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-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 /*      $NetBSD: ufs_lookup.c,v 1.156 2022/08/06 18:26:42 andvar Exp $  */
    2 
    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        8.9 (Berkeley) 8/11/94
   37  */
   38 
   39 #include <sys/cdefs.h>
   40 __KERNEL_RCSID(0, "$NetBSD: ufs_lookup.c,v 1.156 2022/08/06 18:26:42 andvar Exp $");
   41 
   42 #ifdef _KERNEL_OPT
   43 #include "opt_ffs.h"
   44 #endif
   45 
   46 #include <sys/param.h>
   47 #include <sys/systm.h>
   48 #include <sys/namei.h>
   49 #include <sys/buf.h>
   50 #include <sys/file.h>
   51 #include <sys/stat.h>
   52 #include <sys/mount.h>
   53 #include <sys/vnode.h>
   54 #include <sys/kernel.h>
   55 #include <sys/kauth.h>
   56 #include <sys/wapbl.h>
   57 #include <sys/proc.h>
   58 #include <sys/kmem.h>
   59 
   60 #include <ufs/ufs/inode.h>
   61 #include <ufs/ufs/dir.h>
   62 #ifdef UFS_DIRHASH
   63 #include <ufs/ufs/dirhash.h>
   64 #endif
   65 #include <ufs/ufs/ufsmount.h>
   66 #include <ufs/ufs/ufs_extern.h>
   67 #include <ufs/ufs/ufs_bswap.h>
   68 #include <ufs/ufs/ufs_wapbl.h>
   69 
   70 #include <miscfs/genfs/genfs.h>
   71 
   72 #ifdef DIAGNOSTIC
   73 int     dirchk = 1;
   74 #else
   75 int     dirchk = 0;
   76 #endif
   77 
   78 #if BYTE_ORDER == LITTLE_ENDIAN
   79 # define ENDIANSWAP(needswap) ((needswap) == 0)
   80 #else
   81 # define ENDIANSWAP(needswap) ((needswap) != 0)
   82 #endif
   83 
   84 #define NAMLEN(fsfmt, needswap, dp) \
   85     ((fsfmt) && ENDIANSWAP(needswap) ? (dp)->d_type : (dp)->d_namlen)
   86 
   87 static void
   88 ufs_dirswap(struct direct *dirp)
   89 {
   90         uint8_t tmp = dirp->d_namlen;
   91         dirp->d_namlen = dirp->d_type;
   92         dirp->d_type = tmp;
   93 }
   94 
   95 struct slotinfo {
   96         enum {
   97                 NONE,           /* need to search a slot for our new entry */
   98                 COMPACT,        /* a compaction can make a slot in the current
   99                                    DIRBLKSIZ block */
  100                 FOUND,          /* found a slot (or no need to search) */
  101         } status;
  102         doff_t offset;          /* offset of area with free space.
  103                                    a special value -1 for invalid */
  104         int size;               /* size of area at slotoffset */
  105         int freespace;          /* accumulated amount of space free in
  106                                    the current DIRBLKSIZ block */
  107         int needed;             /* size of the entry we're seeking */
  108 };
  109 
  110 static void
  111 calc_count(struct ufs_lookup_results *results, int dirblksiz, doff_t prevoff)
  112 {
  113         if ((results->ulr_offset & (dirblksiz - 1)) == 0)
  114                 results->ulr_count = 0;
  115         else
  116                 results->ulr_count = results->ulr_offset - prevoff;
  117 }
  118 
  119 static void
  120 slot_init(struct slotinfo *slot)
  121 {
  122         slot->status = FOUND;
  123         slot->offset = -1;
  124         slot->freespace = slot->size = slot->needed = 0;
  125 }
  126 
  127 #ifdef UFS_DIRHASH
  128 static doff_t
  129 slot_findfree(struct slotinfo *slot, struct inode *dp)
  130 {
  131         if (slot->status == FOUND)
  132                 return dp->i_size;
  133 
  134         slot->offset = ufsdirhash_findfree(dp, slot->needed, &slot->size);
  135         if (slot->offset < 0)
  136                 return dp->i_size;
  137 
  138         slot->status = COMPACT;
  139         doff_t enduseful = ufsdirhash_enduseful(dp);
  140         if (enduseful < 0)
  141                 return dp->i_size;
  142         return enduseful;
  143 }
  144 #endif
  145 
  146 static void
  147 slot_white(struct slotinfo *slot, uint16_t reclen,
  148     struct ufs_lookup_results *results)
  149 {
  150         slot->status = FOUND;
  151         slot->offset = results->ulr_offset;
  152         slot->size = reclen;
  153         results->ulr_reclen = slot->size;
  154 }
  155 
  156 static void
  157 slot_update(struct slotinfo *slot, int size, uint16_t reclen, doff_t offset)
  158 {
  159         if (size >= slot->needed) {
  160                 slot->status = FOUND;
  161                 slot->offset = offset;
  162                 slot->size = reclen;
  163         } else if (slot->status == NONE) {
  164                 slot->freespace += size;
  165                 if (slot->offset == -1)
  166                         slot->offset = offset;
  167                 if (slot->freespace >= slot->needed) {
  168                         slot->status = COMPACT;
  169                         slot->size = offset + reclen - slot->offset;
  170                 }
  171         }
  172 }
  173 
  174 /*
  175  * Return an indication of where the new directory entry should be put.
  176  * If we didn't find a slot, then set results->ulr_count to 0 indicating
  177  * that the new slot belongs at the end of the directory. If we found a slot,
  178  * then the new entry can be put in the range from results->ulr_offset to
  179  * results->ulr_offset + results->ulr_count.
  180  */
  181 static int
  182 slot_estimate(const struct slotinfo *slot, int dirblksiz, int nameiop,
  183     doff_t prevoff, doff_t enduseful, const struct inode *ip,
  184     struct ufs_lookup_results *results)
  185 {
  186         if (slot->status == NONE) {
  187                 results->ulr_offset = roundup(ip->i_size, dirblksiz);
  188                 results->ulr_count = 0;
  189                 enduseful = results->ulr_offset;
  190         } else if (nameiop == DELETE) {
  191                 results->ulr_offset = slot->offset;
  192                 calc_count(results, dirblksiz, prevoff);
  193         } else {
  194                 results->ulr_offset = slot->offset;
  195                 results->ulr_count = slot->size;
  196                 if (enduseful < slot->offset + slot->size)
  197                         enduseful = slot->offset + slot->size;
  198         }
  199         results->ulr_endoff = roundup(enduseful, dirblksiz);
  200 #if 0 /* commented out by dbj. none of the on disk fields changed */
  201         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  202 #endif
  203         return EJUSTRETURN;
  204 }
  205 
  206 /*
  207  * Check if we can delete inode tdp in directory vdp with inode ip and creds.
  208  */
  209 static int
  210 ufs_can_delete(struct vnode *tdp, struct vnode *vdp, struct inode *ip,
  211     kauth_cred_t cred)
  212 {
  213         int error;
  214 
  215 #ifdef UFS_ACL
  216         /*
  217          * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
  218          *
  219          * 3.16.2.1. ACE4_DELETE vs. ACE4_DELETE_CHILD
  220          */
  221 
  222         /*
  223          * XXX: Is this check required?
