sys/ufs/ext2fs/ext2fs_lookup.c

   1 /*      $NetBSD: ext2fs_lookup.c,v 1.73 2013/01/22 09:39:15 dholland Exp $      */
   2
   3 /*
   4  * Modified for NetBSD 1.2E
   5  * May 1997, Manuel Bouyer
   6  * Laboratoire d'informatique de Paris VI
   7  */
   8 /*
   9  *  modified for Lites 1.1
  10  *
  11  *  Aug 1995, Godmar Back (gback@cs.utah.edu)
  12  *  University of Utah, Department of Computer Science
  13  */
  14 /*
  15  * Copyright (c) 1989, 1993
  16  *      The Regents of the University of California.  All rights reserved.
  17  * (c) UNIX System Laboratories, Inc.
  18  * All or some portions of this file are derived from material licensed
  19  * to the University of California by American Telephone and Telegraph
  20  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  21  * the permission of UNIX System Laboratories, Inc.
  22  *
  23  * Redistribution and use in source and binary forms, with or without
  24  * modification, are permitted provided that the following conditions
  25  * are met:
  26  * 1. Redistributions of source code must retain the above copyright
  27  *    notice, this list of conditions and the following disclaimer.
  28  * 2. Redistributions in binary form must reproduce the above copyright
  29  *    notice, this list of conditions and the following disclaimer in the
  30  *    documentation and/or other materials provided with the distribution.
  31  * 3. Neither the name of the University nor the names of its contributors
  32  *    may be used to endorse or promote products derived from this software
  33  *    without specific prior written permission.
  34  *
  35  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  36  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  38  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  39  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  40  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  41  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  42  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  43  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  44  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  45  * SUCH DAMAGE.
  46  *
  47  *      @(#)ufs_lookup.c        8.6 (Berkeley) 4/1/94
  48  */
  49
  50 #include <sys/cdefs.h>
  51 __KERNEL_RCSID(0, "$NetBSD: ext2fs_lookup.c,v 1.73 2013/01/22 09:39:15 dholland Exp $");
  52
  53 #include <sys/param.h>
  54 #include <sys/systm.h>
  55 #include <sys/namei.h>
  56 #include <sys/buf.h>
  57 #include <sys/file.h>
  58 #include <sys/mount.h>
  59 #include <sys/vnode.h>
  60 #include <sys/kmem.h>
  61 #include <sys/malloc.h>
  62 #include <sys/dirent.h>
  63 #include <sys/kauth.h>
  64 #include <sys/proc.h>
  65
  66 #include <ufs/ufs/inode.h>
  67 #include <ufs/ufs/ufsmount.h>
  68 #include <ufs/ufs/ufs_extern.h>
  69
  70 #include <ufs/ext2fs/ext2fs_extern.h>
  71 #include <ufs/ext2fs/ext2fs_dir.h>
  72 #include <ufs/ext2fs/ext2fs.h>
  73
  74 #include <miscfs/genfs/genfs.h>
  75
  76 extern  int dirchk;
  77
  78 static void     ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir,
  79                                           struct dirent *ffsdir);
  80 static int      ext2fs_dirbadentry(struct vnode *dp,
  81                                           struct ext2fs_direct *de,
  82                                           int entryoffsetinblock);
  83
  84 /*
  85  * the problem that is tackled below is the fact that FFS
  86  * includes the terminating zero on disk while EXT2FS doesn't
  87  * this implies that we need to introduce some padding.
  88  * For instance, a filename "sbin" has normally a reclen 12
  89  * in EXT2, but 16 in FFS.
  90  * This reminds me of that Pepsi commercial: 'Kid saved a lousy nine cents...'
  91  * If it wasn't for that, the complete ufs code for directories would
  92  * have worked w/o changes (except for the difference in DIRBLKSIZ)
  93  */
  94 static void
  95 ext2fs_dirconv2ffs(struct ext2fs_direct *e2dir, struct dirent *ffsdir)
  96 {
  97         memset(ffsdir, 0, sizeof(struct dirent));
  98         ffsdir->d_fileno = fs2h32(e2dir->e2d_ino);
  99         ffsdir->d_namlen = e2dir->e2d_namlen;
 100
 101         ffsdir->d_type = DT_UNKNOWN;            /* don't know more here */
 102 #ifdef DIAGNOSTIC
 103 #if MAXNAMLEN < E2FS_MAXNAMLEN
 104         /*
 105          * we should handle this more gracefully !
 106          */
 107         if (e2dir->e2d_namlen > MAXNAMLEN)
 108                 panic("ext2fs: e2dir->e2d_namlen");
 109 #endif
 110 #endif
 111         strncpy(ffsdir->d_name, e2dir->e2d_name, ffsdir->d_namlen);
 112
 113         /* Godmar thinks: since e2dir->e2d_reclen can be big and means
 114            nothing anyway, we compute our own reclen according to what
 115            we think is right
 116          */
 117         ffsdir->d_reclen = _DIRENT_SIZE(ffsdir);
 118 }
 119
 120 /*
 121  * Vnode op for reading directories.
 122  *
 123  * Convert the on-disk entries to <sys/dirent.h> entries.
 124  * the problem is that the conversion will blow up some entries by four bytes,
 125  * so it can't be done in place. This is too bad. Right now the conversion is
 126  * done entry by entry, the converted entry is sent via uiomove.
