linux/fs/ntfs/mft.c
<<
>>
Prefs
   1/**
   2 * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project.
   3 *
   4 * Copyright (c) 2001-2012 Anton Altaparmakov and Tuxera Inc.
   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/buffer_head.h>
  24#include <linux/slab.h>
  25#include <linux/swap.h>
  26#include <linux/bio.h>
  27
  28#include "attrib.h"
  29#include "aops.h"
  30#include "bitmap.h"
  31#include "debug.h"
  32#include "dir.h"
  33#include "lcnalloc.h"
  34#include "malloc.h"
  35#include "mft.h"
  36#include "ntfs.h"
  37
  38/**
  39 * map_mft_record_page - map the page in which a specific mft record resides
  40 * @ni:         ntfs inode whose mft record page to map
  41 *
  42 * This maps the page in which the mft record of the ntfs inode @ni is situated
  43 * and returns a pointer to the mft record within the mapped page.
  44 *
  45 * Return value needs to be checked with IS_ERR() and if that is true PTR_ERR()
  46 * contains the negative error code returned.
  47 */
  48static inline MFT_RECORD *map_mft_record_page(ntfs_inode *ni)
  49{
  50        loff_t i_size;
  51        ntfs_volume *vol = ni->vol;
  52        struct inode *mft_vi = vol->mft_ino;
  53        struct page *page;
  54        unsigned long index, end_index;
  55        unsigned ofs;
  56
  57        BUG_ON(ni->page);
  58        /*
  59         * The index into the page cache and the offset within the page cache
  60         * page of the wanted mft record. FIXME: We need to check for
  61         * overflowing the unsigned long, but I don't think we would ever get
  62         * here if the volume was that big...
  63         */
  64        index = (u64)ni->mft_no << vol->mft_record_size_bits >>
  65                        PAGE_SHIFT;
  66        ofs = (ni->mft_no << vol->mft_record_size_bits) & ~PAGE_MASK;
  67
  68        i_size = i_size_read(mft_vi);
  69        /* The maximum valid index into the page cache for $MFT's data. */
  70        end_index = i_size >> PAGE_SHIFT;
  71
  72        /* If the wanted index is out of bounds the mft record doesn't exist. */
  73        if (unlikely(index >= end_index)) {
  74                if (index > end_index || (i_size & ~PAGE_MASK) < ofs +
  75                                vol->mft_record_size) {
  76                        page = ERR_PTR(-ENOENT);
  77                        ntfs_error(vol->sb, "Attempt to read mft record 0x%lx, "
  78                                        "which is beyond the end of the mft.  "
  79                                        "This is probably a bug in the ntfs "
  80                                        "driver.", ni->mft_no);
  81                        goto err_out;
  82                }
  83        }
  84        /* Read, map, and pin the page. */
  85        page = ntfs_map_page(mft_vi->i_mapping, index);
  86        if (likely(!IS_ERR(page))) {
  87                /* Catch multi sector transfer fixup errors. */
  88                if (likely(ntfs_is_mft_recordp((le32*)(page_address(page) +
  89                                ofs)))) {
  90                        ni->page = page;
  91                        ni->page_ofs = ofs;
  92                        return page_address(page) + ofs;
  93                }
  94                ntfs_error(vol->sb, "Mft record 0x%lx is corrupt.  "
  95                                "Run chkdsk.", ni->mft_no);
  96                ntfs_unmap_page(page);
  97                page = ERR_PTR(-EIO);
  98                NVolSetErrors(vol);
  99        }
 100err_out:
 101        ni->page = NULL;
 102        ni->page_ofs = 0;
 103        return (void*)page;
 104}
 105
 106/**
 107 * map_mft_record - map, pin and lock an mft record
 108 * @ni:         ntfs inode whose MFT record to map
 109 *
 110 * First, take the mrec_lock mutex.  We might now be sleeping, while waiting
 111 * for the mutex if it was already locked by someone else.
 112 *
 113 * The page of the record is mapped using map_mft_record_page() before being
 114 * returned to the caller.
 115 *
 116 * This in turn uses ntfs_map_page() to get the page containing the wanted mft
 117 * record (it in turn calls read_cache_page() which reads it in from disk if
 118 * necessary, increments the use count on the page so that it cannot disappear
 119 * under us and returns a reference to the page cache page).
 120 *
 121 * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it
 122 * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed
 123 * and the post-read mst fixups on each mft record in the page have been
 124 * performed, the page gets PG_uptodate set and PG_locked cleared (this is done
 125 * in our asynchronous I/O completion handler end_buffer_read_mft_async()).
 126 * ntfs_map_page() waits for PG_locked to become clear and checks if
 127 * PG_uptodate is set and returns an error code if not. This provides
 128 * sufficient protection against races when reading/using the page.
 129 *
 130 * However there is the write mapping to think about. Doing the above described
 131 * checking here will be fine, because when initiating the write we will set
 132 * PG_locked and clear PG_uptodate making sure nobody is touching the page
 133 * contents. Doing the locking this way means that the commit to disk code in
 134 * the page cache code paths is automatically sufficiently locked with us as
 135 * we will not touch a page that has been locked or is not uptodate. The only
 136 * locking problem then is them locking the page while we are accessing it.
 137 *
 138 * So that code will end up having to own the mrec_lock of all mft
 139 * records/inodes present in the page before I/O can proceed. In that case we
 140 * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be
 141 * accessing anything without owning the mrec_lock mutex.  But we do need to
 142 * use them because of the read_cache_page() invocation and the code becomes so
 143 * much simpler this way that it is well worth it.