  224          */
  225         error = VOP_ACCESS(vdp, VEXEC, cred);
  226         if (error)
  227                 goto out;
  228 
  229 #if 0
  230         /* Moved to ufs_remove, ufs_rmdir because they hold the lock */
  231         error = VOP_ACCESSX(tdp, VDELETE, cred);
  232         if (error == 0)
  233                 return (0);
  234 #endif
  235 
  236         error = VOP_ACCESSX(vdp, VDELETE_CHILD, cred);
  237         if (error == 0)
  238                 return (0);
  239 
  240         error = VOP_ACCESSX(vdp, VEXPLICIT_DENY | VDELETE_CHILD, cred);
  241         if (error)
  242                 goto out;
  243 
  244 #endif /* !UFS_ACL */
  245 
  246         /*
  247          * Write access to directory required to delete files.
  248          */
  249         error = VOP_ACCESS(vdp, VWRITE, cred);
  250         if (error)
  251                 goto out;
  252 
  253         if (!(ip->i_mode & ISVTX)) 
  254                 return 0;
  255 
  256         /*
  257          * If directory is "sticky", then user must own
  258          * the directory, or the file in it, else she
  259          * may not delete it (unless she's root). This
  260          * implements append-only directories.
  261          */
  262         error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE, tdp, vdp,
  263             genfs_can_sticky(vdp, cred, ip->i_uid, VTOI(tdp)->i_uid));
  264         if (error) {
  265                 error = EPERM;  // Why override?
  266                 goto out;
  267         }
  268         return 0;
  269 out:
  270         vrele(tdp);
  271         return error;
  272 }
  273 
  274 static int
  275 ufs_getino(struct vnode *vdp, struct inode *ip, ino_t foundino,
  276     struct vnode **tdp, bool same)
  277 {
  278         if (ip->i_number == foundino) {
  279                 if (same)
  280                         return EISDIR;
  281                 vref(vdp);
  282                 *tdp = vdp;
  283                 return 0;
  284         }
  285         return vcache_get(vdp->v_mount, &foundino, sizeof(foundino), tdp);
  286 }
  287 
  288 
  289 /*
  290  * Convert a component of a pathname into a pointer to a locked inode.
  291  * This is a very central and rather complicated routine.
  292  * If the file system is not maintained in a strict tree hierarchy,
  293  * this can result in a deadlock situation (see comments in code below).
  294  *
  295  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
  296  * on whether the name is to be looked up, created, renamed, or deleted.
  297  * When CREATE, RENAME, or DELETE is specified, information usable in
  298  * creating, renaming, or deleting a directory entry may be calculated.
  299  * If flag has LOCKPARENT or'ed into it and the target of the pathname
  300  * exists, lookup returns both the target and its parent directory locked.
  301  * When creating or renaming and LOCKPARENT is specified, the target may
  302  * not be ".".  When deleting and LOCKPARENT is specified, the target may
  303  * be "."., but the caller must check to ensure it does an vrele and vput
  304  * instead of two vputs.
  305  *
  306  * Overall outline of ufs_lookup:
  307  *
  308  *      check accessibility of directory
  309  *      look for name in cache, if found, then if at end of path
  310  *        and deleting or creating, drop it, else return name
  311  *      search for name in directory, to found or notfound
  312  * notfound:
  313  *      if creating, return locked directory, leaving info on available slots
  314  *      else return error
  315  * found:
  316  *      if at end of path and deleting, return information to allow delete
  317  *      if at end of path and rewriting (RENAME and LOCKPARENT), lock target
  318  *        inode and return info to allow rewrite
  319  *      if not at end, add name to cache; if at end and neither creating
  320  *        nor deleting, add name to cache
  321  */
  322 int
  323 ufs_lookup(void *v)
  324 {
  325         struct vop_lookup_v2_args /* {
  326                 struct vnode *a_dvp;
  327                 struct vnode **a_vpp;
  328                 struct componentname *a_cnp;
  329         } */ *ap = v;
  330         struct vnode *vdp = ap->a_dvp;  /* vnode for directory being searched */
  331         struct inode *dp = VTOI(vdp);   /* inode for directory being searched */
  332         struct buf *bp;                 /* a buffer of directory entries */
  333         struct direct *ep;              /* the current directory entry */
  334         int entryoffsetinblock;         /* offset of ep in bp's buffer */
  335         struct slotinfo slot;
  336         int numdirpasses;               /* strategy for directory search */
  337         doff_t endsearch;               /* offset to end directory search */
  338         doff_t prevoff;                 /* previous value of ulr_offset */
  339         struct vnode *tdp;              /* returned by vcache_get */
  340         doff_t enduseful;               /* pointer past last used dir slot.
  341                                            used for directory truncation. */
  342         u_long bmask;                   /* block offset mask */
  343         int error;
  344         struct vnode **vpp = ap->a_vpp;
  345         struct componentname *cnp = ap->a_cnp;
  346         kauth_cred_t cred = cnp->cn_cred;
  347         int flags;
  348         int nameiop = cnp->cn_nameiop;
  349         struct ufsmount *ump = dp->i_ump;
  350         const int needswap = UFS_MPNEEDSWAP(ump);
  351         int dirblksiz = ump->um_dirblksiz;
  352         ino_t foundino;
  353         struct ufs_lookup_results *results;
  354         int iswhiteout;                 /* temp result from cache_lookup() */
  355         const int fsfmt = FSFMT(vdp);
  356         uint16_t reclen;
  357 
  358         flags = cnp->cn_flags;
  359 
  360         bp = NULL;
  361         *vpp = NULL;
  362         endsearch = 0; /* silence compiler warning */
  363 
  364         /*
  365          * Check accessibility of directory.
  366          */
  367         if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
  368                 return (error);
  369 
  370         if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
  371             (nameiop == DELETE || nameiop == RENAME))
  372                 return (EROFS);
  373 
  374         /*
  375          * We now have a segment name to search for, and a directory to search.
  376          *
  377          * Before tediously performing a linear scan of the directory,
  378          * check the name cache to see if the directory/name pair
  379          * we are looking for is known already.
  380          */
  381         if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
  382             cnp->cn_nameiop, cnp->cn_flags, &iswhiteout, vpp)) {
  383                 if (iswhiteout) {
  384                         cnp->cn_flags |= ISWHITEOUT;
  385                 }
  386                 return *vpp == NULLVP ? ENOENT : 0;
  387         }
  388 
  389         /* May need to restart the lookup with an exclusive lock. */
  390         if (VOP_ISLOCKED(vdp) != LK_EXCLUSIVE) {
  391                 return ENOLCK;
  392         }
  393 
  394         /*
  395          * Produce the auxiliary lookup results into i_crap. Increment
  396          * its serial number so elsewhere we can tell if we're using
  397          * stale results. This should not be done this way. XXX.
  398          */
  399         results = &dp->i_crap;
  400         dp->i_crapcounter++;
  401 
  402         if (iswhiteout) {
  403                 /*
  404                  * The namecache set iswhiteout without finding a
  405                  * cache entry. As of this writing (20121014), this
  406                  * can happen if there was a whiteout entry that has
  407                  * been invalidated by the lookup. It is not clear if
  408                  * it is correct to set ISWHITEOUT in this case or
  409                  * not; however, doing so retains the prior behavior,
  410                  * so we'll go with that until some clearer answer
  411                  * appears. XXX
  412                  */
  413                 cnp->cn_flags |= ISWHITEOUT;
  414         }
  415 
  416         /*
  417          * Suppress search for slots unless creating
  418          * file and at end of pathname, in which case
  419          * we watch for a place to put the new file in
  420          * case it doesn't already exist.
