fs/ntfs/dir.c

   1 /**
   2  * dir.c - NTFS kernel directory operations. Part of the Linux-NTFS project.
   3  *
   4  * Copyright (c) 2001-2004 Anton Altaparmakov
   5  * Copyright (c) 2002 Richard Russon
   6  *
   7  * This program/include file is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License as published
   9  * by the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program/include file is distributed in the hope that it will be
  13  * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  14  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program (in the main directory of the Linux-NTFS
  19  * distribution in the file COPYING); if not, write to the Free Software
  20  * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  */
  22
  23 #include <linux/smp_lock.h>
  24 #include <linux/buffer_head.h>
  25
  26 #include "dir.h"
  27 #include "aops.h"
  28 #include "attrib.h"
  29 #include "mft.h"
  30 #include "debug.h"
  31 #include "ntfs.h"
  32
  33 /**
  34  * The little endian Unicode string $I30 as a global constant.
  35  */
  36 ntfschar I30[5] = { const_cpu_to_le16('$'), const_cpu_to_le16('I'),
  37                 const_cpu_to_le16('3'), const_cpu_to_le16('0'), 0 };
  38
  39 /**
  40  * ntfs_lookup_inode_by_name - find an inode in a directory given its name
  41  * @dir_ni:     ntfs inode of the directory in which to search for the name
  42  * @uname:      Unicode name for which to search in the directory
  43  * @uname_len:  length of the name @uname in Unicode characters
  44  * @res:        return the found file name if necessary (see below)
  45  *
  46  * Look for an inode with name @uname in the directory with inode @dir_ni.
  47  * ntfs_lookup_inode_by_name() walks the contents of the directory looking for
  48  * the Unicode name. If the name is found in the directory, the corresponding
  49  * inode number (>= 0) is returned as a mft reference in cpu format, i.e. it
  50  * is a 64-bit number containing the sequence number.
  51  *
  52  * On error, a negative value is returned corresponding to the error code. In
  53  * particular if the inode is not found -ENOENT is returned. Note that you
  54  * can't just check the return value for being negative, you have to check the
  55  * inode number for being negative which you can extract using MREC(return
  56  * value).
  57  *
  58  * Note, @uname_len does not include the (optional) terminating NULL character.
  59  *
  60  * Note, we look for a case sensitive match first but we also look for a case
  61  * insensitive match at the same time. If we find a case insensitive match, we
  62  * save that for the case that we don't find an exact match, where we return
  63  * the case insensitive match and setup @res (which we allocate!) with the mft
  64  * reference, the file name type, length and with a copy of the little endian
  65  * Unicode file name itself. If we match a file name which is in the DOS name
  66  * space, we only return the mft reference and file name type in @res.
  67  * ntfs_lookup() then uses this to find the long file name in the inode itself.
  68  * This is to avoid polluting the dcache with short file names. We want them to
  69  * work but we don't care for how quickly one can access them. This also fixes
  70  * the dcache aliasing issues.
  71  *
  72  * Locking:  - Caller must hold i_sem on the directory.
  73  *           - Each page cache page in the index allocation mapping must be
  74  *             locked whilst being accessed otherwise we may find a corrupt
  75  *             page due to it being under ->writepage at the moment which
  76  *             applies the mst protection fixups before writing out and then
  77  *             removes them again after the write is complete after which it
  78  *             unlocks the page.
  79  */
  80 MFT_REF ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname,
  81                 const int uname_len, ntfs_name **res)
  82 {
  83         ntfs_volume *vol = dir_ni->vol;
  84         struct super_block *sb = vol->sb;
  85         MFT_RECORD *m;
  86         INDEX_ROOT *ir;
  87         INDEX_ENTRY *ie;
  88         INDEX_ALLOCATION *ia;
  89         u8 *index_end;
  90         u64 mref;
  91         ntfs_attr_search_ctx *ctx;
  92         int err, rc;
  93         VCN vcn, old_vcn;
  94         struct address_space *ia_mapping;
  95         struct page *page;
  96         u8 *kaddr;
  97         ntfs_name *name = NULL;
  98
  99         BUG_ON(!S_ISDIR(VFS_I(dir_ni)->i_mode));
 100         BUG_ON(NInoAttr(dir_ni));
 101         /* Get hold of the mft record for the directory. */
 102         m = map_mft_record(dir_ni);
 103         if (IS_ERR(m)) {
 104                 ntfs_error(sb, "map_mft_record() failed with error code %ld.",
 105                                 -PTR_ERR(m));
 106                 return ERR_MREF(PTR_ERR(m));
 107         }
 108         ctx = ntfs_attr_get_search_ctx(dir_ni, m);
 109         if (unlikely(!ctx)) {
 110                 err = -ENOMEM;
 111                 goto err_out;
 112         }
 113         /* Find the index root attribute in the mft record. */
 114         err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL,
 115                         0, ctx);
 116         if (unlikely(err)) {
 117                 if (err == -ENOENT) {
 118                         ntfs_error(sb, "Index root attribute missing in "
 119                                         "directory inode 0x%lx.",
 120                                         dir_ni->mft_no);
 121                         err = -EIO;
 122                 }
 123                 goto err_out;
 124         }
 125         /* Get to the index root value (it's been verified in read_inode). */
 126         ir = (INDEX_ROOT*)((u8*)ctx->attr +
 127                         le16_to_cpu(ctx->attr->data.resident.value_offset));
 128         index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
 129         /* The first index entry. */
 130         ie = (INDEX_ENTRY*)((u8*)&ir->index +
 131                         le32_to_cpu(ir->index.entries_offset));
 132         /*
 133          * Loop until we exceed valid memory (corruption case) or until we
 134          * reach the last entry.
 135          */
 136         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
 137                 /* Bounds checks. */
 138                 if ((u8*)ie < (u8*)ctx->mrec || (u8*)ie +
 139                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
 140                                 (u8*)ie + le16_to_cpu(ie->key_length) >
 141                                 index_end)
 142                         goto dir_err_out;
 143                 /*
 144                  * The last entry cannot contain a name. It can however contain
 145                  * a pointer to a child node in the B+tree so we just break out.
 146                  */
 147                 if (ie->flags & INDEX_ENTRY_END)
 148                         break;
 149                 /*
 150                  * We perform a case sensitive comparison and if that matches
 151                  * we are done and return the mft reference of the inode (i.e.
 152                  * the inode number together with the sequence number for
 153                  * consistency checking). We convert it to cpu format before
 154                  * returning.
 155                  */
 156                 if (ntfs_are_names_equal(uname, uname_len,
 157                                 (ntfschar*)&ie->key.file_name.file_name,
 158                                 ie->key.file_name.file_name_length,
 159                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len)) {
 160 found_it:
 161                         /*
 162                          * We have a perfect match, so we don't need to care
 163                          * about having matched imperfectly before, so we can
 164                          * free name and set *res to NULL.
 165                          * However, if the perfect match is a short file name,
 166                          * we need to signal this through *res, so that
 167                          * ntfs_lookup() can fix dcache aliasing issues.
 168                          * As an optimization we just reuse an existing
 169                          * allocation of *res.
 170                          */
 171                         if (ie->key.file_name.file_name_type == FILE_NAME_DOS) {
 172                                 if (!name) {
 173                                         name = kmalloc(sizeof(ntfs_name),
 174                                                         GFP_NOFS);
 175                                         if (!name) {
 176                                                 err = -ENOMEM;
 177                                                 goto err_out;
 178                                         }
 179                                 }
 180                                 name->mref = le64_to_cpu(
 181                                                 ie->data.dir.indexed_file);
 182                                 name->type = FILE_NAME_DOS;
 183                                 name->len = 0;
 184                                 *res = name;
 185                         } else {
 186                                 if (name)
 187                                         kfree(name);
 188                                 *res = NULL;
 189                         }
 190                         mref = le64_to_cpu(ie->data.dir.indexed_file);
 191                         ntfs_attr_put_search_ctx(ctx);
 192                         unmap_mft_record(dir_ni);
 193                         return mref;
 194                 }
 195                 /*
 196                  * For a case insensitive mount, we also perform a case
 197                  * insensitive comparison (provided the file name is not in the
 198                  * POSIX namespace). If the comparison matches, and the name is
 199                  * in the WIN32 namespace, we cache the filename in *res so
 200                  * that the caller, ntfs_lookup(), can work on it. If the
 201                  * comparison matches, and the name is in the DOS namespace, we
 202                  * only cache the mft reference and the file name type (we set
 203                  * the name length to zero for simplicity).