 127  *
 128  * XXX allocate a buffer, convert as many entries as possible, then send
 129  * the whole buffer to uiomove
 130  */
 131 int
 132 ext2fs_readdir(void *v)
 133 {
 134         struct vop_readdir_args /* {
 135                 struct vnode *a_vp;
 136                 struct uio *a_uio;
 137                 kauth_cred_t a_cred;
 138                 int **a_eofflag;
 139                 off_t **a_cookies;
 140                 int ncookies;
 141         } */ *ap = v;
 142         struct uio *uio = ap->a_uio;
 143         int error;
 144         size_t e2fs_count, readcnt;
 145         struct vnode *vp = ap->a_vp;
 146         struct m_ext2fs *fs = VTOI(vp)->i_e2fs;
 147
 148         struct ext2fs_direct *dp;
 149         struct dirent *dstd;
 150         struct uio auio;
 151         struct iovec aiov;
 152         void *dirbuf;
 153         off_t off = uio->uio_offset;
 154         off_t *cookies = NULL;
 155         int nc = 0, ncookies = 0;
 156         int e2d_reclen;
 157
 158         if (vp->v_type != VDIR)
 159                 return (ENOTDIR);
 160
 161         e2fs_count = uio->uio_resid;
 162         /* Make sure we don't return partial entries. */
 163         e2fs_count -= (uio->uio_offset + e2fs_count) & (fs->e2fs_bsize -1);
 164         if (e2fs_count <= 0)
 165                 return (EINVAL);
 166
 167         auio = *uio;
 168         auio.uio_iov = &aiov;
 169         auio.uio_iovcnt = 1;
 170         aiov.iov_len = e2fs_count;
 171         auio.uio_resid = e2fs_count;
 172         UIO_SETUP_SYSSPACE(&auio);
 173         dirbuf = kmem_alloc(e2fs_count, KM_SLEEP);
 174         dstd = kmem_zalloc(sizeof(struct dirent), KM_SLEEP);
 175         if (ap->a_ncookies) {
 176                 nc = e2fs_count / _DIRENT_MINSIZE((struct dirent *)0);
 177                 ncookies = nc;
 178                 cookies = malloc(sizeof (off_t) * ncookies, M_TEMP, M_WAITOK);
 179                 *ap->a_cookies = cookies;
 180         }
 181         aiov.iov_base = dirbuf;
 182
 183         error = VOP_READ(ap->a_vp, &auio, 0, ap->a_cred);
 184         if (error == 0) {
 185                 readcnt = e2fs_count - auio.uio_resid;
 186                 for (dp = (struct ext2fs_direct *)dirbuf;
 187                         (char *)dp < (char *)dirbuf + readcnt; ) {
 188                         e2d_reclen = fs2h16(dp->e2d_reclen);
 189                         if (e2d_reclen == 0) {
 190                                 error = EIO;
 191                                 break;
 192                         }
 193                         ext2fs_dirconv2ffs(dp, dstd);
 194                         if(dstd->d_reclen > uio->uio_resid) {
 195                                 break;
 196                         }
 197                         error = uiomove(dstd, dstd->d_reclen, uio);
 198                         if (error != 0) {
 199                                 break;
 200                         }
 201                         off = off + e2d_reclen;
 202                         if (cookies != NULL) {
 203                                 *cookies++ = off;
 204                                 if (--ncookies <= 0){
 205                                         break;  /* out of cookies */
 206                                 }
 207                         }
 208                         /* advance dp */
 209                         dp = (struct ext2fs_direct *) ((char *)dp + e2d_reclen);
 210                 }
 211                 /* we need to correct uio_offset */
 212                 uio->uio_offset = off;
 213         }
 214         kmem_free(dirbuf, e2fs_count);
 215         kmem_free(dstd, sizeof(*dstd));
 216         *ap->a_eofflag = ext2fs_size(VTOI(ap->a_vp)) <= uio->uio_offset;
 217         if (ap->a_ncookies) {
 218                 if (error) {
 219                         free(*ap->a_cookies, M_TEMP);
 220                         *ap->a_ncookies = 0;
 221                         *ap->a_cookies = NULL;
 222                 } else
 223                         *ap->a_ncookies = nc - ncookies;
 224         }
 225         return (error);
 226 }
 227
 228 /*
 229  * Convert a component of a pathname into a pointer to a locked inode.
 230  * This is a very central and rather complicated routine.
 231  * If the file system is not maintained in a strict tree hierarchy,
 232  * this can result in a deadlock situation (see comments in code below).
 233  *
 234  * The cnp->cn_nameiop argument is LOOKUP, CREATE, RENAME, or DELETE depending
 235  * on whether the name is to be looked up, created, renamed, or deleted.
 236  * When CREATE, RENAME, or DELETE is specified, information usable in
 237  * creating, renaming, or deleting a directory entry may be calculated.
 238  * If flag has LOCKPARENT or'ed into it and the target of the pathname
 239  * exists, lookup returns both the target and its parent directory locked.
 240  * When creating or renaming and LOCKPARENT is specified, the target may
 241  * not be ".".  When deleting and LOCKPARENT is specified, the target may
 242  * be "."., but the caller must check to ensure it does an vrele and vput
 243  * instead of two vputs.