  421          */
  422         slot_init(&slot);
  423 
  424         if ((nameiop == CREATE || nameiop == RENAME) && (flags & ISLASTCN)) {
  425                 slot.status = NONE;
  426                 slot.needed = UFS_DIRECTSIZ(cnp->cn_namelen);
  427         }
  428 
  429         /*
  430          * If there is cached information on a previous search of
  431          * this directory, pick up where we last left off.
  432          * We cache only lookups as these are the most common
  433          * and have the greatest payoff. Caching CREATE has little
  434          * benefit as it usually must search the entire directory
  435          * to determine that the entry does not exist. Caching the
  436          * location of the last DELETE or RENAME has not reduced
  437          * profiling time and hence has been removed in the interest
  438          * of simplicity.
  439          */
  440         bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
  441 
  442 #ifdef UFS_DIRHASH
  443         /*
  444          * Use dirhash for fast operations on large directories. The logic
  445          * to determine whether to hash the directory is contained within
  446          * ufsdirhash_build(); a zero return means that it decided to hash
  447          * this directory and it successfully built up the hash table.
  448          */
  449         if (ufsdirhash_build(dp) == 0) {
  450                 /* Look for a free slot if needed. */
  451                 enduseful = slot_findfree(&slot, dp);
  452                 /* Look up the component. */
  453                 numdirpasses = 1;
  454                 entryoffsetinblock = 0; /* silence compiler warning */
  455                 switch (ufsdirhash_lookup(dp, cnp->cn_nameptr, cnp->cn_namelen,
  456                     &results->ulr_offset, &bp,
  457                     nameiop == DELETE ? &prevoff : NULL)) {
  458                 case 0:
  459                         ep = (void *)((char *)bp->b_data +
  460                             (results->ulr_offset & bmask));
  461                         reclen = ufs_rw16(ep->d_reclen, needswap);
  462                         goto foundentry;
  463                 case ENOENT:
  464                         results->ulr_offset = roundup(dp->i_size, dirblksiz);
  465                         goto notfound;
  466                 default:
  467                         /* Something failed; just do a linear search. */
  468                         break;
  469                 }
  470         }
  471 #endif /* UFS_DIRHASH */
  472 
  473         if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
  474             results->ulr_diroff >= dp->i_size) {
  475                 entryoffsetinblock = 0;
  476                 results->ulr_offset = 0;
  477                 numdirpasses = 1;
  478         } else {
  479                 results->ulr_offset = results->ulr_diroff;
  480                 entryoffsetinblock = results->ulr_offset & bmask;
  481                 if (entryoffsetinblock != 0 &&
  482                     (error = ufs_blkatoff(vdp, (off_t)results->ulr_offset,
  483                     NULL, &bp, false)))
  484                         goto out;
  485                 numdirpasses = 2;
  486                 namecache_count_2passes();
  487         }
  488         prevoff = results->ulr_offset;
  489         endsearch = roundup(dp->i_size, dirblksiz);
  490         enduseful = 0;
  491 
  492 searchloop:
  493         while (results->ulr_offset < endsearch) {
  494                 preempt_point();
  495 
  496                 /*
  497                  * If necessary, get the next directory block.
  498                  */
  499                 if ((results->ulr_offset & bmask) == 0) {
  500                         if (bp != NULL)
  501                                 brelse(bp, 0);
  502                         error = ufs_blkatoff(vdp, (off_t)results->ulr_offset,
  503                             NULL, &bp, false);
  504                         if (error)
  505                                 goto out;
  506                         entryoffsetinblock = 0;
  507                 }
  508                 /*
  509                  * If still looking for a slot, and at a DIRBLKSIZ
  510                  * boundary, have to start looking for free space again.
  511                  */
  512                 if (slot.status == NONE &&
  513                     (entryoffsetinblock & (dirblksiz - 1)) == 0) {
  514                         slot.offset = -1;
  515                         slot.freespace = 0;
  516                 }
  517                 /*
  518                  * Get pointer to next entry.
  519                  * Full validation checks are slow, so we only check
  520                  * enough to insure forward progress through the
  521                  * directory. Complete checks can be run by patching
  522                  * "dirchk" to be true.
  523                  */
  524                 KASSERT(bp != NULL);
  525                 ep = (void *)((char *)bp->b_data + entryoffsetinblock);
  526                 const char *msg;
  527                 reclen = ufs_rw16(ep->d_reclen, needswap);
  528                 if ((reclen == 0 && (msg = "null entry")) || (dirchk &&
  529                     (msg = ufs_dirbadentry(vdp, ep, entryoffsetinblock)))) {
  530                         ufs_dirbad(dp, results->ulr_offset, msg);
  531                         reclen = dirblksiz -
  532                             (entryoffsetinblock & (dirblksiz - 1));
  533                         goto next;
  534                 }
  535 
  536                 /*
  537                  * If an appropriate sized slot has not yet been found,
  538                  * check to see if one is available. Also accumulate space
  539                  * in the current block so that we can determine if
  540                  * compaction is viable.
  541                  */
  542                 if (slot.status != FOUND) {
  543                         int size = reclen;
  544                         if (ep->d_ino != 0)
  545                                 size -= UFS_DIRSIZ(fsfmt, ep, needswap);
  546                         if (size > 0)
  547                                 slot_update(&slot, size, reclen,
  548                                     results->ulr_offset);
  549                 }
  550 
  551                 if (ep->d_ino == 0)
  552                         goto next;
  553 
  554                 /*
  555                  * Check for a name match.
  556                  */
  557                 const uint16_t namlen = NAMLEN(fsfmt, needswap, ep);
  558                 if (namlen != cnp->cn_namelen ||
  559                     memcmp(cnp->cn_nameptr, ep->d_name, (size_t)namlen))
  560                         goto next;
  561 
  562 #ifdef UFS_DIRHASH
  563 foundentry:
  564 #endif
  565                 /*
  566                  * Save directory entry's inode number and
  567                  * reclen, and release directory buffer.
  568                  */
  569                 if (!fsfmt && ep->d_type == DT_WHT) {
  570                         slot_white(&slot, reclen, results);
  571                         /*
  572                          * This is used to set results->ulr_endoff, which may
  573                          * be used by ufs_direnter() as a length to truncate
  574                          * the directory to. Therefore, it must point past the
  575                          * end of the last non-empty directory entry. We don't
  576                          * know where that is in this case, so we effectively
  577                          * disable shrinking by using the existing size of the
  578                          * directory.
  579                          *
  580                          * Note that we wouldn't expect to shrink the
  581                          * directory while rewriting an existing entry anyway.