 204                  */
 205                 if (!NVolCaseSensitive(vol) &&
 206                                 ie->key.file_name.file_name_type &&
 207                                 ntfs_are_names_equal(uname, uname_len,
 208                                 (ntfschar*)&ie->key.file_name.file_name,
 209                                 ie->key.file_name.file_name_length,
 210                                 IGNORE_CASE, vol->upcase, vol->upcase_len)) {
 211                         int name_size = sizeof(ntfs_name);
 212                         u8 type = ie->key.file_name.file_name_type;
 213                         u8 len = ie->key.file_name.file_name_length;
 214
 215                         /* Only one case insensitive matching name allowed. */
 216                         if (name) {
 217                                 ntfs_error(sb, "Found already allocated name "
 218                                                 "in phase 1. Please run chkdsk "
 219                                                 "and if that doesn't find any "
 220                                                 "errors please report you saw "
 221                                                 "this message to "
 222                                                 "linux-ntfs-dev@lists."
 223                                                 "sourceforge.net.");
 224                                 goto dir_err_out;
 225                         }
 226
 227                         if (type != FILE_NAME_DOS)
 228                                 name_size += len * sizeof(ntfschar);
 229                         name = kmalloc(name_size, GFP_NOFS);
 230                         if (!name) {
 231                                 err = -ENOMEM;
 232                                 goto err_out;
 233                         }
 234                         name->mref = le64_to_cpu(ie->data.dir.indexed_file);
 235                         name->type = type;
 236                         if (type != FILE_NAME_DOS) {
 237                                 name->len = len;
 238                                 memcpy(name->name, ie->key.file_name.file_name,
 239                                                 len * sizeof(ntfschar));
 240                         } else
 241                                 name->len = 0;
 242                         *res = name;
 243                 }
 244                 /*
 245                  * Not a perfect match, need to do full blown collation so we
 246                  * know which way in the B+tree we have to go.
 247                  */
 248                 rc = ntfs_collate_names(uname, uname_len,
 249                                 (ntfschar*)&ie->key.file_name.file_name,
 250                                 ie->key.file_name.file_name_length, 1,
 251                                 IGNORE_CASE, vol->upcase, vol->upcase_len);
 252                 /*
 253                  * If uname collates before the name of the current entry, there
 254                  * is definitely no such name in this index but we might need to
 255                  * descend into the B+tree so we just break out of the loop.
 256                  */
 257                 if (rc == -1)
 258                         break;
 259                 /* The names are not equal, continue the search. */
 260                 if (rc)
 261                         continue;
 262                 /*
 263                  * Names match with case insensitive comparison, now try the
 264                  * case sensitive comparison, which is required for proper
 265                  * collation.
 266                  */
 267                 rc = ntfs_collate_names(uname, uname_len,
 268                                 (ntfschar*)&ie->key.file_name.file_name,
 269                                 ie->key.file_name.file_name_length, 1,
 270                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len);
 271                 if (rc == -1)
 272                         break;
 273                 if (rc)
 274                         continue;
 275                 /*
 276                  * Perfect match, this will never happen as the
 277                  * ntfs_are_names_equal() call will have gotten a match but we
 278                  * still treat it correctly.
 279                  */
 280                 goto found_it;
 281         }
 282         /*
 283          * We have finished with this index without success. Check for the
 284          * presence of a child node and if not present return -ENOENT, unless
 285          * we have got a matching name cached in name in which case return the
 286          * mft reference associated with it.
 287          */
 288         if (!(ie->flags & INDEX_ENTRY_NODE)) {
 289                 if (name) {
 290                         ntfs_attr_put_search_ctx(ctx);
 291                         unmap_mft_record(dir_ni);
 292                         return name->mref;
 293                 }
 294                 ntfs_debug("Entry not found.");
 295                 err = -ENOENT;
 296                 goto err_out;
 297         } /* Child node present, descend into it. */
 298         /* Consistency check: Verify that an index allocation exists. */
 299         if (!NInoIndexAllocPresent(dir_ni)) {
 300                 ntfs_error(sb, "No index allocation attribute but index entry "
 301                                 "requires one. Directory inode 0x%lx is "
 302                                 "corrupt or driver bug.", dir_ni->mft_no);
 303                 goto err_out;
 304         }
 305         /* Get the starting vcn of the index_block holding the child node. */
 306         vcn = sle64_to_cpup((sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8));
 307         ia_mapping = VFS_I(dir_ni)->i_mapping;
 308         /*
 309          * We are done with the index root and the mft record. Release them,
 310          * otherwise we deadlock with ntfs_map_page().
 311          */
 312         ntfs_attr_put_search_ctx(ctx);
 313         unmap_mft_record(dir_ni);
 314         m = NULL;
 315         ctx = NULL;
 316 descend_into_child_node:
 317         /*
 318          * Convert vcn to index into the index allocation attribute in units
 319          * of PAGE_CACHE_SIZE and map the page cache page, reading it from
 320          * disk if necessary.
 321          */
 322         page = ntfs_map_page(ia_mapping, vcn <<
 323                         dir_ni->itype.index.vcn_size_bits >> PAGE_CACHE_SHIFT);
 324         if (IS_ERR(page)) {
 325                 ntfs_error(sb, "Failed to map directory index page, error %ld.",
 326                                 -PTR_ERR(page));
 327                 err = PTR_ERR(page);
 328                 goto err_out;
 329         }
 330         lock_page(page);
 331         kaddr = (u8*)page_address(page);
 332 fast_descend_into_child_node:
 333         /* Get to the index allocation block. */
 334         ia = (INDEX_ALLOCATION*)(kaddr + ((vcn <<
 335                         dir_ni->itype.index.vcn_size_bits) & ~PAGE_CACHE_MASK));
 336         /* Bounds checks. */
 337         if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) {
 338                 ntfs_error(sb, "Out of bounds check failed. Corrupt directory "
 339                                 "inode 0x%lx or driver bug.", dir_ni->mft_no);
 340                 goto unm_err_out;
 341         }
 342         /* Catch multi sector transfer fixup errors. */
 343         if (unlikely(!ntfs_is_indx_record(ia->magic))) {
 344                 ntfs_error(sb, "Directory index record with vcn 0x%llx is "
 345                                 "corrupt.  Corrupt inode 0x%lx.  Run chkdsk.",
 346                                 (unsigned long long)vcn, dir_ni->mft_no);
 347                 goto unm_err_out;
 348         }
 349         if (sle64_to_cpu(ia->index_block_vcn) != vcn) {
 350                 ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is "
 351                                 "different from expected VCN (0x%llx). "
 352                                 "Directory inode 0x%lx is corrupt or driver "
 353                                 "bug.", (unsigned long long)
 354                                 sle64_to_cpu(ia->index_block_vcn),
 355                                 (unsigned long long)vcn, dir_ni->mft_no);
 356                 goto unm_err_out;
 357         }
 358         if (le32_to_cpu(ia->index.allocated_size) + 0x18 !=
 359                         dir_ni->itype.index.block_size) {
 360                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
 361                                 "0x%lx has a size (%u) differing from the "
 362                                 "directory specified size (%u). Directory "
 363                                 "inode is corrupt or driver bug.",
 364                                 (unsigned long long)vcn, dir_ni->mft_no,
 365                                 le32_to_cpu(ia->index.allocated_size) + 0x18,
 366                                 dir_ni->itype.index.block_size);
 367                 goto unm_err_out;
 368         }
 369         index_end = (u8*)ia + dir_ni->itype.index.block_size;
 370         if (index_end > kaddr + PAGE_CACHE_SIZE) {
 371                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
 372                                 "0x%lx crosses page boundary. Impossible! "
 373                                 "Cannot access! This is probably a bug in the "
 374                                 "driver.", (unsigned long long)vcn,
 375                                 dir_ni->mft_no);
 376                 goto unm_err_out;
 377         }
 378         index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length);
 379         if (index_end > (u8*)ia + dir_ni->itype.index.block_size) {
 380                 ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of directory "
 381                                 "inode 0x%lx exceeds maximum size.",
 382                                 (unsigned long long)vcn, dir_ni->mft_no);
 383                 goto unm_err_out;
 384         }
 385         /* The first index entry. */
 386         ie = (INDEX_ENTRY*)((u8*)&ia->index +
 387                         le32_to_cpu(ia->index.entries_offset));
 388         /*
 389          * Iterate similar to above big loop but applied to index buffer, thus
 390          * loop until we exceed valid memory (corruption case) or until we
 391          * reach the last entry.
 392          */
 393         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
 394                 /* Bounds check. */
 395                 if ((u8*)ie < (u8*)ia || (u8*)ie +
 396                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
 397                                 (u8*)ie + le16_to_cpu(ie->key_length) >
 398                                 index_end) {
 399                         ntfs_error(sb, "Index entry out of bounds in "
 400                                         "directory inode 0x%lx.",
 401                                         dir_ni->mft_no);
 402                         goto unm_err_out;
 403                 }
 404                 /*
 405                  * The last entry cannot contain a name. It can however contain
 406                  * a pointer to a child node in the B+tree so we just break out.