 244  *
 245  * Overall outline of ext2fs_lookup:
 246  *
 247  *      check accessibility of directory
 248  *      look for name in cache, if found, then if at end of path
 249  *        and deleting or creating, drop it, else return name
 250  *      search for name in directory, to found or notfound
 251  * notfound:
 252  *      if creating, return locked directory, leaving info on available slots
 253  *      else return error
 254  * found:
 255  *      if at end of path and deleting, return information to allow delete
 256  *      if at end of path and rewriting (RENAME and LOCKPARENT), lock target
 257  *        inode and return info to allow rewrite
 258  *      if not at end, add name to cache; if at end and neither creating
 259  *        nor deleting, add name to cache
 260  */
 261 int
 262 ext2fs_lookup(void *v)
 263 {
 264         struct vop_lookup_args /* {
 265                 struct vnode *a_dvp;
 266                 struct vnode **a_vpp;
 267                 struct componentname *a_cnp;
 268         } */ *ap = v;
 269         struct vnode *vdp = ap->a_dvp;  /* vnode for directory being searched */
 270         struct inode *dp = VTOI(vdp);   /* inode for directory being searched */
 271         struct buf *bp;                 /* a buffer of directory entries */
 272         struct ext2fs_direct *ep;       /* the current directory entry */
 273         int entryoffsetinblock;         /* offset of ep in bp's buffer */
 274         enum {NONE, COMPACT, FOUND} slotstatus;
 275         doff_t slotoffset;              /* offset of area with free space */
 276         int slotsize;                   /* size of area at slotoffset */
 277         int slotfreespace;              /* amount of space free in slot */
 278         int slotneeded;                 /* size of the entry we're seeking */
 279         int numdirpasses;               /* strategy for directory search */
 280         doff_t endsearch;               /* offset to end directory search */
 281         doff_t prevoff;                 /* prev entry dp->i_offset */
 282         struct vnode *pdp;              /* saved dp during symlink work */
 283         struct vnode *tdp;              /* returned by VFS_VGET */
 284         doff_t enduseful;               /* pointer past last used dir slot */
 285         u_long bmask;                   /* block offset mask */
 286         int namlen, error;
 287         struct vnode **vpp = ap->a_vpp;
 288         struct componentname *cnp = ap->a_cnp;
 289         kauth_cred_t cred = cnp->cn_cred;
 290         int flags;
 291         int nameiop = cnp->cn_nameiop;
 292         struct ufsmount *ump = dp->i_ump;
 293         int dirblksiz = ump->um_dirblksiz;
 294         ino_t foundino;
 295         struct ufs_lookup_results *results;
 296
 297         flags = cnp->cn_flags;
 298
 299         bp = NULL;
 300         slotoffset = -1;
 301         *vpp = NULL;
 302
 303         /*
 304          * Produce the auxiliary lookup results into i_crap. Increment
 305          * its serial number so elsewhere we can tell if we're using
 306          * stale results. This should not be done this way. XXX.
 307          */
 308         results = &dp->i_crap;
 309         dp->i_crapcounter++;
 310
 311         /*
 312          * Check accessiblity of directory.
 313          */
 314         if ((error = VOP_ACCESS(vdp, VEXEC, cred)) != 0)
 315                 return (error);
 316
 317         if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
 318             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
 319                 return (EROFS);
 320
 321         /*
 322          * We now have a segment name to search for, and a directory to search.
 323          *
 324          * Before tediously performing a linear scan of the directory,
 325          * check the name cache to see if the directory/name pair
 326          * we are looking for is known already.
 327          */
 328         if (cache_lookup(vdp, cnp->cn_nameptr, cnp->cn_namelen,
 329                          cnp->cn_nameiop, cnp->cn_flags, NULL, vpp)) {
 330                 return *vpp == NULLVP ? ENOENT : 0;
 331         }
 332
 333         /*
 334          * Suppress search for slots unless creating
 335          * file and at end of pathname, in which case
 336          * we watch for a place to put the new file in
 337          * case it doesn't already exist.
 338          */
 339         slotstatus = FOUND;
 340         slotfreespace = slotsize = slotneeded = 0;
 341         if ((nameiop == CREATE || nameiop == RENAME) &&
 342             (flags & ISLASTCN)) {
 343                 slotstatus = NONE;
 344                 slotneeded = EXT2FS_DIRSIZ(cnp->cn_namelen);
 345         }
 346
 347         /*
 348          * If there is cached information on a previous search of
 349          * this directory, pick up where we last left off.
 350          * We cache only lookups as these are the most common
 351          * and have the greatest payoff. Caching CREATE has little
 352          * benefit as it usually must search the entire directory
 353          * to determine that the entry does not exist. Caching the
 354          * location of the last DELETE or RENAME has not reduced
 355          * profiling time and hence has been removed in the interest
 356          * of simplicity.
 357          */
 358         bmask = vdp->v_mount->mnt_stat.f_iosize - 1;
 359         if (nameiop != LOOKUP || results->ulr_diroff == 0 ||
 360             results->ulr_diroff >= ext2fs_size(dp)) {
 361                 entryoffsetinblock = 0;
 362                 results->ulr_offset = 0;
 363                 numdirpasses = 1;
 364         } else {
 365                 results->ulr_offset = results->ulr_diroff;
 366                 if ((entryoffsetinblock = results->ulr_offset & bmask) &&
 367                     (error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL, &bp)))
 368                         return (error);
 369                 numdirpasses = 2;
 370                 nchstats.ncs_2passes++;
 371         }
 372         prevoff = results->ulr_offset;
 373         endsearch = roundup(ext2fs_size(dp), dirblksiz);
 374         enduseful = 0;
 375
 376 searchloop:
 377         while (results->ulr_offset < endsearch) {
 378                 if (curcpu()->ci_schedstate.spc_flags & SPCF_SHOULDYIELD)
 379                         preempt();
 380                 /*
 381                  * If necessary, get the next directory block.
 382                  */
 383                 if ((results->ulr_offset & bmask) == 0) {
 384                         if (bp != NULL)
 385                                 brelse(bp, 0);
 386                         error = ext2fs_blkatoff(vdp, (off_t)results->ulr_offset, NULL,
 387                             &bp);
 388                         if (error != 0)
 389                                 return (error);
 390                         entryoffsetinblock = 0;
 391                 }
 392                 /*
 393                  * If still looking for a slot, and at a dirblksize
 394                  * boundary, have to start looking for free space again.
 395                  */
 396                 if (slotstatus == NONE &&
 397                     (entryoffsetinblock & (dirblksiz - 1)) == 0) {
 398                         slotoffset = -1;
 399                         slotfreespace = 0;
 400                 }
 401                 /*
 402                  * Get pointer to next entry.