  582                          */
  583                         enduseful = endsearch;
  584                         cnp->cn_flags |= ISWHITEOUT;
  585                         numdirpasses--;
  586                         goto notfound;
  587                 }
  588                 foundino = ufs_rw32(ep->d_ino, needswap);
  589                 results->ulr_reclen = reclen;
  590                 goto found;
  591 next:
  592                 prevoff = results->ulr_offset;
  593                 results->ulr_offset += reclen;
  594                 entryoffsetinblock += reclen;
  595                 if (ep->d_ino)
  596                         enduseful = results->ulr_offset;
  597         }
  598 notfound:
  599         /*
  600          * If we started in the middle of the directory and failed
  601          * to find our target, we must check the beginning as well.
  602          */
  603         if (numdirpasses == 2) {
  604                 numdirpasses--;
  605                 results->ulr_offset = 0;
  606                 endsearch = results->ulr_diroff;
  607                 goto searchloop;
  608         }
  609         if (bp != NULL)
  610                 brelse(bp, 0);
  611         /*
  612          * If creating, and at end of pathname and current
  613          * directory has not been removed, then can consider
  614          * allowing file to be created.
  615          */
  616         if ((nameiop == CREATE || nameiop == RENAME ||
  617              (nameiop == DELETE &&
  618               (cnp->cn_flags & DOWHITEOUT) &&
  619               (cnp->cn_flags & ISWHITEOUT))) &&
  620             (flags & ISLASTCN) && dp->i_nlink != 0) {
  621                 /*
  622                  * Access for write is interpreted as allowing
  623                  * creation of files in the directory.
  624                  */
  625                 if (flags & WILLBEDIR)
  626                         error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred);
  627                 else
  628                         error = VOP_ACCESS(vdp, VWRITE, cred);
  629                 if (error)
  630                         goto out;
  631                 error = slot_estimate(&slot, dirblksiz, nameiop,
  632                     prevoff, enduseful, dp, results);
  633                 /*
  634                  * We return with the directory locked, so that
  635                  * the parameters we set up above will still be
  636                  * valid if we actually decide to do a direnter().
  637                  * We return ni_vp == NULL to indicate that the entry
  638                  * does not currently exist; we leave a pointer to
  639                  * the (locked) directory inode in ndp->ni_dvp.
  640                  *
  641                  * NB - if the directory is unlocked, then this
  642                  * information cannot be used.
  643                  */
  644                 goto out;
  645         }
  646         /*
  647          * Insert name into cache (as non-existent) if appropriate.
  648          */
  649         if (nameiop != CREATE) {
  650                 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
  651                             cnp->cn_flags);
  652         }
  653         error = ENOENT;
  654         goto out;
  655 
  656 found:
  657         if (numdirpasses == 2)
  658                 namecache_count_pass2();
  659         /*
  660          * Check that directory length properly reflects presence
  661          * of this entry.
  662          */
  663         const uint64_t newisize =
  664             results->ulr_offset + UFS_DIRSIZ(fsfmt, ep, needswap);
  665         if (newisize > dp->i_size) {
  666                 ufs_dirbad(dp, results->ulr_offset, "i_size too small");
  667                 dp->i_size = newisize;
  668                 DIP_ASSIGN(dp, size, dp->i_size);
  669                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
  670                 UFS_WAPBL_UPDATE(vdp, NULL, NULL, UPDATE_DIROP);
  671         }
  672         brelse(bp, 0);
  673 
  674         /*
  675          * Found component in pathname.
  676          * If the final component of path name, save information
  677          * in the cache as to where the entry was found.
  678          */
  679         if ((flags & ISLASTCN) && nameiop == LOOKUP)
  680                 results->ulr_diroff = results->ulr_offset & ~(dirblksiz - 1);
  681 
  682         /*
  683          * If deleting, and at end of pathname, return
  684          * parameters which can be used to remove file.
  685          * Lock the inode, being careful with ".".
  686          */
  687         if (nameiop == DELETE && (flags & ISLASTCN)) {
  688                 /*
  689                  * Return pointer to current entry in results->ulr_offset,
  690                  * and distance past previous entry (if there
  691                  * is a previous entry in this block) in results->ulr_count.
  692                  * Save directory inode pointer in ndp->ni_dvp for dirremove().
  693                  */
  694                 calc_count(results, dirblksiz, prevoff);
  695 
  696                 if ((error = ufs_getino(vdp, dp, foundino, &tdp, false)) != 0)
  697                         goto out;
  698 
  699                 if ((error = ufs_can_delete(tdp, vdp, dp, cred)) != 0)
  700                         goto out;
  701 
  702                 *vpp = tdp;
  703                 goto out;
  704         }
  705 
  706         /*
  707          * If rewriting (RENAME), return the inode and the
  708          * information required to rewrite the present directory
  709          * Must get inode of directory entry to verify it's a
  710          * regular file, or empty directory.
  711          */
  712         if (nameiop == RENAME && (flags & ISLASTCN)) {
  713                 if (flags & WILLBEDIR)
  714                         error = VOP_ACCESSX(vdp, VWRITE | VAPPEND, cred);
  715                 else
  716                         error = VOP_ACCESS(vdp, VWRITE, cred);
  717                 if (error)
  718                         goto out;
  719                 /*
  720                  * Careful about locking second inode.
  721                  * This can only occur if the target is ".".
  722                  */
  723                 if ((error = ufs_getino(vdp, dp, foundino, &tdp, true)) != 0)
  724                         goto out;
  725                 *vpp = tdp;
  726                 goto out;
  727         }
  728 
  729         if ((error = ufs_getino(vdp, dp, foundino, &tdp, false)) != 0)
  730                 goto out;
  731 
  732         *vpp = tdp;
  733         /*
  734          * Insert name into cache if appropriate.