 407                  */
 408                 if (ie->flags & INDEX_ENTRY_END)
 409                         break;
 410                 /*
 411                  * We perform a case sensitive comparison and if that matches
 412                  * we are done and return the mft reference of the inode (i.e.
 413                  * the inode number together with the sequence number for
 414                  * consistency checking). We convert it to cpu format before
 415                  * returning.
 416                  */
 417                 if (ntfs_are_names_equal(uname, uname_len,
 418                                 (ntfschar*)&ie->key.file_name.file_name,
 419                                 ie->key.file_name.file_name_length,
 420                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len)) {
 421 found_it2:
 422                         /*
 423                          * We have a perfect match, so we don't need to care
 424                          * about having matched imperfectly before, so we can
 425                          * free name and set *res to NULL.
 426                          * However, if the perfect match is a short file name,
 427                          * we need to signal this through *res, so that
 428                          * ntfs_lookup() can fix dcache aliasing issues.
 429                          * As an optimization we just reuse an existing
 430                          * allocation of *res.
 431                          */
 432                         if (ie->key.file_name.file_name_type == FILE_NAME_DOS) {
 433                                 if (!name) {
 434                                         name = kmalloc(sizeof(ntfs_name),
 435                                                         GFP_NOFS);
 436                                         if (!name) {
 437                                                 err = -ENOMEM;
 438                                                 goto unm_err_out;
 439                                         }
 440                                 }
 441                                 name->mref = le64_to_cpu(
 442                                                 ie->data.dir.indexed_file);
 443                                 name->type = FILE_NAME_DOS;
 444                                 name->len = 0;
 445                                 *res = name;
 446                         } else {
 447                                 if (name)
 448                                         kfree(name);
 449                                 *res = NULL;
 450                         }
 451                         mref = le64_to_cpu(ie->data.dir.indexed_file);
 452                         unlock_page(page);
 453                         ntfs_unmap_page(page);
 454                         return mref;
 455                 }
 456                 /*
 457                  * For a case insensitive mount, we also perform a case
 458                  * insensitive comparison (provided the file name is not in the
 459                  * POSIX namespace). If the comparison matches, and the name is
 460                  * in the WIN32 namespace, we cache the filename in *res so
 461                  * that the caller, ntfs_lookup(), can work on it. If the
 462                  * comparison matches, and the name is in the DOS namespace, we
 463                  * only cache the mft reference and the file name type (we set
 464                  * the name length to zero for simplicity).
 465                  */
 466                 if (!NVolCaseSensitive(vol) &&
 467                                 ie->key.file_name.file_name_type &&
 468                                 ntfs_are_names_equal(uname, uname_len,
 469                                 (ntfschar*)&ie->key.file_name.file_name,
 470                                 ie->key.file_name.file_name_length,
 471                                 IGNORE_CASE, vol->upcase, vol->upcase_len)) {
 472                         int name_size = sizeof(ntfs_name);
 473                         u8 type = ie->key.file_name.file_name_type;
 474                         u8 len = ie->key.file_name.file_name_length;
 475
 476                         /* Only one case insensitive matching name allowed. */
 477                         if (name) {
 478                                 ntfs_error(sb, "Found already allocated name "
 479                                                 "in phase 2. Please run chkdsk "
 480                                                 "and if that doesn't find any "
 481                                                 "errors please report you saw "
 482                                                 "this message to "
 483                                                 "linux-ntfs-dev@lists."
 484                                                 "sourceforge.net.");
 485                                 unlock_page(page);
 486                                 ntfs_unmap_page(page);
 487                                 goto dir_err_out;
 488                         }
 489
 490                         if (type != FILE_NAME_DOS)
 491                                 name_size += len * sizeof(ntfschar);
 492                         name = kmalloc(name_size, GFP_NOFS);
 493                         if (!name) {
 494                                 err = -ENOMEM;
 495                                 goto unm_err_out;
 496                         }
 497                         name->mref = le64_to_cpu(ie->data.dir.indexed_file);
 498                         name->type = type;
 499                         if (type != FILE_NAME_DOS) {
 500                                 name->len = len;
 501                                 memcpy(name->name, ie->key.file_name.file_name,
 502                                                 len * sizeof(ntfschar));
 503                         } else
 504                                 name->len = 0;
 505                         *res = name;
 506                 }
 507                 /*
 508                  * Not a perfect match, need to do full blown collation so we
 509                  * know which way in the B+tree we have to go.
 510                  */
 511                 rc = ntfs_collate_names(uname, uname_len,
 512                                 (ntfschar*)&ie->key.file_name.file_name,
 513                                 ie->key.file_name.file_name_length, 1,
 514                                 IGNORE_CASE, vol->upcase, vol->upcase_len);
 515                 /*
 516                  * If uname collates before the name of the current entry, there
 517                  * is definitely no such name in this index but we might need to
 518                  * descend into the B+tree so we just break out of the loop.
 519                  */
 520                 if (rc == -1)
 521                         break;
 522                 /* The names are not equal, continue the search. */
 523                 if (rc)
 524                         continue;
 525                 /*
 526                  * Names match with case insensitive comparison, now try the
 527                  * case sensitive comparison, which is required for proper
 528                  * collation.
 529                  */
 530                 rc = ntfs_collate_names(uname, uname_len,
 531                                 (ntfschar*)&ie->key.file_name.file_name,
 532                                 ie->key.file_name.file_name_length, 1,
 533                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len);
 534                 if (rc == -1)
 535                         break;
 536                 if (rc)
 537                         continue;
 538                 /*
 539                  * Perfect match, this will never happen as the
 540                  * ntfs_are_names_equal() call will have gotten a match but we
 541                  * still treat it correctly.
 542                  */
 543                 goto found_it2;
 544         }
 545         /*
 546          * We have finished with this index buffer without success. Check for
 547          * the presence of a child node.
 548          */
 549         if (ie->flags & INDEX_ENTRY_NODE) {
 550                 if ((ia->index.flags & NODE_MASK) == LEAF_NODE) {
 551                         ntfs_error(sb, "Index entry with child node found in "
 552                                         "a leaf node in directory inode 0x%lx.",
 553                                         dir_ni->mft_no);
 554                         goto unm_err_out;
 555                 }
 556                 /* Child node present, descend into it. */
 557                 old_vcn = vcn;
 558                 vcn = sle64_to_cpup((sle64*)((u8*)ie +
 559                                 le16_to_cpu(ie->length) - 8));
 560                 if (vcn >= 0) {
 561                         /* If vcn is in the same page cache page as old_vcn we
 562                          * recycle the mapped page. */
 563                         if (old_vcn << vol->cluster_size_bits >>
 564                                         PAGE_CACHE_SHIFT == vcn <<
 565                                         vol->cluster_size_bits >>
 566                                         PAGE_CACHE_SHIFT)
 567                                 goto fast_descend_into_child_node;
 568                         unlock_page(page);
 569                         ntfs_unmap_page(page);
 570                         goto descend_into_child_node;
 571                 }
 572                 ntfs_error(sb, "Negative child node vcn in directory inode "
 573                                 "0x%lx.", dir_ni->mft_no);
 574                 goto unm_err_out;
 575         }
 576         /*
 577          * No child node present, return -ENOENT, unless we have got a matching
 578          * name cached in name in which case return the mft reference
 579          * associated with it.
 580          */
 581         if (name) {
 582                 unlock_page(page);
 583                 ntfs_unmap_page(page);
 584                 return name->mref;
 585         }
 586         ntfs_debug("Entry not found.");
 587         err = -ENOENT;
 588 unm_err_out:
 589         unlock_page(page);
 590         ntfs_unmap_page(page);
 591 err_out:
 592         if (!err)
 593                 err = -EIO;
 594         if (ctx)
 595                 ntfs_attr_put_search_ctx(ctx);
 596         if (m)
 597                 unmap_mft_record(dir_ni);
 598         if (name) {
 599                 kfree(name);
 600                 *res = NULL;
 601         }
 602         return ERR_MREF(err);
 603 dir_err_out:
 604         ntfs_error(sb, "Corrupt directory.  Aborting lookup.");
 605         goto err_out;
 606 }
 607
 608 #if 0
 609
 610 // TODO: (AIA)
 611 // The algorithm embedded in this code will be required for the time when we
 612 // want to support adding of entries to directories, where we require correct
 613 // collation of file names in order not to cause corruption of the file system.
 614
 615 /**
 616  * ntfs_lookup_inode_by_name - find an inode in a directory given its name
 617  * @dir_ni:     ntfs inode of the directory in which to search for the name
 618  * @uname:      Unicode name for which to search in the directory
 619  * @uname_len:  length of the name @uname in Unicode characters
 620  *
 621  * Look for an inode with name @uname in the directory with inode @dir_ni.