 403                  * Full validation checks are slow, so we only check
 404                  * enough to insure forward progress through the
 405                  * directory. Complete checks can be run by patching
 406                  * "dirchk" to be true.
 407                  */
 408                 KASSERT(bp != NULL);
 409                 ep = (struct ext2fs_direct *)
 410                         ((char *)bp->b_data + entryoffsetinblock);
 411                 if (ep->e2d_reclen == 0 ||
 412                     (dirchk &&
 413                      ext2fs_dirbadentry(vdp, ep, entryoffsetinblock))) {
 414                         int i;
 415
 416                         ufs_dirbad(dp, results->ulr_offset, "mangled entry");
 417                         i = dirblksiz - (entryoffsetinblock & (dirblksiz - 1));
 418                         results->ulr_offset += i;
 419                         entryoffsetinblock += i;
 420                         continue;
 421                 }
 422
 423                 /*
 424                  * If an appropriate sized slot has not yet been found,
 425                  * check to see if one is available. Also accumulate space
 426                  * in the current block so that we can determine if
 427                  * compaction is viable.
 428                  */
 429                 if (slotstatus != FOUND) {
 430                         int size = fs2h16(ep->e2d_reclen);
 431
 432                         if (ep->e2d_ino != 0)
 433                                 size -= EXT2FS_DIRSIZ(ep->e2d_namlen);
 434                         if (size > 0) {
 435                                 if (size >= slotneeded) {
 436                                         slotstatus = FOUND;
 437                                         slotoffset = results->ulr_offset;
 438                                         slotsize = fs2h16(ep->e2d_reclen);
 439                                 } else if (slotstatus == NONE) {
 440                                         slotfreespace += size;
 441                                         if (slotoffset == -1)
 442                                                 slotoffset = results->ulr_offset;
 443                                         if (slotfreespace >= slotneeded) {
 444                                                 slotstatus = COMPACT;
 445                                                 slotsize = results->ulr_offset +
 446                                                     fs2h16(ep->e2d_reclen) -
 447                                                     slotoffset;
 448                                         }
 449                                 }
 450                         }
 451                 }
 452
 453                 /*
 454                  * Check for a name match.
 455                  */
 456                 if (ep->e2d_ino) {
 457                         namlen = ep->e2d_namlen;
 458                         if (namlen == cnp->cn_namelen &&
 459                             !memcmp(cnp->cn_nameptr, ep->e2d_name,
 460                             (unsigned)namlen)) {
 461                                 /*
 462                                  * Save directory entry's inode number and
 463                                  * reclen in ndp->ni_ufs area, and release
 464                                  * directory buffer.
 465                                  */
 466                                 foundino = fs2h32(ep->e2d_ino);
 467                                 results->ulr_reclen = fs2h16(ep->e2d_reclen);
 468                                 goto found;
 469                         }
 470                 }
 471                 prevoff = results->ulr_offset;
 472                 results->ulr_offset += fs2h16(ep->e2d_reclen);
 473                 entryoffsetinblock += fs2h16(ep->e2d_reclen);
 474                 if (ep->e2d_ino)
 475                         enduseful = results->ulr_offset;
 476         }
 477 /* notfound: */
 478         /*
 479          * If we started in the middle of the directory and failed
 480          * to find our target, we must check the beginning as well.
 481          */
 482         if (numdirpasses == 2) {
 483                 numdirpasses--;
 484                 results->ulr_offset = 0;
 485                 endsearch = results->ulr_diroff;
 486                 goto searchloop;
 487         }
 488         if (bp != NULL)
 489                 brelse(bp, 0);
 490         /*
 491          * If creating, and at end of pathname and current
 492          * directory has not been removed, then can consider
 493          * allowing file to be created.
 494          */
 495         if ((nameiop == CREATE || nameiop == RENAME) &&
 496             (flags & ISLASTCN) && dp->i_e2fs_nlink != 0) {
 497                 /*
 498                  * Access for write is interpreted as allowing
 499                  * creation of files in the directory.
 500                  */
 501                 error = VOP_ACCESS(vdp, VWRITE, cred);
 502                 if (error)
 503                         return (error);
 504                 /*
 505                  * Return an indication of where the new directory
 506                  * entry should be put.  If we didn't find a slot,
 507                  * then set results->ulr_count to 0 indicating
 508                  * that the new slot belongs at the end of the
 509                  * directory. If we found a slot, then the new entry
 510                  * can be put in the range from results->ulr_offset to
 511                  * results->ulr_offset + results->ulr_count.
 512                  */
 513                 if (slotstatus == NONE) {
 514                         results->ulr_offset = roundup(ext2fs_size(dp), dirblksiz);
 515                         results->ulr_count = 0;
 516                         enduseful = results->ulr_offset;
 517                 } else {
 518                         results->ulr_offset = slotoffset;
 519                         results->ulr_count = slotsize;
 520                         if (enduseful < slotoffset + slotsize)
 521                                 enduseful = slotoffset + slotsize;
 522                 }
 523                 results->ulr_endoff = roundup(enduseful, dirblksiz);
 524 #if 0
 525                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
 526 #endif
 527                 /*
 528                  * We return with the directory locked, so that
 529                  * the parameters we set up above will still be
 530                  * valid if we actually decide to do a direnter().
 531                  * We return ni_vp == NULL to indicate that the entry
 532                  * does not currently exist; we leave a pointer to
 533                  * the (locked) directory inode in ndp->ni_dvp.
 534                  *
 535                  * NB - if the directory is unlocked, then this
 536                  * information cannot be used.