  735          */
  736         cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
  737         error = 0;
  738 
  739 out:
  740         return error;
  741 }
  742 
  743 void
  744 ufs_dirbad(struct inode *ip, doff_t offset, const char *how)
  745 {
  746         struct mount *mp = ITOV(ip)->v_mount;
  747         void (*p)(const char  *, ...) __printflike(1, 2) =
  748             (mp->mnt_flag & MNT_RDONLY) == 0 ? panic : printf;
  749 
  750         (*p)("%s: bad dir ino %ju at offset %d: %s\n",
  751             mp->mnt_stat.f_mntonname, (uintmax_t)ip->i_number,
  752             offset, how);
  753 }
  754 
  755 /*
  756  * Do consistency checking on a directory entry:
  757  *      record length must be multiple of 4
  758  *      entry must fit in rest of its DIRBLKSIZ block
  759  *      record must be large enough to contain entry
  760  *      name is not longer than FFS_MAXNAMLEN
  761  *      name must be as long as advertised, and null terminated
  762  */
  763 const char *
  764 ufs_dirbadentry(const struct vnode *dp, const struct direct *ep,
  765     int entryoffsetinblock)
  766 {
  767         const struct ufsmount *ump = VFSTOUFS(dp->v_mount);
  768         const int needswap = UFS_MPNEEDSWAP(ump);
  769         const int dirblksiz = ump->um_dirblksiz;
  770         const int maxsize = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
  771         const int fsfmt = FSFMT(dp);
  772         const uint8_t namlen = NAMLEN(fsfmt, needswap, ep);
  773         const uint16_t reclen = ufs_rw16(ep->d_reclen, needswap);
  774         const int dirsiz = (int)UFS_DIRSIZ(fsfmt, ep, needswap);
  775         const char *name = ep->d_name;
  776         const char *str;
  777 #ifdef DIAGNOSTIC
  778         static char buf[512];
  779 #endif
  780 
  781         if ((reclen & 0x3) != 0)
  782                 str = "not rounded";
  783         else if (reclen > maxsize)
  784                 str = "too big";
  785         else if (reclen < dirsiz)
  786                 str = "too small";
  787 #if FFS_MAXNAMLEN < 255
  788         else if (namlen > FFS_MAXNAMLEN)
  789                 str = "long name";
  790 #endif
  791         else
  792                 str = NULL;
  793 
  794         if (str) {
  795 #ifdef DIAGNOSTIC
  796                 snprintf(buf, sizeof(buf), "Bad dir (%s), reclen=%#x, "
  797                     "namlen=%d, dirsiz=%d <= reclen=%d <= maxsize=%d, "
  798                     "flags=%#x, entryoffsetinblock=%d, dirblksiz=%d",
  799                     str, reclen, namlen, dirsiz, reclen, maxsize,
  800                     dp->v_mount->mnt_flag, entryoffsetinblock, dirblksiz);
  801                 str = buf;
  802 #endif
  803                 return str;
  804         }
  805 
  806         if (ep->d_ino == 0)
  807                 return NULL;
  808 
  809         for (uint8_t i = 0; i < namlen; i++)
  810                 if (name[i] == '\0') {
  811                         str = "NUL in name";
  812 #ifdef DIAGNOSTIC
  813                         snprintf(buf, sizeof(buf), "%s [%s] i=%d, namlen=%d",
  814                             str, name, i, namlen);
  815                         str = buf;
  816 #endif
  817                         return str;
  818                 }
  819 
  820         if (name[namlen]) {
  821                 str = "missing NUL in name";
  822 #ifdef DIAGNOSTIC
  823                 snprintf(buf, sizeof(buf), "%s [%*.*s] namlen=%d", str, 
  824                     namlen, namlen, name, namlen);
  825                 str = buf;
  826 #endif
  827                 return str;
  828         }
  829         return NULL;
  830 }
  831 
  832 /*
  833  * Construct a new directory entry after a call to namei, using the
  834  * name in the componentname argument cnp. The argument ip is the
  835  * inode to which the new directory entry will refer.
  836  */
  837 void
  838 ufs_makedirentry(struct inode *ip, struct componentname *cnp,
  839     struct direct *newdirp)
  840 {
  841         size_t namelen = cnp->cn_namelen;
  842 
  843         newdirp->d_ino = ip->i_number;
  844         newdirp->d_namlen = namelen;
  845         memcpy(newdirp->d_name, cnp->cn_nameptr, namelen);
  846 
  847         /* NUL terminate and zero out padding */
  848         memset(&newdirp->d_name[namelen], 0, UFS_NAMEPAD(namelen));
  849 
  850         if (FSFMT(ITOV(ip)))
  851                 newdirp->d_type = 0;
  852         else
  853                 newdirp->d_type = IFTODT(ip->i_mode);
  854 }
  855 
  856 
  857 static int
  858 ufs_dirgrow(struct vnode *dvp, const struct ufs_lookup_results *ulr,
  859     struct vnode *tvp, struct direct *dirp,
  860     struct componentname *cnp, struct buf *newdirbp)
  861 {
  862         const kauth_cred_t cr = cnp->cn_cred;
  863         const struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
  864         const int needswap = UFS_MPNEEDSWAP(ump);
  865         const int dirblksiz = ump->um_dirblksiz;
  866         const int fsfmt = FSFMT(dvp);
  867         const u_int newentrysize = UFS_DIRSIZ(0, dirp, 0);
  868         struct inode *dp = VTOI(dvp);
  869         int error, ret, blkoff;
  870         struct timespec ts;
  871         struct buf *bp;
  872 
  873         /*
  874          * If ulr_count is 0, then namei could find no
  875          * space in the directory. Here, ulr_offset will
  876          * be on a directory block boundary and we will write the
  877          * new entry into a fresh block.
  878          */
  879         if (ulr->ulr_offset & (dirblksiz - 1))
  880                 panic("%s: newblk", __func__);
  881         if ((error = UFS_BALLOC(dvp, (off_t)ulr->ulr_offset, dirblksiz,
  882             cr, B_CLRBUF | B_SYNC, &bp)) != 0) {
  883                 return error;
  884         }
  885 
  886         dp->i_size = ulr->ulr_offset + dirblksiz;
  887         DIP_ASSIGN(dp, size, dp->i_size);
  888         dp->i_flag |= IN_CHANGE | IN_UPDATE;
  889         uvm_vnp_setsize(dvp, dp->i_size);
  890         dirp->d_reclen = ufs_rw16(dirblksiz, needswap);
  891         dirp->d_ino = ufs_rw32(dirp->d_ino, needswap);
  892         if (fsfmt && ENDIANSWAP(needswap))
  893                 ufs_dirswap(dirp);
  894         blkoff = ulr->ulr_offset & (ump->um_mountp->mnt_stat.f_iosize - 1);
  895         memcpy((char *)bp->b_data + blkoff, dirp, newentrysize);
  896 #ifdef UFS_DIRHASH
  897         if (dp->i_dirhash != NULL) {
  898                 ufsdirhash_newblk(dp, ulr->ulr_offset);
  899                 ufsdirhash_add(dp, dirp, ulr->ulr_offset);
  900                 ufsdirhash_checkblock(dp, (char *)bp->b_data + blkoff,
  901                     ulr->ulr_offset);
  902         }
  903 #endif
  904         error = VOP_BWRITE(bp->b_vp, bp);
  905         vfs_timestamp(&ts);
  906         ret = UFS_UPDATE(dvp, &ts, &ts, UPDATE_DIROP);
  907         if (error == 0)
  908                 return ret;
  909         return error;
  910 }
  911 
  912 static int
  913 #if __GNUC_PREREQ__(5, 3)
  914 /* This gets miscompiled by gcc 5.3 PR/51094 */
  915 __attribute__((__optimize__("no-tree-vrp")))
  916 #endif
  917 ufs_dircompact(struct vnode *dvp, const struct ufs_lookup_results *ulr,
  918     struct vnode *tvp, struct direct *dirp,
  919     struct componentname *cnp, struct buf *newdirbp)
  920 {
  921         const struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
  922         const int needswap = UFS_MPNEEDSWAP(ump);
  923         const int fsfmt = FSFMT(dvp);
  924         const u_int newentrysize = UFS_DIRSIZ(0, dirp, 0);
  925         struct inode *dp = VTOI(dvp);
  926         struct buf *bp;
  927         u_int dsize;
  928         struct direct *ep, *nep;
  929         int error, loc, spacefree;
  930         char *dirbuf;
  931         uint16_t reclen;
  932 
  933         UFS_WAPBL_JLOCK_ASSERT(dvp->v_mount);
  934 
  935         /*
  936          * If ulr_count is non-zero, then namei found space for the new
  937          * entry in the range ulr_offset to ulr_offset + ulr_count
  938          * in the directory. To use this space, we may have to compact
  939          * the entries located there, by copying them together towards the
  940          * beginning of the block, leaving the free space in one usable
  941          * chunk at the end.
  942          */
  943 
  944         /*
  945          * Increase size of directory if entry eats into new space.