 144 *
 145 * The mft record is now ours and we return a pointer to it. You need to check
 146 * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return
 147 * the error code.
 148 *
 149 * NOTE: Caller is responsible for setting the mft record dirty before calling
 150 * unmap_mft_record(). This is obviously only necessary if the caller really
 151 * modified the mft record...
 152 * Q: Do we want to recycle one of the VFS inode state bits instead?
 153 * A: No, the inode ones mean we want to change the mft record, not we want to
 154 * write it out.
 155 */
 156MFT_RECORD *map_mft_record(ntfs_inode *ni)
 157{
 158        MFT_RECORD *m;
 159
 160        ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
 161
 162        /* Make sure the ntfs inode doesn't go away. */
 163        atomic_inc(&ni->count);
 164
 165        /* Serialize access to this mft record. */
 166        mutex_lock(&ni->mrec_lock);
 167
 168        m = map_mft_record_page(ni);
 169        if (likely(!IS_ERR(m)))
 170                return m;
 171
 172        mutex_unlock(&ni->mrec_lock);
 173        atomic_dec(&ni->count);
 174        ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m));
 175        return m;
 176}
 177
 178/**
 179 * unmap_mft_record_page - unmap the page in which a specific mft record resides
 180 * @ni:         ntfs inode whose mft record page to unmap
 181 *
 182 * This unmaps the page in which the mft record of the ntfs inode @ni is
 183 * situated and returns. This is a NOOP if highmem is not configured.
 184 *
 185 * The unmap happens via ntfs_unmap_page() which in turn decrements the use
 186 * count on the page thus releasing it from the pinned state.
 187 *
 188 * We do not actually unmap the page from memory of course, as that will be
 189 * done by the page cache code itself when memory pressure increases or
 190 * whatever.
 191 */
 192static inline void unmap_mft_record_page(ntfs_inode *ni)
 193{
 194        BUG_ON(!ni->page);
 195
 196        // TODO: If dirty, blah...
 197        ntfs_unmap_page(ni->page);
 198        ni->page = NULL;
 199        ni->page_ofs = 0;
 200        return;
 201}
 202
 203/**
 204 * unmap_mft_record - release a mapped mft record
 205 * @ni:         ntfs inode whose MFT record to unmap
 206 *
 207 * We release the page mapping and the mrec_lock mutex which unmaps the mft
 208 * record and releases it for others to get hold of. We also release the ntfs
 209 * inode by decrementing the ntfs inode reference count.
 210 *
 211 * NOTE: If caller has modified the mft record, it is imperative to set the mft
 212 * record dirty BEFORE calling unmap_mft_record().
 213 */
 214void unmap_mft_record(ntfs_inode *ni)
 215{
 216        struct page *page = ni->page;
 217
 218        BUG_ON(!page);
 219
 220        ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
 221
 222        unmap_mft_record_page(ni);
 223        mutex_unlock(&ni->mrec_lock);
 224        atomic_dec(&ni->count);
 225        /*
 226         * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to
 227         * ntfs_clear_extent_inode() in the extent inode case, and to the
 228         * caller in the non-extent, yet pure ntfs inode case, to do the actual
 229         * tear down of all structures and freeing of all allocated memory.
 230         */
 231        return;
 232}
 233
 234/**
 235 * map_extent_mft_record - load an extent inode and attach it to its base
 236 * @base_ni:    base ntfs inode
 237 * @mref:       mft reference of the extent inode to load
 238 * @ntfs_ino:   on successful return, pointer to the ntfs_inode structure
 239 *
 240 * Load the extent mft record @mref and attach it to its base inode @base_ni.
 241 * Return the mapped extent mft record if IS_ERR(result) is false.  Otherwise
 242 * PTR_ERR(result) gives the negative error code.
 243 *
 244 * On successful return, @ntfs_ino contains a pointer to the ntfs_inode
 245 * structure of the mapped extent inode.
 246 */
 247MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref,
 248                ntfs_inode **ntfs_ino)
 249{
 250        MFT_RECORD *m;
 251        ntfs_inode *ni = NULL;
 252        ntfs_inode **extent_nis = NULL;
 253        int i;
 254        unsigned long mft_no = MREF(mref);
 255        u16 seq_no = MSEQNO(mref);
 256        bool destroy_ni = false;
 257
 258        ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).",
 259                        mft_no, base_ni->mft_no);
 260        /* Make sure the base ntfs inode doesn't go away. */
 261        atomic_inc(&base_ni->count);
 262        /*
 263         * Check if this extent inode has already been added to the base inode,
 264         * in which case just return it. If not found, add it to the base
 265         * inode before returning it.