 622  * ntfs_lookup_inode_by_name() walks the contents of the directory looking for
 623  * the Unicode name. If the name is found in the directory, the corresponding
 624  * inode number (>= 0) is returned as a mft reference in cpu format, i.e. it
 625  * is a 64-bit number containing the sequence number.
 626  *
 627  * On error, a negative value is returned corresponding to the error code. In
 628  * particular if the inode is not found -ENOENT is returned. Note that you
 629  * can't just check the return value for being negative, you have to check the
 630  * inode number for being negative which you can extract using MREC(return
 631  * value).
 632  *
 633  * Note, @uname_len does not include the (optional) terminating NULL character.
 634  */
 635 u64 ntfs_lookup_inode_by_name(ntfs_inode *dir_ni, const ntfschar *uname,
 636                 const int uname_len)
 637 {
 638         ntfs_volume *vol = dir_ni->vol;
 639         struct super_block *sb = vol->sb;
 640         MFT_RECORD *m;
 641         INDEX_ROOT *ir;
 642         INDEX_ENTRY *ie;
 643         INDEX_ALLOCATION *ia;
 644         u8 *index_end;
 645         u64 mref;
 646         ntfs_attr_search_ctx *ctx;
 647         int err, rc;
 648         IGNORE_CASE_BOOL ic;
 649         VCN vcn, old_vcn;
 650         struct address_space *ia_mapping;
 651         struct page *page;
 652         u8 *kaddr;
 653
 654         /* Get hold of the mft record for the directory. */
 655         m = map_mft_record(dir_ni);
 656         if (IS_ERR(m)) {
 657                 ntfs_error(sb, "map_mft_record() failed with error code %ld.",
 658                                 -PTR_ERR(m));
 659                 return ERR_MREF(PTR_ERR(m));
 660         }
 661         ctx = ntfs_attr_get_search_ctx(dir_ni, m);
 662         if (!ctx) {
 663                 err = -ENOMEM;
 664                 goto err_out;
 665         }
 666         /* Find the index root attribute in the mft record. */
 667         err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL,
 668                         0, ctx);
 669         if (unlikely(err)) {
 670                 if (err == -ENOENT) {
 671                         ntfs_error(sb, "Index root attribute missing in "
 672                                         "directory inode 0x%lx.",
 673                                         dir_ni->mft_no);
 674                         err = -EIO;
 675                 }
 676                 goto err_out;
 677         }
 678         /* Get to the index root value (it's been verified in read_inode). */
 679         ir = (INDEX_ROOT*)((u8*)ctx->attr +
 680                         le16_to_cpu(ctx->attr->data.resident.value_offset));
 681         index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
 682         /* The first index entry. */
 683         ie = (INDEX_ENTRY*)((u8*)&ir->index +
 684                         le32_to_cpu(ir->index.entries_offset));
 685         /*
 686          * Loop until we exceed valid memory (corruption case) or until we
 687          * reach the last entry.
 688          */
 689         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
 690                 /* Bounds checks. */
 691                 if ((u8*)ie < (u8*)ctx->mrec || (u8*)ie +
 692                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
 693                                 (u8*)ie + le16_to_cpu(ie->key_length) >
 694                                 index_end)
 695                         goto dir_err_out;
 696                 /*
 697                  * The last entry cannot contain a name. It can however contain
 698                  * a pointer to a child node in the B+tree so we just break out.
 699                  */
 700                 if (ie->flags & INDEX_ENTRY_END)
 701                         break;
 702                 /*
 703                  * If the current entry has a name type of POSIX, the name is
 704                  * case sensitive and not otherwise. This has the effect of us
 705                  * not being able to access any POSIX file names which collate
 706                  * after the non-POSIX one when they only differ in case, but
 707                  * anyone doing screwy stuff like that deserves to burn in
 708                  * hell... Doing that kind of stuff on NT4 actually causes
 709                  * corruption on the partition even when using SP6a and Linux
 710                  * is not involved at all.
 711                  */
 712                 ic = ie->key.file_name.file_name_type ? IGNORE_CASE :
 713                                 CASE_SENSITIVE;
 714                 /*
 715                  * If the names match perfectly, we are done and return the
 716                  * mft reference of the inode (i.e. the inode number together
 717                  * with the sequence number for consistency checking. We
 718                  * convert it to cpu format before returning.
 719                  */
 720                 if (ntfs_are_names_equal(uname, uname_len,
 721                                 (ntfschar*)&ie->key.file_name.file_name,
 722                                 ie->key.file_name.file_name_length, ic,
 723                                 vol->upcase, vol->upcase_len)) {
 724 found_it:
 725                         mref = le64_to_cpu(ie->data.dir.indexed_file);
 726                         ntfs_attr_put_search_ctx(ctx);
 727                         unmap_mft_record(dir_ni);
 728                         return mref;
 729                 }
 730                 /*
 731                  * Not a perfect match, need to do full blown collation so we
 732                  * know which way in the B+tree we have to go.
 733                  */
 734                 rc = ntfs_collate_names(uname, uname_len,
 735                                 (ntfschar*)&ie->key.file_name.file_name,
 736                                 ie->key.file_name.file_name_length, 1,
 737                                 IGNORE_CASE, vol->upcase, vol->upcase_len);
 738                 /*
 739                  * If uname collates before the name of the current entry, there
 740                  * is definitely no such name in this index but we might need to
 741                  * descend into the B+tree so we just break out of the loop.
 742                  */
 743                 if (rc == -1)
 744                         break;
 745                 /* The names are not equal, continue the search. */
 746                 if (rc)
 747                         continue;
 748                 /*
 749                  * Names match with case insensitive comparison, now try the
 750                  * case sensitive comparison, which is required for proper
 751                  * collation.
 752                  */
 753                 rc = ntfs_collate_names(uname, uname_len,
 754                                 (ntfschar*)&ie->key.file_name.file_name,
 755                                 ie->key.file_name.file_name_length, 1,
 756                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len);
 757                 if (rc == -1)
 758                         break;
 759                 if (rc)
 760                         continue;
 761                 /*
 762                  * Perfect match, this will never happen as the
 763                  * ntfs_are_names_equal() call will have gotten a match but we
 764                  * still treat it correctly.
 765                  */
 766                 goto found_it;
 767         }
 768         /*
 769          * We have finished with this index without success. Check for the
 770          * presence of a child node.
 771          */
 772         if (!(ie->flags & INDEX_ENTRY_NODE)) {
 773                 /* No child node, return -ENOENT. */
 774                 err = -ENOENT;
 775                 goto err_out;
 776         } /* Child node present, descend into it. */
 777         /* Consistency check: Verify that an index allocation exists. */
 778         if (!NInoIndexAllocPresent(dir_ni)) {
 779                 ntfs_error(sb, "No index allocation attribute but index entry "
 780                                 "requires one. Directory inode 0x%lx is "
 781                                 "corrupt or driver bug.", dir_ni->mft_no);
 782                 goto err_out;
 783         }
 784         /* Get the starting vcn of the index_block holding the child node. */
 785         vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8);
 786         ia_mapping = VFS_I(dir_ni)->i_mapping;
 787         /*
 788          * We are done with the index root and the mft record. Release them,
 789          * otherwise we deadlock with ntfs_map_page().
 790          */
 791         ntfs_attr_put_search_ctx(ctx);
 792         unmap_mft_record(dir_ni);
 793         m = NULL;
 794         ctx = NULL;
 795 descend_into_child_node:
 796         /*
 797          * Convert vcn to index into the index allocation attribute in units
 798          * of PAGE_CACHE_SIZE and map the page cache page, reading it from
 799          * disk if necessary.