 537                  */
 538                 return (EJUSTRETURN);
 539         }
 540         /*
 541          * Insert name into cache (as non-existent) if appropriate.
 542          */
 543         if (nameiop != CREATE) {
 544                 cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen,
 545                             cnp->cn_flags);
 546         }
 547         return ENOENT;
 548
 549 found:
 550         if (numdirpasses == 2)
 551                 nchstats.ncs_pass2++;
 552         /*
 553          * Check that directory length properly reflects presence
 554          * of this entry.
 555          */
 556         if (results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen) > ext2fs_size(dp)) {
 557                 ufs_dirbad(dp, results->ulr_offset, "i_size too small");
 558                 error = ext2fs_setsize(dp,
 559                                 results->ulr_offset + EXT2FS_DIRSIZ(ep->e2d_namlen));
 560                 if (error) {
 561                         brelse(bp, 0);
 562                         return (error);
 563                 }
 564                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
 565                 uvm_vnp_setsize(vdp, ext2fs_size(dp));
 566         }
 567         brelse(bp, 0);
 568
 569         /*
 570          * Found component in pathname.
 571          * If the final component of path name, save information
 572          * in the cache as to where the entry was found.
 573          */
 574         if ((flags & ISLASTCN) && nameiop == LOOKUP)
 575                 results->ulr_diroff = results->ulr_offset &~ (dirblksiz - 1);
 576
 577         /*
 578          * If deleting, and at end of pathname, return
 579          * parameters which can be used to remove file.
 580          * Lock the inode, being careful with ".".
 581          */
 582         if (nameiop == DELETE && (flags & ISLASTCN)) {
 583                 /*
 584                  * Return pointer to current entry in results->ulr_offset,
 585                  * and distance past previous entry (if there
 586                  * is a previous entry in this block) in results->ulr_count.
 587                  * Save directory inode pointer in ndp->ni_dvp for dirremove().
 588                  */
 589                 if ((results->ulr_offset & (dirblksiz - 1)) == 0)
 590                         results->ulr_count = 0;
 591                 else
 592                         results->ulr_count = results->ulr_offset - prevoff;
 593                 if (dp->i_number == foundino) {
 594                         vref(vdp);
 595                         tdp = vdp;
 596                 } else {
 597                         if (flags & ISDOTDOT)
 598                                 VOP_UNLOCK(vdp); /* race to get the inode */
 599                         error = VFS_VGET(vdp->v_mount, foundino, &tdp);
 600                         if (flags & ISDOTDOT)
 601                                 vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
 602                         if (error)
 603                                 return (error);
 604                 }
 605                 /*
 606                  * Write access to directory required to delete files.
 607                  */
 608                 if ((error = VOP_ACCESS(vdp, VWRITE, cred)) != 0) {
 609                         if (dp->i_number == foundino)
 610                                 vrele(tdp);
 611                         else
 612                                 vput(tdp);
 613                         return (error);
 614                 }
 615                 /*
 616                  * If directory is "sticky", then user must own
 617                  * the directory, or the file in it, else she
 618                  * may not delete it (unless she's root). This
 619                  * implements append-only directories.
 620                  */
 621                 if (dp->i_e2fs_mode & ISVTX) {
 622                         error = kauth_authorize_vnode(cred, KAUTH_VNODE_DELETE,
 623                             tdp, vdp, genfs_can_sticky(cred, dp->i_uid,
 624                             VTOI(tdp)->i_uid));
 625                         if (error) {
 626                                 if (dp->i_number == foundino)
 627                                         vrele(tdp);
 628                                 else
 629                                         vput(tdp);
 630                                 return (EPERM);
 631                         }
 632                 }
 633                 *vpp = tdp;
 634                 return (0);
 635         }
 636
 637         /*
 638          * If rewriting (RENAME), return the inode and the
 639          * information required to rewrite the present directory
 640          * Must get inode of directory entry to verify it's a
 641          * regular file, or empty directory.
 642          */
 643         if (nameiop == RENAME && (flags & ISLASTCN)) {
 644                 error = VOP_ACCESS(vdp, VWRITE, cred);
 645                 if (error)
 646                         return (error);
 647                 /*
 648                  * Careful about locking second inode.
 649                  * This can only occur if the target is ".".
 650                  */
 651                 if (dp->i_number == foundino)
 652                         return (EISDIR);
 653                 if (flags & ISDOTDOT)
 654                         VOP_UNLOCK(vdp); /* race to get the inode */
 655                 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
 656                 if (flags & ISDOTDOT)
 657                         vn_lock(vdp, LK_EXCLUSIVE | LK_RETRY);
 658                 if (error)
 659                         return (error);
 660                 *vpp = tdp;
 661                 return (0);
 662         }
 663
 664         /*
 665          * Step through the translation in the name.  We do not `vput' the
 666          * directory because we may need it again if a symbolic link
 667          * is relative to the current directory.  Instead we save it
 668          * unlocked as "pdp".  We must get the target inode before unlocking
 669          * the directory to insure that the inode will not be removed
 670          * before we get it.  We prevent deadlock by always fetching
 671          * inodes from the root, moving down the directory tree. Thus
 672          * when following backward pointers ".." we must unlock the
 673          * parent directory before getting the requested directory.
 674          * There is a potential race condition here if both the current
 675          * and parent directories are removed before the VFS_VGET for the
 676          * inode associated with ".." returns.  We hope that this occurs
 677          * infrequently since we cannot avoid this race condition without
 678          * implementing a sophisticated deadlock detection algorithm.
 679          * Note also that this simple deadlock detection scheme will not
 680          * work if the file system has any hard links other than ".."
 681          * that point backwards in the directory structure.