 266         */
 267        mutex_lock(&base_ni->extent_lock);
 268        if (base_ni->nr_extents > 0) {
 269                extent_nis = base_ni->ext.extent_ntfs_inos;
 270                for (i = 0; i < base_ni->nr_extents; i++) {
 271                        if (mft_no != extent_nis[i]->mft_no)
 272                                continue;
 273                        ni = extent_nis[i];
 274                        /* Make sure the ntfs inode doesn't go away. */
 275                        atomic_inc(&ni->count);
 276                        break;
 277                }
 278        }
 279        if (likely(ni != NULL)) {
 280                mutex_unlock(&base_ni->extent_lock);
 281                atomic_dec(&base_ni->count);
 282                /* We found the record; just have to map and return it. */
 283                m = map_mft_record(ni);
 284                /* map_mft_record() has incremented this on success. */
 285                atomic_dec(&ni->count);
 286                if (likely(!IS_ERR(m))) {
 287                        /* Verify the sequence number. */
 288                        if (likely(le16_to_cpu(m->sequence_number) == seq_no)) {
 289                                ntfs_debug("Done 1.");
 290                                *ntfs_ino = ni;
 291                                return m;
 292                        }
 293                        unmap_mft_record(ni);
 294                        ntfs_error(base_ni->vol->sb, "Found stale extent mft "
 295                                        "reference! Corrupt filesystem. "
 296                                        "Run chkdsk.");
 297                        return ERR_PTR(-EIO);
 298                }
 299map_err_out:
 300                ntfs_error(base_ni->vol->sb, "Failed to map extent "
 301                                "mft record, error code %ld.", -PTR_ERR(m));
 302                return m;
 303        }
 304        /* Record wasn't there. Get a new ntfs inode and initialize it. */
 305        ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no);
 306        if (unlikely(!ni)) {
 307                mutex_unlock(&base_ni->extent_lock);
 308                atomic_dec(&base_ni->count);
 309                return ERR_PTR(-ENOMEM);
 310        }
 311        ni->vol = base_ni->vol;
 312        ni->seq_no = seq_no;
 313        ni->nr_extents = -1;
 314        ni->ext.base_ntfs_ino = base_ni;
 315        /* Now map the record. */
 316        m = map_mft_record(ni);
 317        if (IS_ERR(m)) {
 318                mutex_unlock(&base_ni->extent_lock);
 319                atomic_dec(&base_ni->count);
 320                ntfs_clear_extent_inode(ni);
 321                goto map_err_out;
 322        }
 323        /* Verify the sequence number if it is present. */
 324        if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) {
 325                ntfs_error(base_ni->vol->sb, "Found stale extent mft "
 326                                "reference! Corrupt filesystem. Run chkdsk.");
 327                destroy_ni = true;
 328                m = ERR_PTR(-EIO);
 329                goto unm_err_out;
 330        }
 331        /* Attach extent inode to base inode, reallocating memory if needed. */
 332        if (!(base_ni->nr_extents & 3)) {
 333                ntfs_inode **tmp;
 334                int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *);
 335
 336                tmp = kmalloc(new_size, GFP_NOFS);
 337                if (unlikely(!tmp)) {
 338                        ntfs_error(base_ni->vol->sb, "Failed to allocate "
 339                                        "internal buffer.");
 340                        destroy_ni = true;
 341                        m = ERR_PTR(-ENOMEM);
 342                        goto unm_err_out;
 343                }
 344                if (base_ni->nr_extents) {
 345                        BUG_ON(!base_ni->ext.extent_ntfs_inos);
 346                        memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size -
 347                                        4 * sizeof(ntfs_inode *));
 348                        kfree(base_ni->ext.extent_ntfs_inos);
 349                }
 350                base_ni->ext.extent_ntfs_inos = tmp;
 351        }
 352        base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni;
 353        mutex_unlock(&base_ni->extent_lock);
 354        atomic_dec(&base_ni->count);
 355        ntfs_debug("Done 2.");
 356        *ntfs_ino = ni;
 357        return m;
 358unm_err_out:
 359        unmap_mft_record(ni);
 360        mutex_unlock(&base_ni->extent_lock);
 361        atomic_dec(&base_ni->count);
 362        /*
 363         * If the extent inode was not attached to the base inode we need to
 364         * release it or we will leak memory.
 365         */
 366        if (destroy_ni)
 367                ntfs_clear_extent_inode(ni);
 368        return m;
 369}
 370
 371#ifdef NTFS_RW
 372
 373/**
 374 * __mark_mft_record_dirty - set the mft record and the page containing it dirty
 375 * @ni:         ntfs inode describing the mapped mft record
 376 *
 377 * Internal function.  Users should call mark_mft_record_dirty() instead.
 378 *
 379 * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni,
 380 * as well as the page containing the mft record, dirty.  Also, mark the base
 381 * vfs inode dirty.  This ensures that any changes to the mft record are
 382 * written out to disk.
 383 *
 384 * NOTE:  We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES)
 385 * on the base vfs inode, because even though file data may have been modified,
 386 * it is dirty in the inode meta data rather than the data page cache of the
 387 * inode, and thus there are no data pages that need writing out.  Therefore, a
 388 * full mark_inode_dirty() is overkill.  A mark_inode_dirty_sync(), on the
 389 * other hand, is not sufficient, because ->write_inode needs to be called even
 390 * in case of fdatasync. This needs to happen or the file data would not
 391 * necessarily hit the device synchronously, even though the vfs inode has the
 392 * O_SYNC flag set.  Also, I_DIRTY_DATASYNC simply "feels" better than just
 393 * I_DIRTY_SYNC, since the file data has not actually hit the block device yet,
 394 * which is not what I_DIRTY_SYNC on its own would suggest.