  946          * This should never push the size past a new multiple of
  947          * DIRBLKSIZ.
  948          *
  949          * N.B. - THIS IS AN ARTIFACT OF 4.2 AND SHOULD NEVER HAPPEN.
  950          */
  951         if (ulr->ulr_offset + ulr->ulr_count > dp->i_size) {
  952 #ifdef DIAGNOSTIC
  953                 printf("%s: reached 4.2-only block, not supposed to happen\n",
  954                     __func__);
  955 #endif
  956                 dp->i_size = ulr->ulr_offset + ulr->ulr_count;
  957                 DIP_ASSIGN(dp, size, dp->i_size);
  958                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
  959                 UFS_WAPBL_UPDATE(dvp, NULL, NULL, UPDATE_DIROP);
  960         }
  961         /*
  962          * Get the block containing the space for the new directory entry.
  963          */
  964         error = ufs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp, true);
  965         if (error)
  966                 return error;
  967 
  968         /*
  969          * Find space for the new entry. In the simple case, the entry at
  970          * offset base will have the space. If it does not, then namei
  971          * arranged that compacting the region ulr_offset to
  972          * ulr_offset + ulr_count would yield the space.
  973          */
  974         ep = (void *)dirbuf;
  975         dsize = (ep->d_ino != 0) ? UFS_DIRSIZ(fsfmt, ep, needswap) : 0;
  976         reclen = ufs_rw16(ep->d_reclen, needswap);
  977         spacefree = reclen - dsize;
  978         for (loc = reclen; loc < ulr->ulr_count; ) {
  979                 nep = (void *)(dirbuf + loc);
  980 
  981                 /* Trim the existing slot (NB: dsize may be zero). */
  982                 ep->d_reclen = ufs_rw16(dsize, needswap);
  983                 ep = (void *)((char *)ep + dsize);
  984 
  985                 reclen = ufs_rw16(nep->d_reclen, needswap);
  986                 loc += reclen;
  987                 if (nep->d_ino == 0) {
  988                         /*
  989                          * A mid-block unused entry. Such entries are
  990                          * never created by the kernel, but fsck_ffs
  991                          * can create them (and it doesn't fix them).
  992                          *
  993                          * Add up the free space, and initialise the
  994                          * relocated entry since we don't memcpy it.
  995                          */
  996                         spacefree += reclen;
  997                         ep->d_ino = 0;
  998                         dsize = 0;
  999                         continue;
 1000                 }
 1001                 dsize = UFS_DIRSIZ(fsfmt, nep, needswap);
 1002                 spacefree += reclen - dsize;
 1003 #ifdef UFS_DIRHASH
 1004                 if (dp->i_dirhash != NULL)
 1005                         ufsdirhash_move(dp, nep,
 1006                             ulr->ulr_offset + ((char *)nep - dirbuf),
 1007                             ulr->ulr_offset + ((char *)ep - dirbuf));
 1008 #endif
 1009                 memcpy(ep, nep, dsize);
 1010         }
 1011         /*
 1012          * Here, `ep' points to a directory entry containing `dsize' in-use
 1013          * bytes followed by `spacefree' unused bytes. If ep->d_ino == 0,
 1014          * then the entry is completely unused (dsize == 0). The value
 1015          * of ep->d_reclen is always indeterminate.
 1016          *
 1017          * Update the pointer fields in the previous entry (if any),
 1018          * copy in the new entry, and write out the block.
 1019          */
 1020         if (ep->d_ino == 0 ||
 1021             (ufs_rw32(ep->d_ino, needswap) == UFS_WINO &&
 1022              memcmp(ep->d_name, dirp->d_name, dirp->d_namlen) == 0)) {
 1023                 if (spacefree + dsize < newentrysize)
 1024                         panic("%s: too big", __func__);
 1025                 dirp->d_reclen = spacefree + dsize;
 1026         } else {
 1027                 if (spacefree < newentrysize)
 1028                         panic("%s: nospace", __func__);
 1029                 dirp->d_reclen = spacefree;
 1030                 ep->d_reclen = ufs_rw16(dsize, needswap);
 1031                 ep = (void *)((char *)ep + dsize);
 1032         }
 1033 
 1034         dirp->d_reclen = ufs_rw16(dirp->d_reclen, needswap);
 1035         dirp->d_ino = ufs_rw32(dirp->d_ino, needswap);
 1036         if (fsfmt && ENDIANSWAP(needswap))
 1037                 ufs_dirswap(dirp);
 1038 #ifdef UFS_DIRHASH
 1039         if (dp->i_dirhash != NULL && (ep->d_ino == 0 ||
 1040             dirp->d_reclen == spacefree))
 1041                 ufsdirhash_add(dp, dirp, ulr->ulr_offset + ((char *)ep - dirbuf));
 1042 #endif
 1043         memcpy(ep, dirp, newentrysize);
 1044 #ifdef UFS_DIRHASH
 1045         if (dp->i_dirhash != NULL) {
 1046                 const int dirblkmsk = ump->um_dirblksiz - 1;
 1047                 ufsdirhash_checkblock(dp, dirbuf -
 1048                     (ulr->ulr_offset & dirblkmsk),
 1049                     ulr->ulr_offset & ~dirblkmsk);
 1050         }
 1051 #endif
 1052         error = VOP_BWRITE(bp->b_vp, bp);
 1053         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 1054         /*
 1055          * If all went well, and the directory can be shortened, proceed
 1056          * with the truncation. Note that we have to unlock the inode for
 1057          * the entry that we just entered, as the truncation may need to
 1058          * lock other inodes which can lead to deadlock if we also hold a
 1059          * lock on the newly entered node.
 1060          */
 1061         if (error == 0 && ulr->ulr_endoff && ulr->ulr_endoff < dp->i_size) {
 1062                 const kauth_cred_t cr = cnp->cn_cred;
 1063 #ifdef UFS_DIRHASH
 1064                 if (dp->i_dirhash != NULL)
 1065                         ufsdirhash_dirtrunc(dp, ulr->ulr_endoff);
 1066 #endif
 1067                 (void) UFS_TRUNCATE(dvp, (off_t)ulr->ulr_endoff, IO_SYNC, cr);
 1068         }
 1069         UFS_WAPBL_UPDATE(dvp, NULL, NULL, UPDATE_DIROP);
 1070         return error;
 1071 }
 1072 
 1073 /*
 1074  * Write a directory entry after a call to namei, using the parameters
 1075  * that ufs_lookup left in nameidata and in the ufs_lookup_results.
 1076  *
 1077  * DVP is the directory to be updated. It must be locked.
 1078  * ULR is the ufs_lookup_results structure from the final lookup step.
 1079  * TVP is not used. (XXX: why is it here? remove it)
 1080  * DIRP is the new directory entry contents.
 1081  * CNP is the componentname from the final lookup step.
 1082  * NEWDIRBP is not used and (XXX) should be removed. The previous
 1083  * comment here said it was used by the now-removed softupdates code.
 1084  *
 1085  * The link count of the target inode is *not* incremented; the
 1086  * caller does that.
 1087  *
 1088  * If ulr->ulr_count is 0, ufs_lookup did not find space to insert the
 1089  * directory entry. ulr_offset, which is the place to put the entry,
 1090  * should be on a block boundary (and should be at the end of the
 1091  * directory AFAIK) and a fresh block is allocated to put the new
 1092  * directory entry in.