 682          */
 683         pdp = vdp;
 684         if (flags & ISDOTDOT) {
 685                 VOP_UNLOCK(pdp);        /* race to get the inode */
 686                 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
 687                 vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
 688                 if (error) {
 689                         return (error);
 690                 }
 691                 *vpp = tdp;
 692         } else if (dp->i_number == foundino) {
 693                 vref(vdp);      /* we want ourself, ie "." */
 694                 *vpp = vdp;
 695         } else {
 696                 error = VFS_VGET(vdp->v_mount, foundino, &tdp);
 697                 if (error)
 698                         return (error);
 699                 *vpp = tdp;
 700         }
 701
 702         /*
 703          * Insert name into cache if appropriate.
 704          */
 705         cache_enter(vdp, *vpp, cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_flags);
 706         return 0;
 707 }
 708
 709 /*
 710  * Do consistency checking on a directory entry:
 711  *      record length must be multiple of 4
 712  *      entry must fit in rest of its dirblksize block
 713  *      record must be large enough to contain entry
 714  *      name is not longer than EXT2FS_MAXNAMLEN
 715  *      name must be as long as advertised, and null terminated
 716  */
 717 /*
 718  *      changed so that it confirms to ext2fs_check_dir_entry
 719  */
 720 static int
 721 ext2fs_dirbadentry(struct vnode *dp, struct ext2fs_direct *de,
 722                 int entryoffsetinblock)
 723 {
 724         struct ufsmount *ump = VFSTOUFS(dp->v_mount);
 725         int dirblksiz = ump->um_dirblksiz;
 726
 727                 const char *error_msg = NULL;
 728                 int reclen = fs2h16(de->e2d_reclen);
 729                 int namlen = de->e2d_namlen;
 730
 731                 if (reclen < EXT2FS_DIRSIZ(1)) /* e2d_namlen = 1 */
 732                         error_msg = "rec_len is smaller than minimal";
 733                 else if (reclen % 4 != 0)
 734                         error_msg = "rec_len % 4 != 0";
 735                 else if (namlen > EXT2FS_MAXNAMLEN)
 736                         error_msg = "namlen > EXT2FS_MAXNAMLEN";
 737                 else if (reclen < EXT2FS_DIRSIZ(namlen))
 738                         error_msg = "reclen is too small for name_len";
 739                 else if (entryoffsetinblock + reclen > dirblksiz)
 740                         error_msg = "directory entry across blocks";
 741                 else if (fs2h32(de->e2d_ino) >
 742                     VTOI(dp)->i_e2fs->e2fs.e2fs_icount)
 743                         error_msg = "inode out of bounds";
 744
 745                 if (error_msg != NULL) {
 746                         printf( "bad directory entry: %s\n"
 747                             "offset=%d, inode=%lu, rec_len=%d, name_len=%d \n",
 748                             error_msg, entryoffsetinblock,
 749                             (unsigned long) fs2h32(de->e2d_ino),
 750                             reclen, namlen);
 751                         panic("ext2fs_dirbadentry");
 752                 }
 753                 return error_msg == NULL ? 0 : 1;
 754 }
 755
 756 /*
 757  * Write a directory entry after a call to namei, using the parameters
 758  * that it left in nameidata.  The argument ip is the inode which the new
 759  * directory entry will refer to.  Dvp is a pointer to the directory to
 760  * be written, which was left locked by namei. Remaining parameters
 761  * (ulr_offset, ulr_count) indicate how the space for the new
 762  * entry is to be obtained.
 763  */
 764 int
 765 ext2fs_direnter(struct inode *ip, struct vnode *dvp,
 766                 const struct ufs_lookup_results *ulr,
 767                 struct componentname *cnp)
 768 {
 769         struct ext2fs_direct *ep, *nep;
 770         struct inode *dp;
 771         struct buf *bp;
 772         struct ext2fs_direct newdir;
 773         struct iovec aiov;
 774         struct uio auio;
 775         u_int dsize;
 776         int error, loc, newentrysize, spacefree;
 777         char *dirbuf;
 778         struct ufsmount *ump = VFSTOUFS(dvp->v_mount);
 779         int dirblksiz = ump->um_dirblksiz;
 780
 781         dp = VTOI(dvp);
 782
 783         newdir.e2d_ino = h2fs32(ip->i_number);
 784         newdir.e2d_namlen = cnp->cn_namelen;
 785         if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
 786             (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
 787                 newdir.e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
 788         } else {
 789                 newdir.e2d_type = 0;
 790         }
 791         memcpy(newdir.e2d_name, cnp->cn_nameptr, (unsigned)cnp->cn_namelen + 1);
 792         newentrysize = EXT2FS_DIRSIZ(cnp->cn_namelen);
 793         if (ulr->ulr_count == 0) {
 794                 /*
 795                  * If ulr_count is 0, then namei could find no
 796                  * space in the directory. Here, ulr_offset will
 797                  * be on a directory block boundary and we will write the
 798                  * new entry into a fresh block.
 799                  */
 800                 if (ulr->ulr_offset & (dirblksiz - 1))
 801                         panic("ext2fs_direnter: newblk");
 802                 auio.uio_offset = ulr->ulr_offset;
 803                 newdir.e2d_reclen = h2fs16(dirblksiz);
 804                 auio.uio_resid = newentrysize;
 805                 aiov.iov_len = newentrysize;
 806                 aiov.iov_base = (void *)&newdir;
 807                 auio.uio_iov = &aiov;
 808                 auio.uio_iovcnt = 1;
 809                 auio.uio_rw = UIO_WRITE;
 810                 UIO_SETUP_SYSSPACE(&auio);
 811                 error = VOP_WRITE(dvp, &auio, IO_SYNC, cnp->cn_cred);
 812                 if (dirblksiz > dvp->v_mount->mnt_stat.f_bsize)
 813                         /* XXX should grow with balloc() */
 814                         panic("ext2fs_direnter: frag size");
 815                 else if (!error) {
 816                         error = ext2fs_setsize(dp,
 817                                 roundup(ext2fs_size(dp), dirblksiz));
 818                         if (error)
 819                                 return (error);
 820                         dp->i_flag |= IN_CHANGE;
 821                         uvm_vnp_setsize(dvp, ext2fs_size(dp));
 822                 }
 823                 return (error);
 824         }
 825
 826         /*
 827          * If ulr_count is non-zero, then namei found space
 828          * for the new entry in the range ulr_offset to
 829          * ulr_offset + ulr_count in the directory.