 395 */
 396void __mark_mft_record_dirty(ntfs_inode *ni)
 397{
 398        ntfs_inode *base_ni;
 399
 400        ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
 401        BUG_ON(NInoAttr(ni));
 402        mark_ntfs_record_dirty(ni->page, ni->page_ofs);
 403        /* Determine the base vfs inode and mark it dirty, too. */
 404        mutex_lock(&ni->extent_lock);
 405        if (likely(ni->nr_extents >= 0))
 406                base_ni = ni;
 407        else
 408                base_ni = ni->ext.base_ntfs_ino;
 409        mutex_unlock(&ni->extent_lock);
 410        __mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC);
 411}
 412
 413static const char *ntfs_please_email = "Please email "
 414                "linux-ntfs-dev@lists.sourceforge.net and say that you saw "
 415                "this message.  Thank you.";
 416
 417/**
 418 * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror
 419 * @vol:        ntfs volume on which the mft record to synchronize resides
 420 * @mft_no:     mft record number of mft record to synchronize
 421 * @m:          mapped, mst protected (extent) mft record to synchronize
 422 *
 423 * Write the mapped, mst protected (extent) mft record @m with mft record
 424 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol,
 425 * bypassing the page cache and the $MFTMirr inode itself.
 426 *
 427 * This function is only for use at umount time when the mft mirror inode has
 428 * already been disposed off.  We BUG() if we are called while the mft mirror
 429 * inode is still attached to the volume.
 430 *
 431 * On success return 0.  On error return -errno.
 432 *
 433 * NOTE:  This function is not implemented yet as I am not convinced it can
 434 * actually be triggered considering the sequence of commits we do in super.c::
 435 * ntfs_put_super().  But just in case we provide this place holder as the
 436 * alternative would be either to BUG() or to get a NULL pointer dereference
 437 * and Oops.
 438 */
 439static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol,
 440                const unsigned long mft_no, MFT_RECORD *m)
 441{
 442        BUG_ON(vol->mftmirr_ino);
 443        ntfs_error(vol->sb, "Umount time mft mirror syncing is not "
 444                        "implemented yet.  %s", ntfs_please_email);
 445        return -EOPNOTSUPP;
 446}
 447
 448/**
 449 * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror
 450 * @vol:        ntfs volume on which the mft record to synchronize resides
 451 * @mft_no:     mft record number of mft record to synchronize
 452 * @m:          mapped, mst protected (extent) mft record to synchronize
 453 * @sync:       if true, wait for i/o completion
 454 *
 455 * Write the mapped, mst protected (extent) mft record @m with mft record
 456 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol.
 457 *
 458 * On success return 0.  On error return -errno and set the volume errors flag
 459 * in the ntfs volume @vol.
 460 *
 461 * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
 462 *
 463 * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
 464 * schedule i/o via ->writepage or do it via kntfsd or whatever.
 465 */
 466int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no,
 467                MFT_RECORD *m, int sync)
 468{
 469        struct page *page;
 470        unsigned int blocksize = vol->sb->s_blocksize;
 471        int max_bhs = vol->mft_record_size / blocksize;
 472        struct buffer_head *bhs[max_bhs];
 473        struct buffer_head *bh, *head;
 474        u8 *kmirr;
 475        runlist_element *rl;
 476        unsigned int block_start, block_end, m_start, m_end, page_ofs;
 477        int i_bhs, nr_bhs, err = 0;
 478        unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
 479
 480        ntfs_debug("Entering for inode 0x%lx.", mft_no);
 481        BUG_ON(!max_bhs);
 482        if (unlikely(!vol->mftmirr_ino)) {
 483                /* This could happen during umount... */
 484                err = ntfs_sync_mft_mirror_umount(vol, mft_no, m);
 485                if (likely(!err))
 486                        return err;
 487                goto err_out;
 488        }
 489        /* Get the page containing the mirror copy of the mft record @m. */
 490        page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >>
 491                        (PAGE_SHIFT - vol->mft_record_size_bits));
 492        if (IS_ERR(page)) {
 493                ntfs_error(vol->sb, "Failed to map mft mirror page.");
 494                err = PTR_ERR(page);
 495                goto err_out;
 496        }
 497        lock_page(page);
 498        BUG_ON(!PageUptodate(page));
 499        ClearPageUptodate(page);
 500        /* Offset of the mft mirror record inside the page. */
 501        page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_MASK;
 502        /* The address in the page of the mirror copy of the mft record @m. */
 503        kmirr = page_address(page) + page_ofs;
 504        /* Copy the mst protected mft record to the mirror. */
 505        memcpy(kmirr, m, vol->mft_record_size);
 506        /* Create uptodate buffers if not present. */
 507        if (unlikely(!page_has_buffers(page))) {
 508                struct buffer_head *tail;
 509
 510                bh = head = alloc_page_buffers(page, blocksize, 1);
 511                do {
 512                        set_buffer_uptodate(bh);
 513                        tail = bh;
 514                        bh = bh->b_this_page;
 515                } while (bh);
 516                tail->b_this_page = head;
 517                attach_page_buffers(page, head);
 518        }
 519        bh = head = page_buffers(page);
 520        BUG_ON(!bh);
 521        rl = NULL;
 522        nr_bhs = 0;
 523        block_start = 0;
 524        m_start = kmirr - (u8*)page_address(page);
 525        m_end = m_start + vol->mft_record_size;
 526        do {
 527                block_end = block_start + blocksize;
 528                /* If the buffer is outside the mft record, skip it. */
 529                if (block_end <= m_start)
 530                        continue;
 531                if (unlikely(block_start >= m_end))
 532                        break;
 533                /* Need to map the buffer if it is not mapped already. */
 534                if (unlikely(!buffer_mapped(bh))) {
 535                        VCN vcn;
 536                        LCN lcn;
 537                        unsigned int vcn_ofs;
 538
 539                        bh->b_bdev = vol->sb->s_bdev;
 540                        /* Obtain the vcn and offset of the current block. */
 541                        vcn = ((VCN)mft_no << vol->mft_record_size_bits) +
 542                                        (block_start - m_start);
 543                        vcn_ofs = vcn & vol->cluster_size_mask;
 544                        vcn >>= vol->cluster_size_bits;
 545                        if (!rl) {
 546                                down_read(&NTFS_I(vol->mftmirr_ino)->
 547                                                runlist.lock);
 548                                rl = NTFS_I(vol->mftmirr_ino)->runlist.rl;
 549                                /*
 550                                 * $MFTMirr always has the whole of its runlist
 551                                 * in memory.