 800          */
 801         page = ntfs_map_page(ia_mapping, vcn <<
 802                         dir_ni->itype.index.vcn_size_bits >> PAGE_CACHE_SHIFT);
 803         if (IS_ERR(page)) {
 804                 ntfs_error(sb, "Failed to map directory index page, error %ld.",
 805                                 -PTR_ERR(page));
 806                 err = PTR_ERR(page);
 807                 goto err_out;
 808         }
 809         lock_page(page);
 810         kaddr = (u8*)page_address(page);
 811 fast_descend_into_child_node:
 812         /* Get to the index allocation block. */
 813         ia = (INDEX_ALLOCATION*)(kaddr + ((vcn <<
 814                         dir_ni->itype.index.vcn_size_bits) & ~PAGE_CACHE_MASK));
 815         /* Bounds checks. */
 816         if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) {
 817                 ntfs_error(sb, "Out of bounds check failed. Corrupt directory "
 818                                 "inode 0x%lx or driver bug.", dir_ni->mft_no);
 819                 goto unm_err_out;
 820         }
 821         /* Catch multi sector transfer fixup errors. */
 822         if (unlikely(!ntfs_is_indx_record(ia->magic))) {
 823                 ntfs_error(sb, "Directory index record with vcn 0x%llx is "
 824                                 "corrupt.  Corrupt inode 0x%lx.  Run chkdsk.",
 825                                 (unsigned long long)vcn, dir_ni->mft_no);
 826                 goto unm_err_out;
 827         }
 828         if (sle64_to_cpu(ia->index_block_vcn) != vcn) {
 829                 ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is "
 830                                 "different from expected VCN (0x%llx). "
 831                                 "Directory inode 0x%lx is corrupt or driver "
 832                                 "bug.", (unsigned long long)
 833                                 sle64_to_cpu(ia->index_block_vcn),
 834                                 (unsigned long long)vcn, dir_ni->mft_no);
 835                 goto unm_err_out;
 836         }
 837         if (le32_to_cpu(ia->index.allocated_size) + 0x18 !=
 838                         dir_ni->itype.index.block_size) {
 839                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
 840                                 "0x%lx has a size (%u) differing from the "
 841                                 "directory specified size (%u). Directory "
 842                                 "inode is corrupt or driver bug.",
 843                                 (unsigned long long)vcn, dir_ni->mft_no,
 844                                 le32_to_cpu(ia->index.allocated_size) + 0x18,
 845                                 dir_ni->itype.index.block_size);
 846                 goto unm_err_out;
 847         }
 848         index_end = (u8*)ia + dir_ni->itype.index.block_size;
 849         if (index_end > kaddr + PAGE_CACHE_SIZE) {
 850                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
 851                                 "0x%lx crosses page boundary. Impossible! "
 852                                 "Cannot access! This is probably a bug in the "
 853                                 "driver.", (unsigned long long)vcn,
 854                                 dir_ni->mft_no);
 855                 goto unm_err_out;
 856         }
 857         index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length);
 858         if (index_end > (u8*)ia + dir_ni->itype.index.block_size) {
 859                 ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of directory "
 860                                 "inode 0x%lx exceeds maximum size.",
 861                                 (unsigned long long)vcn, dir_ni->mft_no);
 862                 goto unm_err_out;
 863         }
 864         /* The first index entry. */
 865         ie = (INDEX_ENTRY*)((u8*)&ia->index +
 866                         le32_to_cpu(ia->index.entries_offset));
 867         /*
 868          * Iterate similar to above big loop but applied to index buffer, thus
 869          * loop until we exceed valid memory (corruption case) or until we
 870          * reach the last entry.
 871          */
 872         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
 873                 /* Bounds check. */
 874                 if ((u8*)ie < (u8*)ia || (u8*)ie +
 875                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
 876                                 (u8*)ie + le16_to_cpu(ie->key_length) >
 877                                 index_end) {
 878                         ntfs_error(sb, "Index entry out of bounds in "
 879                                         "directory inode 0x%lx.",
 880                                         dir_ni->mft_no);
 881                         goto unm_err_out;
 882                 }
 883                 /*
 884                  * The last entry cannot contain a name. It can however contain
 885                  * a pointer to a child node in the B+tree so we just break out.
 886                  */
 887                 if (ie->flags & INDEX_ENTRY_END)
 888                         break;
 889                 /*
 890                  * If the current entry has a name type of POSIX, the name is
 891                  * case sensitive and not otherwise. This has the effect of us
 892                  * not being able to access any POSIX file names which collate
 893                  * after the non-POSIX one when they only differ in case, but
 894                  * anyone doing screwy stuff like that deserves to burn in
 895                  * hell... Doing that kind of stuff on NT4 actually causes
 896                  * corruption on the partition even when using SP6a and Linux
 897                  * is not involved at all.
 898                  */
 899                 ic = ie->key.file_name.file_name_type ? IGNORE_CASE :
 900                                 CASE_SENSITIVE;
 901                 /*
 902                  * If the names match perfectly, we are done and return the
 903                  * mft reference of the inode (i.e. the inode number together
 904                  * with the sequence number for consistency checking. We
 905                  * convert it to cpu format before returning.
 906                  */
 907                 if (ntfs_are_names_equal(uname, uname_len,
 908                                 (ntfschar*)&ie->key.file_name.file_name,
 909                                 ie->key.file_name.file_name_length, ic,
 910                                 vol->upcase, vol->upcase_len)) {
 911 found_it2:
 912                         mref = le64_to_cpu(ie->data.dir.indexed_file);
 913                         unlock_page(page);
 914                         ntfs_unmap_page(page);
 915                         return mref;
 916                 }
 917                 /*
 918                  * Not a perfect match, need to do full blown collation so we
 919                  * know which way in the B+tree we have to go.
 920                  */
 921                 rc = ntfs_collate_names(uname, uname_len,
 922                                 (ntfschar*)&ie->key.file_name.file_name,
 923                                 ie->key.file_name.file_name_length, 1,
 924                                 IGNORE_CASE, vol->upcase, vol->upcase_len);
 925                 /*
 926                  * If uname collates before the name of the current entry, there
 927                  * is definitely no such name in this index but we might need to
 928                  * descend into the B+tree so we just break out of the loop.
 929                  */
 930                 if (rc == -1)
 931                         break;
 932                 /* The names are not equal, continue the search. */
 933                 if (rc)
 934                         continue;
 935                 /*
 936                  * Names match with case insensitive comparison, now try the
 937                  * case sensitive comparison, which is required for proper
 938                  * collation.
 939                  */
 940                 rc = ntfs_collate_names(uname, uname_len,
 941                                 (ntfschar*)&ie->key.file_name.file_name,
 942                                 ie->key.file_name.file_name_length, 1,
 943                                 CASE_SENSITIVE, vol->upcase, vol->upcase_len);
 944                 if (rc == -1)
 945                         break;
 946                 if (rc)
 947                         continue;
 948                 /*
 949                  * Perfect match, this will never happen as the
 950                  * ntfs_are_names_equal() call will have gotten a match but we
 951                  * still treat it correctly.
 952                  */
 953                 goto found_it2;
 954         }
 955         /*
 956          * We have finished with this index buffer without success. Check for
 957          * the presence of a child node.
 958          */
 959         if (ie->flags & INDEX_ENTRY_NODE) {
 960                 if ((ia->index.flags & NODE_MASK) == LEAF_NODE) {
 961                         ntfs_error(sb, "Index entry with child node found in "
 962                                         "a leaf node in directory inode 0x%lx.",
 963                                         dir_ni->mft_no);
 964                         goto unm_err_out;
 965                 }
 966                 /* Child node present, descend into it. */
 967                 old_vcn = vcn;
 968                 vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8);
 969                 if (vcn >= 0) {
 970                         /* If vcn is in the same page cache page as old_vcn we
 971                          * recycle the mapped page. */
 972                         if (old_vcn << vol->cluster_size_bits >>
 973                                         PAGE_CACHE_SHIFT == vcn <<
 974                                         vol->cluster_size_bits >>
 975                                         PAGE_CACHE_SHIFT)
 976                                 goto fast_descend_into_child_node;
 977                         unlock_page(page);
 978                         ntfs_unmap_page(page);
 979                         goto descend_into_child_node;
 980                 }
 981                 ntfs_error(sb, "Negative child node vcn in directory inode "
 982                                 "0x%lx.", dir_ni->mft_no);
 983                 goto unm_err_out;
 984         }
 985         /* No child node, return -ENOENT. */
 986         ntfs_debug("Entry not found.");
 987         err = -ENOENT;
 988 unm_err_out:
 989         unlock_page(page);
 990         ntfs_unmap_page(page);
 991 err_out:
 992         if (!err)
 993                 err = -EIO;
 994         if (ctx)
 995                 ntfs_attr_put_search_ctx(ctx);
 996         if (m)
 997                 unmap_mft_record(dir_ni);
 998         return ERR_MREF(err);
 999 dir_err_out:
1000         ntfs_error(sb, "Corrupt directory. Aborting lookup.");
1001         goto err_out;
1002 }
1003
1004 #endif
1005
1006 /**
1007  * ntfs_filldir - ntfs specific filldir method
1008  * @vol:        current ntfs volume
1009  * @fpos:       position in the directory
1010  * @ndir:       ntfs inode of current directory
1011  * @ia_page:    page in which the index allocation buffer @ie is in resides
1012  * @ie:         current index entry
1013  * @name:       buffer to use for the converted name
1014  * @dirent:     vfs filldir callback context
1015  * @filldir:    vfs filldir callback
1016  *
1017  * Convert the Unicode @name to the loaded NLS and pass it to the @filldir
1018  * callback.
1019  *
1020  * If @ia_page is not NULL it is the locked page containing the index
1021  * allocation block containing the index entry @ie.