 830          * To use this space, we may have to compact the entries located
 831          * there, by copying them together towards the beginning of the
 832          * block, leaving the free space in one usable chunk at the end.
 833          */
 834
 835         /*
 836          * Get the block containing the space for the new directory entry.
 837          */
 838         if ((error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset, &dirbuf, &bp)) != 0)
 839                 return (error);
 840         /*
 841          * Find space for the new entry. In the simple case, the entry at
 842          * offset base will have the space. If it does not, then namei
 843          * arranged that compacting the region ulr_offset to
 844          * ulr_offset + ulr_count would yield the
 845          * space.
 846          */
 847         ep = (struct ext2fs_direct *)dirbuf;
 848         dsize = EXT2FS_DIRSIZ(ep->e2d_namlen);
 849         spacefree = fs2h16(ep->e2d_reclen) - dsize;
 850         for (loc = fs2h16(ep->e2d_reclen); loc < ulr->ulr_count; ) {
 851                 nep = (struct ext2fs_direct *)(dirbuf + loc);
 852                 if (ep->e2d_ino) {
 853                         /* trim the existing slot */
 854                         ep->e2d_reclen = h2fs16(dsize);
 855                         ep = (struct ext2fs_direct *)((char *)ep + dsize);
 856                 } else {
 857                         /* overwrite; nothing there; header is ours */
 858                         spacefree += dsize;
 859                 }
 860                 dsize = EXT2FS_DIRSIZ(nep->e2d_namlen);
 861                 spacefree += fs2h16(nep->e2d_reclen) - dsize;
 862                 loc += fs2h16(nep->e2d_reclen);
 863                 memcpy((void *)ep, (void *)nep, dsize);
 864         }
 865         /*
 866          * Update the pointer fields in the previous entry (if any),
 867          * copy in the new entry, and write out the block.
 868          */
 869         if (ep->e2d_ino == 0) {
 870 #ifdef DIAGNOSTIC
 871                 if (spacefree + dsize < newentrysize)
 872                         panic("ext2fs_direnter: compact1");
 873 #endif
 874                 newdir.e2d_reclen = h2fs16(spacefree + dsize);
 875         } else {
 876 #ifdef DIAGNOSTIC
 877                 if (spacefree < newentrysize) {
 878                         printf("ext2fs_direnter: compact2 %u %u",
 879                             (u_int)spacefree, (u_int)newentrysize);
 880                         panic("ext2fs_direnter: compact2");
 881                 }
 882 #endif
 883                 newdir.e2d_reclen = h2fs16(spacefree);
 884                 ep->e2d_reclen = h2fs16(dsize);
 885                 ep = (struct ext2fs_direct *)((char *)ep + dsize);
 886         }
 887         memcpy((void *)ep, (void *)&newdir, (u_int)newentrysize);
 888         error = VOP_BWRITE(bp->b_vp, bp);
 889         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 890         if (!error && ulr->ulr_endoff && ulr->ulr_endoff < ext2fs_size(dp))
 891                 error = ext2fs_truncate(dvp, (off_t)ulr->ulr_endoff, IO_SYNC,
 892                     cnp->cn_cred);
 893         return (error);
 894 }
 895
 896 /*
 897  * Remove a directory entry after a call to namei, using
 898  * the auxiliary results it provided. The entry
 899  * ulr_offset contains the offset into the directory of the
 900  * entry to be eliminated.  The ulr_count field contains the
 901  * size of the previous record in the directory.  If this
 902  * is 0, the first entry is being deleted, so we need only
 903  * zero the inode number to mark the entry as free.  If the
 904  * entry is not the first in the directory, we must reclaim
 905  * the space of the now empty record by adding the record size
 906  * to the size of the previous entry.
 907  */
 908 int
 909 ext2fs_dirremove(struct vnode *dvp, const struct ufs_lookup_results *ulr,
 910                  struct componentname *cnp)
 911 {
 912         struct inode *dp;
 913         struct ext2fs_direct *ep;
 914         struct buf *bp;
 915         int error;
 916
 917         dp = VTOI(dvp);
 918
 919         if (ulr->ulr_count == 0) {
 920                 /*
 921                  * First entry in block: set d_ino to zero.
 922                  */
 923                 error = ext2fs_blkatoff(dvp, (off_t)ulr->ulr_offset,
 924                     (void *)&ep, &bp);
 925                 if (error != 0)
 926                         return (error);
 927                 ep->e2d_ino = 0;
 928                 error = VOP_BWRITE(bp->b_vp, bp);
 929                 dp->i_flag |= IN_CHANGE | IN_UPDATE;
 930                 return (error);
 931         }
 932         /*
 933          * Collapse new free space into previous entry.
 934          */
 935         error = ext2fs_blkatoff(dvp, (off_t)(ulr->ulr_offset - ulr->ulr_count),
 936             (void *)&ep, &bp);
 937         if (error != 0)
 938                 return (error);
 939         ep->e2d_reclen = h2fs16(fs2h16(ep->e2d_reclen) + ulr->ulr_reclen);
 940         error = VOP_BWRITE(bp->b_vp, bp);
 941         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 942         return (error);
 943 }
 944
 945 /*
 946  * Rewrite an existing directory entry to point at the inode
 947  * supplied.  The parameters describing the directory entry are
 948  * set up by a call to namei.