 552                                 */
 553                                BUG_ON(!rl);
 554                        }
 555                        /* Seek to element containing target vcn. */
 556                        while (rl->length && rl[1].vcn <= vcn)
 557                                rl++;
 558                        lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
 559                        /* For $MFTMirr, only lcn >= 0 is a successful remap. */
 560                        if (likely(lcn >= 0)) {
 561                                /* Setup buffer head to correct block. */
 562                                bh->b_blocknr = ((lcn <<
 563                                                vol->cluster_size_bits) +
 564                                                vcn_ofs) >> blocksize_bits;
 565                                set_buffer_mapped(bh);
 566                        } else {
 567                                bh->b_blocknr = -1;
 568                                ntfs_error(vol->sb, "Cannot write mft mirror "
 569                                                "record 0x%lx because its "
 570                                                "location on disk could not "
 571                                                "be determined (error code "
 572                                                "%lli).", mft_no,
 573                                                (long long)lcn);
 574                                err = -EIO;
 575                        }
 576                }
 577                BUG_ON(!buffer_uptodate(bh));
 578                BUG_ON(!nr_bhs && (m_start != block_start));
 579                BUG_ON(nr_bhs >= max_bhs);
 580                bhs[nr_bhs++] = bh;
 581                BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
 582        } while (block_start = block_end, (bh = bh->b_this_page) != head);
 583        if (unlikely(rl))
 584                up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock);
 585        if (likely(!err)) {
 586                /* Lock buffers and start synchronous write i/o on them. */
 587                for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
 588                        struct buffer_head *tbh = bhs[i_bhs];
 589
 590                        if (!trylock_buffer(tbh))
 591                                BUG();
 592                        BUG_ON(!buffer_uptodate(tbh));
 593                        clear_buffer_dirty(tbh);
 594                        get_bh(tbh);
 595                        tbh->b_end_io = end_buffer_write_sync;
 596                        submit_bh(REQ_OP_WRITE, 0, tbh);
 597                }
 598                /* Wait on i/o completion of buffers. */
 599                for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
 600                        struct buffer_head *tbh = bhs[i_bhs];
 601
 602                        wait_on_buffer(tbh);
 603                        if (unlikely(!buffer_uptodate(tbh))) {
 604                                err = -EIO;
 605                                /*
 606                                 * Set the buffer uptodate so the page and
 607                                 * buffer states do not become out of sync.
 608                                 */
 609                                set_buffer_uptodate(tbh);
 610                        }
 611                }
 612        } else /* if (unlikely(err)) */ {
 613                /* Clean the buffers. */
 614                for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
 615                        clear_buffer_dirty(bhs[i_bhs]);
 616        }
 617        /* Current state: all buffers are clean, unlocked, and uptodate. */
 618        /* Remove the mst protection fixups again. */
 619        post_write_mst_fixup((NTFS_RECORD*)kmirr);
 620        flush_dcache_page(page);
 621        SetPageUptodate(page);
 622        unlock_page(page);
 623        ntfs_unmap_page(page);
 624        if (likely(!err)) {
 625                ntfs_debug("Done.");
 626        } else {
 627                ntfs_error(vol->sb, "I/O error while writing mft mirror "
 628                                "record 0x%lx!", mft_no);
 629err_out:
 630                ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error "
 631                                "code %i).  Volume will be left marked dirty "
 632                                "on umount.  Run ntfsfix on the partition "
 633                                "after umounting to correct this.", -err);
 634                NVolSetErrors(vol);
 635        }
 636        return err;
 637}
 638
 639/**
 640 * write_mft_record_nolock - write out a mapped (extent) mft record
 641 * @ni:         ntfs inode describing the mapped (extent) mft record
 642 * @m:          mapped (extent) mft record to write
 643 * @sync:       if true, wait for i/o completion
 644 *
 645 * Write the mapped (extent) mft record @m described by the (regular or extent)
 646 * ntfs inode @ni to backing store.  If the mft record @m has a counterpart in
 647 * the mft mirror, that is also updated.
 648 *
 649 * We only write the mft record if the ntfs inode @ni is dirty and the first
 650 * buffer belonging to its mft record is dirty, too.  We ignore the dirty state
 651 * of subsequent buffers because we could have raced with
 652 * fs/ntfs/aops.c::mark_ntfs_record_dirty().
 653 *
 654 * On success, clean the mft record and return 0.  On error, leave the mft
 655 * record dirty and return -errno.