 1093  *
 1094  * If ulr->ulr_count is not zero, ufs_lookup found a slot to insert
 1095  * the entry into. This slot ranges from ulr_offset to ulr_offset +
 1096  * ulr_count. However, this slot may already be partially populated
 1097  * requiring compaction. See notes below.
 1098  *
 1099  * Furthermore, if ulr_count is not zero and ulr_endoff is not the
 1100  * same as i_size, the directory is truncated to size ulr_endoff.
 1101  */
 1102 int
 1103 ufs_direnter(struct vnode *dvp, const struct ufs_lookup_results *ulr,
 1104     struct vnode *tvp, struct direct *dirp,
 1105     struct componentname *cnp, struct buf *newdirbp)
 1106 {
 1107         if (ulr->ulr_count == 0)
 1108                 return ufs_dirgrow(dvp, ulr, tvp, dirp, cnp, newdirbp);
 1109         else
 1110                 return ufs_dircompact(dvp, ulr, tvp, dirp, cnp, newdirbp);
 1111 }
 1112 
 1113 /*
 1114  * Remove a directory entry after a call to namei, using the
 1115  * parameters that ufs_lookup left in nameidata and in the
 1116  * ufs_lookup_results.
 1117  *
 1118  * DVP is the directory to be updated. It must be locked.
 1119  * ULR is the ufs_lookup_results structure from the final lookup step.
 1120  * IP, if not null, is the inode being unlinked.
 1121  * FLAGS may contain DOWHITEOUT.
 1122  * ISRMDIR is not used and (XXX) should be removed.
 1123  *
 1124  * If FLAGS contains DOWHITEOUT the entry is replaced with a whiteout
 1125  * instead of being cleared.
 1126  *
 1127  * ulr->ulr_offset contains the position of the directory entry
 1128  * to be removed.
 1129  *
 1130  * ulr->ulr_reclen contains the size of the directory entry to be
 1131  * removed.
 1132  *
 1133  * ulr->ulr_count contains the size of the *previous* directory
 1134  * entry. This allows finding it, for free space management. If
 1135  * ulr_count is 0, the target entry is at the beginning of the
 1136  * directory. (Does this ever happen? The first entry should be ".",
 1137  * which should only be removed at rmdir time. Does rmdir come here
 1138  * to clear out the "." and ".." entries? Perhaps, but I doubt it.)
 1139  *
 1140  * The space is marked free by adding it to the record length (not
 1141  * name length) of the preceding entry. If the first entry becomes
 1142  * free, it is marked free by setting the inode number to 0.
 1143  *
 1144  * The link count of IP is decremented. Note that this is not the
 1145  * inverse behavior of ufs_direnter, which does not adjust link
 1146  * counts. Sigh.
 1147  */
 1148 int
 1149 ufs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
 1150     struct inode *ip, int flags, int isrmdir)
 1151 {
 1152         struct inode *dp = VTOI(dvp);
 1153         struct direct *ep;
 1154         struct buf *bp;
 1155         int error;
 1156         const int needswap = UFS_MPNEEDSWAP(dp->i_ump);
 1157         uint16_t reclen;
 1158 
 1159         UFS_WAPBL_JLOCK_ASSERT(dvp->v_mount);
 1160 
 1161         if (flags & DOWHITEOUT) {
 1162                 /*
 1163                  * Whiteout entry: set d_ino to UFS_WINO.
 1164                  */
 1165                 error = ufs_blkatoff(dvp, (off_t)ulr->ulr_offset, &ep,
 1166                                      &bp, true);
 1167                 if (error)
 1168                         return (error);
 1169                 ep->d_ino = ufs_rw32(UFS_WINO, needswap);
 1170                 ep->d_type = DT_WHT;
 1171                 goto out;
 1172         }
 1173 
 1174         if ((error = ufs_blkatoff(dvp,
 1175             (off_t)(ulr->ulr_offset - ulr->ulr_count), &ep, &bp, true)) != 0)
 1176                 return (error);
 1177 
 1178         reclen = ufs_rw16(ep->d_reclen, needswap);
 1179 #ifdef UFS_DIRHASH
 1180         /*
 1181          * Remove the dirhash entry. This is complicated by the fact
 1182          * that `ep' is the previous entry when ulr_count != 0.
 1183          */
 1184         if (dp->i_dirhash != NULL)
 1185                 ufsdirhash_remove(dp, (ulr->ulr_count == 0) ? ep :
 1186                    (void *)((char *)ep + reclen), ulr->ulr_offset);
 1187 #endif
 1188 
 1189         if (ulr->ulr_count == 0) {
 1190                 /*
 1191                  * First entry in block: set d_ino to zero.
 1192                  */
 1193                 ep->d_ino = 0;
 1194         } else {
 1195                 /*
 1196                  * Collapse new free space into previous entry.
 1197                  */
 1198                 ep->d_reclen = ufs_rw16(reclen + ulr->ulr_reclen, needswap);
 1199         }
 1200 
 1201 #ifdef UFS_DIRHASH
 1202         if (dp->i_dirhash != NULL) {
 1203                 int dirblksiz = ip->i_ump->um_dirblksiz;
 1204                 ufsdirhash_checkblock(dp, (char *)ep -
 1205                     ((ulr->ulr_offset - ulr->ulr_count) & (dirblksiz - 1)),
 1206                     ulr->ulr_offset & ~(dirblksiz - 1));
 1207         }
 1208 #endif
 1209 
 1210 out:
 1211         if (ip) {
 1212                 ip->i_nlink--;
 1213                 DIP_ASSIGN(ip, nlink, ip->i_nlink);
 1214                 ip->i_flag |= IN_CHANGE;
 1215                 UFS_WAPBL_UPDATE(ITOV(ip), NULL, NULL, 0);
 1216         }
 1217         /*
 1218          * XXX did it ever occur to anyone that it might be a good
 1219          * idea to restore ip->i_nlink if this fails? Or something?
 1220          * Currently on error return from this function the state of
 1221          * ip->i_nlink depends on what happened, and callers
 1222          * definitely do not take this into account.
 1223          */
 1224         error = VOP_BWRITE(bp->b_vp, bp);
 1225         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 1226         /*
 1227          * If the last named reference to a snapshot goes away,
 1228          * drop its snapshot reference so that it will be reclaimed
 1229          * when last open reference goes away.
 1230          */
 1231         if (ip != 0 && (ip->i_flags & SF_SNAPSHOT) != 0 &&
 1232             ip->i_nlink == 0)
 1233                 UFS_SNAPGONE(ITOV(ip));
 1234         UFS_WAPBL_UPDATE(dvp, NULL, NULL, 0);
 1235         return (error);
 1236 }
 1237 
 1238 /*
 1239  * Rewrite an existing directory entry to point at the inode supplied.
 1240  *
 1241  * DP is the directory to update.
 1242  * OFFSET is the position of the entry in question. It may come
 1243  * from ulr_offset of a ufs_lookup_results.
 1244  * OIP is the old inode the directory previously pointed to.
 1245  * NEWINUM is the number of the new inode.
 1246  * NEWTYPE is the new value for the type field of the directory entry.
