goes away.
 1253          */
 1254         if (ip != NULL && (ip->i_flags & SF_SNAPSHOT) != 0 &&
 1255             ip->i_effnlink == 0)
 1256                 UFS_SNAPGONE(ip);
 1257         return (error);
 1258 }
 1259 
 1260 /*
 1261  * Rewrite an existing directory entry to point at the inode
 1262  * supplied.  The parameters describing the directory entry are
 1263  * set up by a call to namei.
 1264  */
 1265 int
 1266 ufs_dirrewrite(dp, oip, newinum, newtype, isrmdir)
 1267         struct inode *dp, *oip;
 1268         ino_t newinum;
 1269         int newtype;
 1270         int isrmdir;
 1271 {
 1272         struct buf *bp;
 1273         struct direct *ep;
 1274         struct vnode *vdp = ITOV(dp);
 1275         int error;
 1276 
 1277         /*
 1278          * Drop the link before we lock the buf so softdep can block if
 1279          * necessary.
 1280          */
 1281         oip->i_effnlink--;
 1282         if (DOINGSOFTDEP(vdp)) {
 1283                 softdep_setup_unlink(dp, oip);
 1284         } else {
 1285                 oip->i_nlink--;
 1286                 DIP_SET(oip, i_nlink, oip->i_nlink);
 1287                 oip->i_flag |= IN_CHANGE;
 1288         }
 1289 
 1290         error = UFS_BLKATOFF(vdp, (off_t)dp->i_offset, (char **)&ep, &bp);
 1291         if (error)
 1292                 return (error);
 1293         if (ep->d_namlen == 2 && ep->d_name[1] == '.' && ep->d_name[0] == '.' &&
 1294             ep->d_ino != oip->i_number) {
 1295                 brelse(bp);
 1296                 return (EIDRM);
 1297         }
 1298         ep->d_ino = newinum;
 1299         if (!OFSFMT(vdp))
 1300                 ep->d_type = newtype;
 1301         if (DOINGSOFTDEP(vdp)) {
 1302                 softdep_setup_directory_change(bp, dp, oip, newinum, isrmdir);
 1303                 bdwrite(bp);
 1304         } else {
 1305                 if (DOINGASYNC(vdp)) {
 1306                         bdwrite(bp);
 1307                         error = 0;
 1308                 } else {
 1309                         error = bwrite(bp);
 1310                 }
 1311         }
 1312         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 1313         /*
 1314          * If the last named reference to a snapshot goes away,
 1315          * drop its snapshot reference so that it will be reclaimed
 1316          * when last open reference goes away.
 1317          */
 1318         if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_effnlink == 0)
 1319                 UFS_SNAPGONE(oip);
 1320         return (error);
 1321 }
 1322 
 1323 /*
 1324  * Check if a directory is empty or not.
 1325  * Inode supplied must be locked.
 1326  *
 1327  * Using a struct dirtemplate here is not precisely
 1328  * what we want, but better than using a struct direct.
 1329  *
 1330  * NB: does not handle corrupted directories.
 1331  */
 1332 int
 1333 ufs_dirempty(ip, parentino, cred)
 1334         struct inode *ip;
 1335         ino_t parentino;
 1336         struct ucred *cred;
 1337 {
 1338         doff_t off;
 1339         struct dirtemplate dbuf;
 1340         struct direct *dp = (struct direct *)&dbuf;
 1341         int error, namlen;
 1342         ssize_t count;
 1343 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
 1344 
 1345         for (off = 0; off < ip->i_size; off += dp->d_reclen) {
 1346                 error = vn_rdwr(UIO_READ, ITOV(ip), (caddr_t)dp, MINDIRSIZ,
 1347                     off, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, cred,
 1348                     NOCRED, &count, (struct thread *)0);
 1349                 /*
 1350                  * Since we read MINDIRSIZ, residual must
 1351                  * be 0 unless we're at end of file.
 1352                  */
 1353                 if (error || count != 0)
 1354                         return (0);
 1355                 /* avoid infinite loops */
 1356                 if (dp->d_reclen == 0)
 1357                         return (0);
 1358                 /* skip empty entries */
 1359                 if (dp->d_ino == 0 || dp->d_ino == UFS_WINO)
 1360                         continue;
 1361                 /* accept only "." and ".." */
 1362 #               if (BYTE_ORDER == LITTLE_ENDIAN)
 1363                         if (OFSFMT(ITOV(ip)))
 1364                                 namlen = dp->d_type;
 1365                         else
 1366                                 namlen = dp->d_namlen;
 1367 #               else
 1368                         namlen = dp->d_namlen;
 1369 #               endif
 1370                 if (namlen > 2)
 1371                         return (0);
 1372                 if (dp->d_name[0] != '.')
 1373                         return (0);
 1374                 /*
 1375                  * At this point namlen must be 1 or 2.