 656 *
 657 * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
 658 * However, if the mft record has a counterpart in the mft mirror and @sync is
 659 * true, we write the mft record, wait for i/o completion, and only then write
 660 * the mft mirror copy.  This ensures that if the system crashes either the mft
 661 * or the mft mirror will contain a self-consistent mft record @m.  If @sync is
 662 * false on the other hand, we start i/o on both and then wait for completion
 663 * on them.  This provides a speedup but no longer guarantees that you will end
 664 * up with a self-consistent mft record in the case of a crash but if you asked
 665 * for asynchronous writing you probably do not care about that anyway.
 666 *
 667 * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
 668 * schedule i/o via ->writepage or do it via kntfsd or whatever.
 669 */
 670int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync)
 671{
 672        ntfs_volume *vol = ni->vol;
 673        struct page *page = ni->page;
 674        unsigned int blocksize = vol->sb->s_blocksize;
 675        unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
 676        int max_bhs = vol->mft_record_size / blocksize;
 677        struct buffer_head *bhs[max_bhs];
 678        struct buffer_head *bh, *head;
 679        runlist_element *rl;
 680        unsigned int block_start, block_end, m_start, m_end;
 681        int i_bhs, nr_bhs, err = 0;
 682
 683        ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
 684        BUG_ON(NInoAttr(ni));
 685        BUG_ON(!max_bhs);
 686        BUG_ON(!PageLocked(page));
 687        /*
 688         * If the ntfs_inode is clean no need to do anything.  If it is dirty,
 689         * mark it as clean now so that it can be redirtied later on if needed.
 690         * There is no danger of races since the caller is holding the locks
 691         * for the mft record @m and the page it is in.
 692         */
 693        if (!NInoTestClearDirty(ni))
 694                goto done;
 695        bh = head = page_buffers(page);
 696        BUG_ON(!bh);
 697        rl = NULL;
 698        nr_bhs = 0;
 699        block_start = 0;
 700        m_start = ni->page_ofs;
 701        m_end = m_start + vol->mft_record_size;
 702        do {
 703                block_end = block_start + blocksize;
 704                /* If the buffer is outside the mft record, skip it. */
 705                if (block_end <= m_start)
 706                        continue;
 707                if (unlikely(block_start >= m_end))
 708                        break;
 709                /*
 710                 * If this block is not the first one in the record, we ignore
 711                 * the buffer's dirty state because we could have raced with a
 712                 * parallel mark_ntfs_record_dirty().
 713                 */
 714                if (block_start == m_start) {
 715                        /* This block is the first one in the record. */
 716                        if (!buffer_dirty(bh)) {
 717                                BUG_ON(nr_bhs);
 718                                /* Clean records are not written out. */
 719                                break;
 720                        }
 721                }
 722                /* Need to map the buffer if it is not mapped already. */
 723                if (unlikely(!buffer_mapped(bh))) {
 724                        VCN vcn;
 725                        LCN lcn;
 726                        unsigned int vcn_ofs;
 727
 728                        bh->b_bdev = vol->sb->s_bdev;
 729                        /* Obtain the vcn and offset of the current block. */
 730                        vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) +
 731                                        (block_start - m_start);
 732                        vcn_ofs = vcn & vol->cluster_size_mask;
 733                        vcn >>= vol->cluster_size_bits;
 734                        if (!rl) {
 735                                down_read(&NTFS_I(vol->mft_ino)->runlist.lock);
 736                                rl = NTFS_I(vol->mft_ino)->runlist.rl;
 737                                BUG_ON(!rl);
 738                        }
 739                        /* Seek to element containing target vcn. */
 740                        while (rl->length && rl[1].vcn <= vcn)
 741                                rl++;
 742                        lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
 743                        /* For $MFT, only lcn >= 0 is a successful remap. */
 744                        if (likely(lcn >= 0)) {
 745                                /* Setup buffer head to correct block. */
 746                                bh->b_blocknr = ((lcn <<
 747                                                vol->cluster_size_bits) +
 748                                                vcn_ofs) >> blocksize_bits;
 749                                set_buffer_mapped(bh);
 750                        } else {
 751                                bh->b_blocknr = -1;
 752                                ntfs_error(vol->sb, "Cannot write mft record "
 753                                                "0x%lx because its location "
 754                                                "on disk could not be "
 755                                                "determined (error code %lli).",
 756                                                ni->mft_no, (long long)lcn);
 757                                err = -EIO;
 758                        }
 759                }
 760                BUG_ON(!buffer_uptodate(bh));
 761                BUG_ON(!nr_bhs && (m_start != block_start));
 762                BUG_ON(nr_bhs >= max_bhs);
 763                bhs[nr_bhs++] = bh;
 764                BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
 765        } while (block_start = block_end, (bh = bh->b_this_page) != head);
 766        if (unlikely(rl))
 767                up_read(&NTFS_I(vol->mft_ino)->runlist.lock);
 768        if (!nr_bhs)
 769                goto done;
 770        if (unlikely(err))
 771                goto cleanup_out;
 772        /* Apply the mst protection fixups. */
 773        err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size);
 774        if (err) {
 775                ntfs_error(vol->sb, "Failed to apply mst fixups!");
 776                goto cleanup_out;
 777        }
 778        flush_dcache_mft_record_page(ni);
 779        /* Lock buffers and start synchronous write i/o on them. */
 780        for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
 781                struct buffer_head *tbh = bhs[i_bhs];
 782
 783                if (!trylock_buffer(tbh))
 784                        BUG();
 785                BUG_ON(!buffer_uptodate(tbh));
 786                clear_buffer_dirty(tbh);
 787                get_bh(tbh);
 788                tbh->b_end_io = end_buffer_write_sync;
 789                submit_bh(REQ_OP_WRITE, 0, tbh);
 790        }
 791        /* Synchronize the mft mirror now if not @sync. */
 792        if (!sync && ni->mft_no < vol->mftmirr_size)
 793                ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
 794        /* Wait on i/o completion of buffers. */
 795        for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
 796                struct buffer_head *tbh = bhs[i_bhs];
 797
 798                wait_on_buffer(tbh);
 799                if (unlikely(!buffer_uptodate(tbh))) {
 800                        err = -EIO;
 801                        /*
 802                         * Set the buffer uptodate so the page and buffer
 803                         * states do not become out of sync.