1022  *
1023  * Note, we drop (and then reacquire) the page lock on @ia_page across the
1024  * @filldir() call otherwise we would deadlock with NFSd when it calls ->lookup
1025  * since ntfs_lookup() will lock the same page.  As an optimization, we do not
1026  * retake the lock if we are returning a non-zero value as ntfs_readdir()
1027  * would need to drop the lock immediately anyway.
1028  */
1029 static inline int ntfs_filldir(ntfs_volume *vol, loff_t fpos,
1030                 ntfs_inode *ndir, struct page *ia_page, INDEX_ENTRY *ie,
1031                 u8 *name, void *dirent, filldir_t filldir)
1032 {
1033         unsigned long mref;
1034         int name_len, rc;
1035         unsigned dt_type;
1036         FILE_NAME_TYPE_FLAGS name_type;
1037
1038         name_type = ie->key.file_name.file_name_type;
1039         if (name_type == FILE_NAME_DOS) {
1040                 ntfs_debug("Skipping DOS name space entry.");
1041                 return 0;
1042         }
1043         if (MREF_LE(ie->data.dir.indexed_file) == FILE_root) {
1044                 ntfs_debug("Skipping root directory self reference entry.");
1045                 return 0;
1046         }
1047         if (MREF_LE(ie->data.dir.indexed_file) < FILE_first_user &&
1048                         !NVolShowSystemFiles(vol)) {
1049                 ntfs_debug("Skipping system file.");
1050                 return 0;
1051         }
1052         name_len = ntfs_ucstonls(vol, (ntfschar*)&ie->key.file_name.file_name,
1053                         ie->key.file_name.file_name_length, &name,
1054                         NTFS_MAX_NAME_LEN * NLS_MAX_CHARSET_SIZE + 1);
1055         if (name_len <= 0) {
1056                 ntfs_debug("Skipping unrepresentable file.");
1057                 return 0;
1058         }
1059         if (ie->key.file_name.file_attributes &
1060                         FILE_ATTR_DUP_FILE_NAME_INDEX_PRESENT)
1061                 dt_type = DT_DIR;
1062         else
1063                 dt_type = DT_REG;
1064         mref = MREF_LE(ie->data.dir.indexed_file);
1065         /*
1066          * Drop the page lock otherwise we deadlock with NFS when it calls
1067          * ->lookup since ntfs_lookup() will lock the same page.
1068          */
1069         if (ia_page)
1070                 unlock_page(ia_page);
1071         ntfs_debug("Calling filldir for %s with len %i, fpos 0x%llx, inode "
1072                         "0x%lx, DT_%s.", name, name_len, fpos, mref,
1073                         dt_type == DT_DIR ? "DIR" : "REG");
1074         rc = filldir(dirent, name, name_len, fpos, mref, dt_type);
1075         /* Relock the page but not if we are aborting ->readdir. */
1076         if (!rc && ia_page)
1077                 lock_page(ia_page);
1078         return rc;
1079 }
1080
1081 /*
1082  * We use the same basic approach as the old NTFS driver, i.e. we parse the
1083  * index root entries and then the index allocation entries that are marked
1084  * as in use in the index bitmap.
1085  *
1086  * While this will return the names in random order this doesn't matter for
1087  * ->readdir but OTOH results in a faster ->readdir.
1088  *
1089  * VFS calls ->readdir without BKL but with i_sem held. This protects the VFS
1090  * parts (e.g. ->f_pos and ->i_size, and it also protects against directory
1091  * modifications).
1092  *
1093  * Locking:  - Caller must hold i_sem on the directory.
1094  *           - Each page cache page in the index allocation mapping must be
1095  *             locked whilst being accessed otherwise we may find a corrupt
1096  *             page due to it being under ->writepage at the moment which
1097  *             applies the mst protection fixups before writing out and then
1098  *             removes them again after the write is complete after which it
1099  *             unlocks the page.
1100  */
1101 static int ntfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1102 {
1103         s64 ia_pos, ia_start, prev_ia_pos, bmp_pos;
1104         loff_t fpos;
1105         struct inode *bmp_vi, *vdir = filp->f_dentry->d_inode;
1106         struct super_block *sb = vdir->i_sb;
1107         ntfs_inode *ndir = NTFS_I(vdir);
1108         ntfs_volume *vol = NTFS_SB(sb);
1109         MFT_RECORD *m;
1110         INDEX_ROOT *ir = NULL;
1111         INDEX_ENTRY *ie;
1112         INDEX_ALLOCATION *ia;
1113         u8 *name = NULL;
1114         int rc, err, ir_pos, cur_bmp_pos;
1115         struct address_space *ia_mapping, *bmp_mapping;
1116         struct page *bmp_page = NULL, *ia_page = NULL;
1117         u8 *kaddr, *bmp, *index_end;
1118         ntfs_attr_search_ctx *ctx;
1119
1120         fpos = filp->f_pos;
1121         ntfs_debug("Entering for inode 0x%lx, fpos 0x%llx.",
1122                         vdir->i_ino, fpos);
1123         rc = err = 0;
1124         /* Are we at end of dir yet? */
1125         if (fpos >= vdir->i_size + vol->mft_record_size)
1126                 goto done;
1127         /* Emulate . and .. for all directories. */
1128         if (!fpos) {
1129                 ntfs_debug("Calling filldir for . with len 1, fpos 0x0, "
1130                                 "inode 0x%lx, DT_DIR.", vdir->i_ino);
1131                 rc = filldir(dirent, ".", 1, fpos, vdir->i_ino, DT_DIR);
1132                 if (rc)
1133                         goto done;
1134                 fpos++;
1135         }
1136         if (fpos == 1) {
1137                 ntfs_debug("Calling filldir for .. with len 2, fpos 0x1, "
1138                                 "inode 0x%lx, DT_DIR.",
1139                                 parent_ino(filp->f_dentry));
1140                 rc = filldir(dirent, "..", 2, fpos,
1141                                 parent_ino(filp->f_dentry), DT_DIR);
1142                 if (rc)
1143                         goto done;
1144                 fpos++;
1145         }
1146         m = NULL;
1147         ctx = NULL;
1148         /*
1149          * Allocate a buffer to store the current name being processed
1150          * converted to format determined by current NLS.
1151          */
1152         name = (u8*)kmalloc(NTFS_MAX_NAME_LEN * NLS_MAX_CHARSET_SIZE + 1,
1153                         GFP_NOFS);
1154         if (unlikely(!name)) {
1155                 err = -ENOMEM;
1156                 goto err_out;
1157         }
1158         /* Are we jumping straight into the index allocation attribute? */
1159         if (fpos >= vol->mft_record_size)
1160                 goto skip_index_root;
1161         /* Get hold of the mft record for the directory. */
1162         m = map_mft_record(ndir);
1163         if (IS_ERR(m)) {
1164                 err = PTR_ERR(m);
1165                 m = NULL;
1166                 goto err_out;
1167         }
1168         ctx = ntfs_attr_get_search_ctx(ndir, m);
1169         if (unlikely(!ctx)) {
1170                 err = -ENOMEM;
1171                 goto err_out;
1172         }
1173         /* Get the offset into the index root attribute. */
1174         ir_pos = (s64)fpos;
1175         /* Find the index root attribute in the mft record. */
1176         err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE, 0, NULL,
1177                         0, ctx);
1178         if (unlikely(err)) {
1179                 ntfs_error(sb, "Index root attribute missing in directory "
1180                                 "inode 0x%lx.", vdir->i_ino);
1181                 goto err_out;
1182         }
1183         /*
1184          * Copy the index root attribute value to a buffer so that we can put
1185          * the search context and unmap the mft record before calling the
1186          * filldir() callback.  We need to do this because of NFSd which calls
1187          * ->lookup() from its filldir callback() and this causes NTFS to
1188          * deadlock as ntfs_lookup() maps the mft record of the directory and
1189          * we have got it mapped here already.  The only solution is for us to
1190          * unmap the mft record here so that a call to ntfs_lookup() is able to
1191          * map the mft record without deadlocking.
1192          */
1193         rc = le32_to_cpu(ctx->attr->data.resident.value_length);
1194         ir = (INDEX_ROOT*)kmalloc(rc, GFP_NOFS);
1195         if (unlikely(!ir)) {
1196                 err = -ENOMEM;
1197                 goto err_out;
1198         }
1199         /* Copy the index root value (it has been verified in read_inode). */
1200         memcpy(ir, (u8*)ctx->attr +
1201                         le16_to_cpu(ctx->attr->data.resident.value_offset), rc);
1202         ntfs_attr_put_search_ctx(ctx);
1203         unmap_mft_record(ndir);
1204         ctx = NULL;
1205         m = NULL;
1206         index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
1207         /* The first index entry. */
1208         ie = (INDEX_ENTRY*)((u8*)&ir->index +
1209                         le32_to_cpu(ir->index.entries_offset));
1210         /*
1211          * Loop until we exceed valid memory (corruption case) or until we
1212          * reach the last entry or until filldir tells us it has had enough
1213          * or signals an error (both covered by the rc test).