 949  */
 950 int
 951 ext2fs_dirrewrite(struct inode *dp, const struct ufs_lookup_results *ulr,
 952     struct inode *ip, struct componentname *cnp)
 953 {
 954         struct buf *bp;
 955         struct ext2fs_direct *ep;
 956         struct vnode *vdp = ITOV(dp);
 957         int error;
 958
 959         error = ext2fs_blkatoff(vdp, (off_t)ulr->ulr_offset, (void *)&ep, &bp);
 960         if (error != 0)
 961                 return (error);
 962         ep->e2d_ino = h2fs32(ip->i_number);
 963         if (ip->i_e2fs->e2fs.e2fs_rev > E2FS_REV0 &&
 964             (ip->i_e2fs->e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE)) {
 965                 ep->e2d_type = inot2ext2dt(IFTODT(ip->i_e2fs_mode));
 966         } else {
 967                 ep->e2d_type = 0;
 968         }
 969         error = VOP_BWRITE(bp->b_vp, bp);
 970         dp->i_flag |= IN_CHANGE | IN_UPDATE;
 971         return (error);
 972 }
 973
 974 /*
 975  * Check if a directory is empty or not.
 976  * Inode supplied must be locked.
 977  *
 978  * Using a struct dirtemplate here is not precisely
 979  * what we want, but better than using a struct ext2fs_direct.
 980  *
 981  * NB: does not handle corrupted directories.
 982  */
 983 int
 984 ext2fs_dirempty(struct inode *ip, ino_t parentino, kauth_cred_t cred)
 985 {
 986         off_t off;
 987         struct ext2fs_dirtemplate dbuf;
 988         struct ext2fs_direct *dp = (struct ext2fs_direct *)&dbuf;
 989         int error, namlen;
 990         size_t count;
 991
 992 #define MINDIRSIZ (sizeof (struct ext2fs_dirtemplate) / 2)
 993
 994         for (off = 0; off < ext2fs_size(ip); off += fs2h16(dp->e2d_reclen)) {
 995                 error = vn_rdwr(UIO_READ, ITOV(ip), (void *)dp, MINDIRSIZ, off,
 996                    UIO_SYSSPACE, IO_NODELOCKED, cred, &count, NULL);
 997                 /*
 998                  * Since we read MINDIRSIZ, residual must
 999                  * be 0 unless we're at end of file.
1000                  */
1001                 if (error || count != 0)
1002                         return (0);
1003                 /* avoid infinite loops */
1004                 if (dp->e2d_reclen == 0)
1005                         return (0);
1006                 /* skip empty entries */
1007                 if (dp->e2d_ino == 0)
1008                         continue;
1009                 /* accept only "." and ".." */
1010                 namlen = dp->e2d_namlen;
1011                 if (namlen > 2)
1012                         return (0);
1013                 if (dp->e2d_name[0] != '.')
1014                         return (0);
1015                 /*
1016                  * At this point namlen must be 1 or 2.
1017                  * 1 implies ".", 2 implies ".." if second
1018                  * char is also "."
1019                  */
1020                 if (namlen == 1)
1021                         continue;
1022                 if (dp->e2d_name[1] == '.' && fs2h32(dp->e2d_ino) == parentino)
1023                         continue;
1024                 return (0);
1025         }
1026         return (1);
1027 }
1028
1029 /*
1030  * Check if source directory is in the path of the target directory.
1031  * Target is supplied locked, source is unlocked.
1032  * The target is always vput before returning.
1033  */
1034 int
1035 ext2fs_checkpath(struct inode *source, struct inode *target,
1036         kauth_cred_t cred)
1037 {
1038         struct vnode *vp;
1039         int error, rootino, namlen;
1040         struct ext2fs_dirtemplate dirbuf;
1041         uint32_t ino;
1042
1043         vp = ITOV(target);
1044         if (target->i_number == source->i_number) {
1045                 error = EEXIST;
1046                 goto out;
1047         }
1048         rootino = UFS_ROOTINO;
1049         error = 0;
1050         if (target->i_number == rootino)
1051                 goto out;
1052
1053         for (;;) {
1054                 if (vp->v_type != VDIR) {
1055                         error = ENOTDIR;
1056                         break;
1057                 }
1058                 error = vn_rdwr(UIO_READ, vp, (void *)&dirbuf,
1059                         sizeof (struct ext2fs_dirtemplate), (off_t)0,
1060                         UIO_SYSSPACE, IO_NODELOCKED, cred, (size_t *)0,
1061                         NULL);
1062                 if (error != 0)
1063                         break;
1064                 namlen = dirbuf.dotdot_namlen;
1065                 if (namlen != 2 ||
1066                         dirbuf.dotdot_name[0] != '.' ||
1067                         dirbuf.dotdot_name[1] != '.') {
1068                         error = ENOTDIR;
1069                         break;
1070                 }
1071                 ino = fs2h32(dirbuf.dotdot_ino);
1072                 if (ino == source->i_number) {
1073                         error = EINVAL;
1074                         break;
1075                 }
1076                 if (ino == rootino)
1077                         break;
1078                 vput(vp);
1079                 error = VFS_VGET(vp->v_mount, ino, &vp);
1080                 if (error != 0) {
1081                         vp = NULL;
1082                         break;
1083                 }
1084         }
1085
1086 out:
1087         if (error == ENOTDIR) {
1088                 printf("checkpath: .. not a directory\n");
1089                 panic("checkpath");
1090         }
1091         if (vp != NULL)
1092                 vput(vp);
1093         return (error);
1094 }