 804                         */
 805                        if (PageUptodate(page))
 806                                set_buffer_uptodate(tbh);
 807                }
 808        }
 809        /* If @sync, now synchronize the mft mirror. */
 810        if (sync && ni->mft_no < vol->mftmirr_size)
 811                ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
 812        /* Remove the mst protection fixups again. */
 813        post_write_mst_fixup((NTFS_RECORD*)m);
 814        flush_dcache_mft_record_page(ni);
 815        if (unlikely(err)) {
 816                /* I/O error during writing.  This is really bad! */
 817                ntfs_error(vol->sb, "I/O error while writing mft record "
 818                                "0x%lx!  Marking base inode as bad.  You "
 819                                "should unmount the volume and run chkdsk.",
 820                                ni->mft_no);
 821                goto err_out;
 822        }
 823done:
 824        ntfs_debug("Done.");
 825        return 0;
 826cleanup_out:
 827        /* Clean the buffers. */
 828        for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
 829                clear_buffer_dirty(bhs[i_bhs]);
 830err_out:
 831        /*
 832         * Current state: all buffers are clean, unlocked, and uptodate.
 833         * The caller should mark the base inode as bad so that no more i/o
 834         * happens.  ->clear_inode() will still be invoked so all extent inodes
 835         * and other allocated memory will be freed.
 836         */
 837        if (err == -ENOMEM) {
 838                ntfs_error(vol->sb, "Not enough memory to write mft record.  "
 839                                "Redirtying so the write is retried later.");
 840                mark_mft_record_dirty(ni);
 841                err = 0;
 842        } else
 843                NVolSetErrors(vol);
 844        return err;
 845}
 846
 847/**
 848 * ntfs_may_write_mft_record - check if an mft record may be written out
 849 * @vol:        [IN]  ntfs volume on which the mft record to check resides
 850 * @mft_no:     [IN]  mft record number of the mft record to check
 851 * @m:          [IN]  mapped mft record to check
 852 * @locked_ni:  [OUT] caller has to unlock this ntfs inode if one is returned
 853 *
 854 * Check if the mapped (base or extent) mft record @m with mft record number
 855 * @mft_no belonging to the ntfs volume @vol may be written out.  If necessary
 856 * and possible the ntfs inode of the mft record is locked and the base vfs
 857 * inode is pinned.  The locked ntfs inode is then returned in @locked_ni.  The
 858 * caller is responsible for unlocking the ntfs inode and unpinning the base
 859 * vfs inode.
 860 *
 861 * Return 'true' if the mft record may be written out and 'false' if not.
 862 *
 863 * The caller has locked the page and cleared the uptodate flag on it which
 864 * means that we can safely write out any dirty mft records that do not have
 865 * their inodes in icache as determined by ilookup5() as anyone
 866 * opening/creating such an inode would block when attempting to map the mft
 867 * record in read_cache_page() until we are finished with the write out.
 868 *
 869 * Here is a description of the tests we perform:
 870 *
 871 * If the inode is found in icache we know the mft record must be a base mft
 872 * record.  If it is dirty, we do not write it and return 'false' as the vfs
 873 * inode write paths will result in the access times being updated which would
 874 * cause the base mft record to be redirtied and written out again.  (We know
 875 * the access time update will modify the base mft record because Windows
 876 * chkdsk complains if the standard information attribute is not in the base
 877 * mft record.)
 878 *
 879 * If the inode is in icache and not dirty, we attempt to lock the mft record
 880 * and if we find the lock was already taken, it is not safe to write the mft
 881 * record and we return 'false'.
 882 *
 883 * If we manage to obtain the lock we have exclusive access to the mft record,
 884 * which also allows us safe writeout of the mft record.  We then set
 885 * @locked_ni to the locked ntfs inode and return 'true'.
 886 *
 887 * Note we cannot just lock the mft record and sleep while waiting for the lock
 888 * because this would deadlock due to lock reversal (normally the mft record is
 889 * locked before the page is locked but we already have the page locked here
 890 * when we try to lock the mft record).
 891 *
 892 * If the inode is not in icache we need to perform further checks.
 893 *
 894 * If the mft record is not a FILE record or it is a base mft record, we can
 895 * safely write it and return 'true'.
 896 *
 897 * We now know the mft record is an extent mft record.  We check if the inode
 898 * corresponding to its base mft record is in icache and obtain a reference to
 899 * it if it is.  If it is not, we can safely write it and return 'true'.
 900 *
 901 * We now have the base inode for the extent mft record.  We check if it has an
 902 * ntfs inode for the extent mft record attached and if not it is safe to write
 903 * the extent mft record and we return 'true'.