1214          */
1215         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
1216                 ntfs_debug("In index root, offset 0x%zx.", (u8*)ie - (u8*)ir);
1217                 /* Bounds checks. */
1218                 if (unlikely((u8*)ie < (u8*)ir || (u8*)ie +
1219                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
1220                                 (u8*)ie + le16_to_cpu(ie->key_length) >
1221                                 index_end))
1222                         goto err_out;
1223                 /* The last entry cannot contain a name. */
1224                 if (ie->flags & INDEX_ENTRY_END)
1225                         break;
1226                 /* Skip index root entry if continuing previous readdir. */
1227                 if (ir_pos > (u8*)ie - (u8*)ir)
1228                         continue;
1229                 /* Advance the position even if going to skip the entry. */
1230                 fpos = (u8*)ie - (u8*)ir;
1231                 /* Submit the name to the filldir callback. */
1232                 rc = ntfs_filldir(vol, fpos, ndir, NULL, ie, name, dirent,
1233                                 filldir);
1234                 if (rc) {
1235                         kfree(ir);
1236                         goto abort;
1237                 }
1238         }
1239         /* We are done with the index root and can free the buffer. */
1240         kfree(ir);
1241         ir = NULL;
1242         /* If there is no index allocation attribute we are finished. */
1243         if (!NInoIndexAllocPresent(ndir))
1244                 goto EOD;
1245         /* Advance fpos to the beginning of the index allocation. */
1246         fpos = vol->mft_record_size;
1247 skip_index_root:
1248         kaddr = NULL;
1249         prev_ia_pos = -1LL;
1250         /* Get the offset into the index allocation attribute. */
1251         ia_pos = (s64)fpos - vol->mft_record_size;
1252         ia_mapping = vdir->i_mapping;
1253         bmp_vi = ndir->itype.index.bmp_ino;
1254         if (unlikely(!bmp_vi)) {
1255                 ntfs_debug("Inode 0x%lx, regetting index bitmap.", vdir->i_ino);
1256                 bmp_vi = ntfs_attr_iget(vdir, AT_BITMAP, I30, 4);
1257                 if (IS_ERR(bmp_vi)) {
1258                         ntfs_error(sb, "Failed to get bitmap attribute.");
1259                         err = PTR_ERR(bmp_vi);
1260                         goto err_out;
1261                 }
1262                 ndir->itype.index.bmp_ino = bmp_vi;
1263         }
1264         bmp_mapping = bmp_vi->i_mapping;
1265         /* Get the starting bitmap bit position and sanity check it. */
1266         bmp_pos = ia_pos >> ndir->itype.index.block_size_bits;
1267         if (unlikely(bmp_pos >> 3 >= bmp_vi->i_size)) {
1268                 ntfs_error(sb, "Current index allocation position exceeds "
1269                                 "index bitmap size.");
1270                 goto err_out;
1271         }
1272         /* Get the starting bit position in the current bitmap page. */
1273         cur_bmp_pos = bmp_pos & ((PAGE_CACHE_SIZE * 8) - 1);
1274         bmp_pos &= ~(u64)((PAGE_CACHE_SIZE * 8) - 1);
1275 get_next_bmp_page:
1276         ntfs_debug("Reading bitmap with page index 0x%llx, bit ofs 0x%llx",
1277                         (unsigned long long)bmp_pos >> (3 + PAGE_CACHE_SHIFT),
1278                         (unsigned long long)bmp_pos &
1279                         (unsigned long long)((PAGE_CACHE_SIZE * 8) - 1));
1280         bmp_page = ntfs_map_page(bmp_mapping,
1281                         bmp_pos >> (3 + PAGE_CACHE_SHIFT));
1282         if (IS_ERR(bmp_page)) {
1283                 ntfs_error(sb, "Reading index bitmap failed.");
1284                 err = PTR_ERR(bmp_page);
1285                 bmp_page = NULL;
1286                 goto err_out;
1287         }
1288         bmp = (u8*)page_address(bmp_page);
1289         /* Find next index block in use. */
1290         while (!(bmp[cur_bmp_pos >> 3] & (1 << (cur_bmp_pos & 7)))) {
1291 find_next_index_buffer:
1292                 cur_bmp_pos++;
1293                 /*
1294                  * If we have reached the end of the bitmap page, get the next
1295                  * page, and put away the old one.
1296                  */
1297                 if (unlikely((cur_bmp_pos >> 3) >= PAGE_CACHE_SIZE)) {
1298                         ntfs_unmap_page(bmp_page);
1299                         bmp_pos += PAGE_CACHE_SIZE * 8;
1300                         cur_bmp_pos = 0;
1301                         goto get_next_bmp_page;
1302                 }
1303                 /* If we have reached the end of the bitmap, we are done. */
1304                 if (unlikely(((bmp_pos + cur_bmp_pos) >> 3) >= vdir->i_size))
1305                         goto unm_EOD;
1306                 ia_pos = (bmp_pos + cur_bmp_pos) <<
1307                                 ndir->itype.index.block_size_bits;
1308         }
1309         ntfs_debug("Handling index buffer 0x%llx.",
1310                         (unsigned long long)bmp_pos + cur_bmp_pos);
1311         /* If the current index buffer is in the same page we reuse the page. */
1312         if ((prev_ia_pos & PAGE_CACHE_MASK) != (ia_pos & PAGE_CACHE_MASK)) {
1313                 prev_ia_pos = ia_pos;
1314                 if (likely(ia_page != NULL)) {
1315                         unlock_page(ia_page);
1316                         ntfs_unmap_page(ia_page);
1317                 }
1318                 /*
1319                  * Map the page cache page containing the current ia_pos,
1320                  * reading it from disk if necessary.
1321                  */
1322                 ia_page = ntfs_map_page(ia_mapping, ia_pos >> PAGE_CACHE_SHIFT);
1323                 if (IS_ERR(ia_page)) {
1324                         ntfs_error(sb, "Reading index allocation data failed.");
1325                         err = PTR_ERR(ia_page);
1326                         ia_page = NULL;
1327                         goto err_out;
1328                 }
1329                 lock_page(ia_page);
1330                 kaddr = (u8*)page_address(ia_page);
1331         }
1332         /* Get the current index buffer. */
1333         ia = (INDEX_ALLOCATION*)(kaddr + (ia_pos & ~PAGE_CACHE_MASK &
1334                         ~(s64)(ndir->itype.index.block_size - 1)));
1335         /* Bounds checks. */
1336         if (unlikely((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE)) {
1337                 ntfs_error(sb, "Out of bounds check failed. Corrupt directory "
1338                                 "inode 0x%lx or driver bug.", vdir->i_ino);
1339                 goto err_out;
1340         }
1341         /* Catch multi sector transfer fixup errors. */
1342         if (unlikely(!ntfs_is_indx_record(ia->magic))) {
1343                 ntfs_error(sb, "Directory index record with vcn 0x%llx is "
1344                                 "corrupt.  Corrupt inode 0x%lx.  Run chkdsk.",
1345                                 (unsigned long long)ia_pos >>
1346                                 ndir->itype.index.vcn_size_bits, vdir->i_ino);
1347                 goto err_out;
1348         }
1349         if (unlikely(sle64_to_cpu(ia->index_block_vcn) != (ia_pos &
1350                         ~(s64)(ndir->itype.index.block_size - 1)) >>
1351                         ndir->itype.index.vcn_size_bits)) {
1352                 ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is "
1353                                 "different from expected VCN (0x%llx). "
1354                                 "Directory inode 0x%lx is corrupt or driver "
1355                                 "bug. ", (unsigned long long)
1356                                 sle64_to_cpu(ia->index_block_vcn),
1357                                 (unsigned long long)ia_pos >>
1358                                 ndir->itype.index.vcn_size_bits, vdir->i_ino);
1359                 goto err_out;
1360         }
1361         if (unlikely(le32_to_cpu(ia->index.allocated_size) + 0x18 !=
1362                         ndir->itype.index.block_size)) {
1363                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
1364                                 "0x%lx has a size (%u) differing from the "
1365                                 "directory specified size (%u). Directory "
1366                                 "inode is corrupt or driver bug.",
1367                                 (unsigned long long)ia_pos >>
1368                                 ndir->itype.index.vcn_size_bits, vdir->i_ino,
1369                                 le32_to_cpu(ia->index.allocated_size) + 0x18,
1370                                 ndir->itype.index.block_size);
1371                 goto err_out;
1372         }
1373         index_end = (u8*)ia + ndir->itype.index.block_size;
1374         if (unlikely(index_end > kaddr + PAGE_CACHE_SIZE)) {
1375                 ntfs_error(sb, "Index buffer (VCN 0x%llx) of directory inode "
1376                                 "0x%lx crosses page boundary. Impossible! "
1377                                 "Cannot access! This is probably a bug in the "
1378                                 "driver.", (unsigned long long)ia_pos >>
1379                                 ndir->itype.index.vcn_size_bits, vdir->i_ino);
1380                 goto err_out;
1381         }
1382         ia_start = ia_pos & ~(s64)(ndir->itype.index.block_size - 1);
1383         index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length);
1384         if (unlikely(index_end > (u8*)ia + ndir->itype.index.block_size)) {
1385                 ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of directory "
1386                                 "inode 0x%lx exceeds maximum size.",
1387                                 (unsigned long long)ia_pos >>
1388                                 ndir->itype.index.vcn_size_bits, vdir->i_ino);
1389                 goto err_out;
1390         }
1391         /* The first index entry in this index buffer. */
1392         ie = (INDEX_ENTRY*)((u8*)&ia->index +
1393                         le32_to_cpu(ia->index.entries_offset));
1394         /*
1395          * Loop until we exceed valid memory (corruption case) or until we
1396          * reach the last entry or until filldir tells us it has had enough
1397          * or signals an error (both covered by the rc test).