 1376                  * 1 implies ".", 2 implies ".." if second
 1377                  * char is also "."
 1378                  */
 1379                 if (namlen == 1 && dp->d_ino == ip->i_number)
 1380                         continue;
 1381                 if (dp->d_name[1] == '.' && dp->d_ino == parentino)
 1382                         continue;
 1383                 return (0);
 1384         }
 1385         return (1);
 1386 }
 1387 
 1388 static int
 1389 ufs_dir_dd_ino(struct vnode *vp, struct ucred *cred, ino_t *dd_ino,
 1390     struct vnode **dd_vp)
 1391 {
 1392         struct dirtemplate dirbuf;
 1393         struct vnode *ddvp;
 1394         int error, namlen;
 1395 
 1396         ASSERT_VOP_LOCKED(vp, "ufs_dir_dd_ino");
 1397         if (vp->v_type != VDIR)
 1398                 return (ENOTDIR);
 1399         /*
 1400          * First check to see if we have it in the name cache.
 1401          */
 1402         if ((ddvp = vn_dir_dd_ino(vp)) != NULL) {
 1403                 KASSERT(ddvp->v_mount == vp->v_mount,
 1404                     ("ufs_dir_dd_ino: Unexpected mount point crossing"));
 1405                 *dd_ino = VTOI(ddvp)->i_number;
 1406                 *dd_vp = ddvp;
 1407                 return (0);
 1408         }
 1409         /*
 1410          * Have to read the directory.
 1411          */
 1412         error = vn_rdwr(UIO_READ, vp, (caddr_t)&dirbuf,
 1413             sizeof (struct dirtemplate), (off_t)0, UIO_SYSSPACE,
 1414             IO_NODELOCKED | IO_NOMACCHECK, cred, NOCRED, NULL, NULL);
 1415         if (error != 0)
 1416                 return (error);
 1417 #if (BYTE_ORDER == LITTLE_ENDIAN)
 1418         if (OFSFMT(vp))
 1419                 namlen = dirbuf.dotdot_type;
 1420         else
 1421                 namlen = dirbuf.dotdot_namlen;
 1422 #else
 1423         namlen = dirbuf.dotdot_namlen;
 1424 #endif
 1425         if (namlen != 2 || dirbuf.dotdot_name[0] != '.' ||
 1426             dirbuf.dotdot_name[1] != '.')
 1427                 return (ENOTDIR);
 1428         *dd_ino = dirbuf.dotdot_ino;
 1429         *dd_vp = NULL;
 1430         return (0);
 1431 }
 1432 
 1433 /*
 1434  * Check if source directory is in the path of the target directory.
 1435  */
 1436 int
 1437 ufs_checkpath(ino_t source_ino, ino_t parent_ino, struct inode *target, struct ucred *cred, ino_t *wait_ino)
 1438 {
 1439         struct mount *mp;
 1440         struct vnode *tvp, *vp, *vp1;
 1441         int error;
 1442         ino_t dd_ino;
 1443 
 1444         vp = tvp = ITOV(target);
 1445         mp = vp->v_mount;
 1446         *wait_ino = 0;
 1447         if (target->i_number == source_ino)
 1448                 return (EEXIST);
 1449         if (target->i_number == parent_ino)
 1450                 return (0);
 1451         if (target->i_number == UFS_ROOTINO)
 1452                 return (0);
 1453         for (;;) {
 1454                 error = ufs_dir_dd_ino(vp, cred, &dd_ino, &vp1);
 1455                 if (error != 0)
 1456                         break;
 1457                 if (dd_ino == source_ino) {
 1458                         error = EINVAL;
 1459                         break;
 1460                 }
 1461                 if (dd_ino == UFS_ROOTINO)
 1462                         break;
 1463                 if (dd_ino == parent_ino)
 1464                         break;
 1465                 if (vp1 == NULL) {
 1466                         error = VFS_VGET(mp, dd_ino, LK_SHARED | LK_NOWAIT,
 1467                             &vp1);
 1468                         if (error != 0) {
 1469                                 *wait_ino = dd_ino;
 1470                                 break;
 1471                         }
 1472                 }
 1473                 KASSERT(dd_ino == VTOI(vp1)->i_number,
 1474                     ("directory %ju reparented\n",
 1475                     (uintmax_t)VTOI(vp1)->i_number));
 1476                 if (vp != tvp)
 1477                         vput(vp);
 1478                 vp = vp1;
 1479         }
 1480 
 1481         if (error == ENOTDIR)
 1482                 panic("checkpath: .. not a directory\n");
 1483         if (vp1 != NULL)
 1484                 vput(vp1);
 1485         if (vp != tvp)
 1486                 vput(vp);
 1487         return (error);
 1488 }

Cache object: 957d9f24d389d91324d99123e314af3e


[ 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.