 1247  * (This is ignored if the fs doesn't support that.)
 1248  * ISRMDIR is not used and (XXX) should be removed.
 1249  * IFLAGS are added to DP's inode flags.
 1250  *
 1251  * The link count of OIP is decremented. Note that the link count of
 1252  * the new inode is *not* incremented. Yay for symmetry.
 1253  */
 1254 int
 1255 ufs_dirrewrite(struct inode *dp, off_t offset,
 1256     struct inode *oip, ino_t newinum, int newtype,
 1257     int isrmdir, int iflags)
 1258 {
 1259         struct buf *bp;
 1260         struct direct *ep;
 1261         struct vnode *vdp = ITOV(dp);
 1262         int error;
 1263 
 1264         error = ufs_blkatoff(vdp, offset, &ep, &bp, true);
 1265         if (error)
 1266                 return (error);
 1267         ep->d_ino = ufs_rw32(newinum, UFS_MPNEEDSWAP(dp->i_ump));
 1268         if (!FSFMT(vdp))
 1269                 ep->d_type = newtype;
 1270         oip->i_nlink--;
 1271         DIP_ASSIGN(oip, nlink, oip->i_nlink);
 1272         oip->i_flag |= IN_CHANGE;
 1273         UFS_WAPBL_UPDATE(ITOV(oip), NULL, NULL, UPDATE_DIROP);
 1274         error = VOP_BWRITE(bp->b_vp, bp);
 1275         dp->i_flag |= iflags;
 1276         /*
 1277          * If the last named reference to a snapshot goes away,
 1278          * drop its snapshot reference so that it will be reclaimed
 1279          * when last open reference goes away.
 1280          */
 1281         if ((oip->i_flags & SF_SNAPSHOT) != 0 && oip->i_nlink == 0)
 1282                 UFS_SNAPGONE(ITOV(oip));
 1283         UFS_WAPBL_UPDATE(vdp, NULL, NULL, UPDATE_DIROP);
 1284         return (error);
 1285 }
 1286 
 1287 /*
 1288  * Check if a directory is empty or not.
 1289  * Inode supplied must be locked.
 1290  *
 1291  * Using a struct dirtemplate here is not precisely
 1292  * what we want, but better than using a struct direct.
 1293  *
 1294  * NB: does not handle corrupted directories.
 1295  */
 1296 int
 1297 ufs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
 1298 {
 1299         doff_t off;
 1300         struct dirtemplate dbuf;
 1301         struct direct *dp = (void *)&dbuf;
 1302         int error;
 1303         size_t count;
 1304         const int needswap = UFS_IPNEEDSWAP(ip);
 1305         const int fsfmt = FSFMT(ITOV(ip));
 1306 #define MINDIRSIZ (sizeof (struct dirtemplate) / 2)
 1307 
 1308         for (off = 0; off < ip->i_size;
 1309             off += ufs_rw16(dp->d_reclen, needswap)) {
 1310                 error = ufs_bufio(UIO_READ, ITOV(ip), dp, MINDIRSIZ,
 1311                     off, IO_NODELOCKED, cred, &count, NULL);
 1312                 /*
 1313                  * Since we read MINDIRSIZ, residual must
 1314                  * be 0 unless we're at end of file.
 1315                  */
 1316                 if (error || count != 0)
 1317                         return (0);
 1318                 /* avoid infinite loops */
 1319                 if (dp->d_reclen == 0)
 1320                         return (0);
 1321                 /* skip empty entries */
 1322                 ino_t ino = ufs_rw32(dp->d_ino, needswap);
 1323                 if (ino == 0 || ino == UFS_WINO)
 1324                         continue;
 1325                 /* accept only "." and ".." */
 1326                 const uint8_t namlen = NAMLEN(fsfmt, needswap, dp);
 1327                 if (namlen > 2)
 1328                         return (0);
 1329                 if (dp->d_name[0] != '.')
 1330                         return (0);
 1331                 /*
 1332                  * At this point namlen must be 1 or 2.
 1333                  * 1 implies ".", 2 implies ".." if second
 1334                  * char is also "."
 1335                  */
 1336                 if (namlen == 1 && ino == ip->i_number)
 1337                         continue;
 1338                 if (dp->d_name[1] == '.' && ino == parentino)
 1339                         continue;
 1340                 return (0);
 1341         }
 1342         return (1);
 1343 }
 1344 
 1345 #define UFS_DIRRABLKS 0
 1346 int ufs_dirrablks = UFS_DIRRABLKS;
 1347 
 1348 /*
 1349  * ufs_blkatoff: Return buffer with the contents of block "offset" from
 1350  * the beginning of directory "vp".  If "res" is non-NULL, fill it in with
 1351  * a pointer to the remaining space in the directory.  If the caller intends
 1352  * to modify the buffer returned, "modify" must be true.
 1353  */
 1354 
 1355 int
 1356 ufs_blkatoff(struct vnode *vp, off_t offset, void *v, struct buf **bpp,
 1357     bool modify)
 1358 {
 1359         char **res = v;
 1360         struct inode *ip __diagused;
 1361         struct buf *bp;
 1362         daddr_t lbn;
 1363         const int dirrablks = ufs_dirrablks;
 1364         daddr_t *blks;
 1365         int *blksizes;
 1366         int run, error;
 1367         struct mount *mp = vp->v_mount;
 1368         const int bshift = mp->mnt_fs_bshift;
 1369         const int bsize = 1 << bshift;
 1370         off_t eof;
 1371 
 1372         blks = kmem_alloc((1 + dirrablks) * sizeof(daddr_t), KM_SLEEP);
 1373         blksizes = kmem_alloc((1 + dirrablks) * sizeof(int), KM_SLEEP);
 1374         ip = VTOI(vp);
 1375         KASSERT(vp->v_size == ip->i_size);
 1376         GOP_SIZE(vp, vp->v_size, &eof, 0);
 1377         lbn = offset >> bshift;
 1378 
 1379         for (run = 0; run <= dirrablks;) {
 1380                 const off_t curoff = lbn << bshift;
 1381                 const int size = MIN(eof - curoff, bsize);
 1382 
 1383                 if (size == 0) {
 1384                         break;
 1385                 }
 1386                 KASSERT(curoff < eof);
 1387                 blks[run] = lbn;
 1388                 blksizes[run] = size;
 1389                 lbn++;
 1390                 run++;
 1391                 if (size != bsize) {
 1392                         break;
 1393                 }
 1394         }
 1395         KASSERT(run >= 1);
 1396         error = breadn(vp, blks[0], blksizes[0], &blks[1], &blksizes[1],
 1397             run - 1, (modify ? B_MODIFY : 0), &bp);
 1398         if (error != 0) {
 1399                 *bpp = NULL;
 1400                 goto out;
 1401         }
 1402         if (res) {
 1403                 *res = (char *)bp->b_data + (offset & (bsize - 1));
 1404         }
 1405         *bpp = bp;
 1406 
 1407  out:
 1408         kmem_free(blks, (1 + dirrablks) * sizeof(daddr_t));
 1409         kmem_free(blksizes, (1 + dirrablks) * sizeof(int));
 1410         return error;
 1411 }

Cache object: d07256c31d95d0c97b68643c98a16473


[ 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.