1398          */
1399         for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
1400                 ntfs_debug("In index allocation, offset 0x%llx.",
1401                                 (unsigned long long)ia_start +
1402                                 (unsigned long long)((u8*)ie - (u8*)ia));
1403                 /* Bounds checks. */
1404                 if (unlikely((u8*)ie < (u8*)ia || (u8*)ie +
1405                                 sizeof(INDEX_ENTRY_HEADER) > index_end ||
1406                                 (u8*)ie + le16_to_cpu(ie->key_length) >
1407                                 index_end))
1408                         goto err_out;
1409                 /* The last entry cannot contain a name. */
1410                 if (ie->flags & INDEX_ENTRY_END)
1411                         break;
1412                 /* Skip index block entry if continuing previous readdir. */
1413                 if (ia_pos - ia_start > (u8*)ie - (u8*)ia)
1414                         continue;
1415                 /* Advance the position even if going to skip the entry. */
1416                 fpos = (u8*)ie - (u8*)ia +
1417                                 (sle64_to_cpu(ia->index_block_vcn) <<
1418                                 ndir->itype.index.vcn_size_bits) +
1419                                 vol->mft_record_size;
1420                 /*
1421                  * Submit the name to the @filldir callback.  Note,
1422                  * ntfs_filldir() drops the lock on @ia_page but it retakes it
1423                  * before returning, unless a non-zero value is returned in
1424                  * which case the page is left unlocked.
1425                  */
1426                 rc = ntfs_filldir(vol, fpos, ndir, ia_page, ie, name, dirent,
1427                                 filldir);
1428                 if (rc) {
1429                         /* @ia_page is already unlocked in this case. */
1430                         ntfs_unmap_page(ia_page);
1431                         ntfs_unmap_page(bmp_page);
1432                         goto abort;
1433                 }
1434         }
1435         goto find_next_index_buffer;
1436 unm_EOD:
1437         if (ia_page) {
1438                 unlock_page(ia_page);
1439                 ntfs_unmap_page(ia_page);
1440         }
1441         ntfs_unmap_page(bmp_page);
1442 EOD:
1443         /* We are finished, set fpos to EOD. */
1444         fpos = vdir->i_size + vol->mft_record_size;
1445 abort:
1446         kfree(name);
1447 done:
1448 #ifdef DEBUG
1449         if (!rc)
1450                 ntfs_debug("EOD, fpos 0x%llx, returning 0.", fpos);
1451         else
1452                 ntfs_debug("filldir returned %i, fpos 0x%llx, returning 0.",
1453                                 rc, fpos);
1454 #endif
1455         filp->f_pos = fpos;
1456         return 0;
1457 err_out:
1458         if (bmp_page)
1459                 ntfs_unmap_page(bmp_page);
1460         if (ia_page) {
1461                 unlock_page(ia_page);
1462                 ntfs_unmap_page(ia_page);
1463         }
1464         if (ir)
1465                 kfree(ir);
1466         if (name)
1467                 kfree(name);
1468         if (ctx)
1469                 ntfs_attr_put_search_ctx(ctx);
1470         if (m)
1471                 unmap_mft_record(ndir);
1472         if (!err)
1473                 err = -EIO;
1474         ntfs_debug("Failed. Returning error code %i.", -err);
1475         filp->f_pos = fpos;
1476         return err;
1477 }
1478
1479 /**
1480  * ntfs_dir_open - called when an inode is about to be opened
1481  * @vi:         inode to be opened
1482  * @filp:       file structure describing the inode
1483  *
1484  * Limit directory size to the page cache limit on architectures where unsigned
1485  * long is 32-bits. This is the most we can do for now without overflowing the
1486  * page cache page index. Doing it this way means we don't run into problems
1487  * because of existing too large directories. It would be better to allow the
1488  * user to read the accessible part of the directory but I doubt very much
1489  * anyone is going to hit this check on a 32-bit architecture, so there is no
1490  * point in adding the extra complexity required to support this.
1491  *
1492  * On 64-bit architectures, the check is hopefully optimized away by the
1493  * compiler.
1494  */
1495 static int ntfs_dir_open(struct inode *vi, struct file *filp)
1496 {
1497         if (sizeof(unsigned long) < 8) {
1498                 if (vi->i_size > MAX_LFS_FILESIZE)
1499                         return -EFBIG;
1500         }
1501         return 0;
1502 }
1503
1504 #ifdef NTFS_RW
1505
1506 /**
1507  * ntfs_dir_fsync - sync a directory to disk
1508  * @filp:       directory to be synced
1509  * @dentry:     dentry describing the directory to sync
1510  * @datasync:   if non-zero only flush user data and not metadata
1511  *
1512  * Data integrity sync of a directory to disk.  Used for fsync, fdatasync, and
1513  * msync system calls.  This function is based on file.c::ntfs_file_fsync().
1514  *
1515  * Write the mft record and all associated extent mft records as well as the
1516  * $INDEX_ALLOCATION and $BITMAP attributes and then sync the block device.
1517  *
1518  * If @datasync is true, we do not wait on the inode(s) to be written out
1519  * but we always wait on the page cache pages to be written out.
1520  *
1521  * Note: In the past @filp could be NULL so we ignore it as we don't need it
1522  * anyway.
1523  *
1524  * Locking: Caller must hold i_sem on the inode.
1525  *
1526  * TODO: We should probably also write all attribute/index inodes associated
1527  * with this inode but since we have no simple way of getting to them we ignore
1528  * this problem for now.  We do write the $BITMAP attribute if it is present
1529  * which is the important one for a directory so things are not too bad.
1530  */
1531 static int ntfs_dir_fsync(struct file *filp, struct dentry *dentry,
1532                 int datasync)
1533 {
1534         struct inode *vi = dentry->d_inode;
1535         ntfs_inode *ni = NTFS_I(vi);
1536         int err, ret;
1537
1538         ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
1539         BUG_ON(!S_ISDIR(vi->i_mode));
1540         if (NInoIndexAllocPresent(ni) && ni->itype.index.bmp_ino)
1541                 write_inode_now(ni->itype.index.bmp_ino, !datasync);
1542         ret = ntfs_write_inode(vi, 1);
1543         write_inode_now(vi, !datasync);
1544         err = sync_blockdev(vi->i_sb->s_bdev);
1545         if (unlikely(err && !ret))
1546                 ret = err;
1547         if (likely(!ret))
1548                 ntfs_debug("Done.");
1549         else
1550                 ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx.  Error "
1551                                 "%u.", datasync ? "data" : "", vi->i_ino, -ret);
1552         return ret;
1553 }
1554
1555 #endif /* NTFS_RW */
1556
1557 struct file_operations ntfs_dir_ops = {
1558         .llseek         = generic_file_llseek,  /* Seek inside directory. */
1559         .read           = generic_read_dir,     /* Return -EISDIR. */
1560         .readdir        = ntfs_readdir,         /* Read directory contents. */
1561 #ifdef NTFS_RW
1562         .fsync          = ntfs_dir_fsync,       /* Sync a directory to disk. */
1563         /*.aio_fsync    = ,*/                   /* Sync all outstanding async
1564                                                    i/o operations on a kiocb. */
1565 #endif /* NTFS_RW */
1566         /*.ioctl        = ,*/                   /* Perform function on the
1567                                                    mounted filesystem. */
1568         .open           = ntfs_dir_open,        /* Open directory. */
1569 };