 904 *
 905 * The ntfs inode for the extent mft record is attached to the base inode so we
 906 * attempt to lock the extent mft record and if we find the lock was already
 907 * taken, it is not safe to write the extent mft record and we return 'false'.
 908 *
 909 * If we manage to obtain the lock we have exclusive access to the extent mft
 910 * record, which also allows us safe writeout of the extent mft record.  We
 911 * set the ntfs inode of the extent mft record clean and then set @locked_ni to
 912 * the now locked ntfs inode and return 'true'.
 913 *
 914 * Note, the reason for actually writing dirty mft records here and not just
 915 * relying on the vfs inode dirty code paths is that we can have mft records
 916 * modified without them ever having actual inodes in memory.  Also we can have
 917 * dirty mft records with clean ntfs inodes in memory.  None of the described
 918 * cases would result in the dirty mft records being written out if we only
 919 * relied on the vfs inode dirty code paths.  And these cases can really occur
 920 * during allocation of new mft records and in particular when the
 921 * initialized_size of the $MFT/$DATA attribute is extended and the new space
 922 * is initialized using ntfs_mft_record_format().  The clean inode can then
 923 * appear if the mft record is reused for a new inode before it got written
 924 * out.
 925 */
 926bool ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
 927                const MFT_RECORD *m, ntfs_inode **locked_ni)
 928{
 929        struct super_block *sb = vol->sb;
 930        struct inode *mft_vi = vol->mft_ino;
 931        struct inode *vi;
 932        ntfs_inode *ni, *eni, **extent_nis;
 933        int i;
 934        ntfs_attr na;
 935
 936        ntfs_debug("Entering for inode 0x%lx.", mft_no);
 937        /*
 938         * Normally we do not return a locked inode so set @locked_ni to NULL.
 939         */
 940        BUG_ON(!locked_ni);
 941        *locked_ni = NULL;
 942        /*
 943         * Check if the inode corresponding to this mft record is in the VFS
 944         * inode cache and obtain a reference to it if it is.
 945         */
 946        ntfs_debug("Looking for inode 0x%lx in icache.", mft_no);
 947        na.mft_no = mft_no;
 948        na.name = NULL;
 949        na.name_len = 0;
 950        na.type = AT_UNUSED;
 951        /*
 952         * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
 953         * we get here for it rather often.
 954         */
 955        if (!mft_no) {
 956                /* Balance the below iput(). */
 957                vi = igrab(mft_vi);
 958                BUG_ON(vi != mft_vi);
 959        } else {
 960                /*
 961                 * Have to use ilookup5_nowait() since ilookup5() waits for the
 962                 * inode lock which causes ntfs to deadlock when a concurrent
 963                 * inode write via the inode dirty code paths and the page
 964                 * dirty code path of the inode dirty code path when writing
 965                 * $MFT occurs.
 966                 */
 967                vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
 968        }
 969        if (vi) {
 970                ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
 971                /* The inode is in icache. */
 972                ni = NTFS_I(vi);
 973                /* Take a reference to the ntfs inode. */
 974                atomic_inc(&ni->count);
 975                /* If the inode is dirty, do not write this record. */
 976                if (NInoDirty(ni)) {
 977                        ntfs_debug("Inode 0x%lx is dirty, do not write it.",
 978                                        mft_no);
 979                        atomic_dec(&ni->count);
 980                        iput(vi);
 981                        return false;
 982                }
 983                ntfs_debug("Inode 0x%lx is not dirty.", mft_no);
 984                /* The inode is not dirty, try to take the mft record lock. */
 985                if (unlikely(!mutex_trylock(&ni->mrec_lock))) {
 986                        ntfs_debug("Mft record 0x%lx is already locked, do "
 987                                        "not write it.", mft_no);
 988                        atomic_dec(&ni->count);
 989                        iput(vi);
 990                        return false;
 991                }
 992                ntfs_debug("Managed to lock mft record 0x%lx, write it.",
 993                                mft_no);
 994                /*
 995                 * The write has to occur while we hold the mft record lock so
 996                 * return the locked ntfs inode.
 997                 */
 998                *locked_ni = ni;
 999                return true;
1000        }
1001        ntfs_debug("Inode 0x%lx is not in icache.", mft_no);
1002        /* The inode is not in icache. */
1003        /* Write the record if it is not a mft record (type "FILE"). */
1004        if (!ntfs_is_mft_record(m->magic)) {
1005                ntfs_debug("Mft record 0x%lx is not a FILE record, write it.",
1006                                mft_no);
1007                return true;
1008        }
1009        /* Write the mft record if it is a base inode. */
1010        if (!m->base_mft_record) {
1011                ntfs_debug("Mft record 0x%lx is a base record, write it.",
1012                                mft_no);
1013                return true;
1014        }
1015        /*
1016         * This is an extent mft record.  Check if the inode corresponding to
1017         * its base mft record is in icache and obtain a reference to it if it
1018         * is.
1019         */
1020        na.mft_no = MREF_LE(m->base_mft_record);
1021        ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
1022                        "inode 0x%lx in icache.", mft_no, na.mft_no);
1023        if (!na.mft_no) {
1024                /* Balance the below iput(). */
1025                vi = igrab(mft_vi);
1026                BUG_ON(vi != mft_vi);
1027        } else
1028                vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
1029                                &na);
1030        if (!vi) {
1031                /*
1032                 * The base inode is not in icache, write this extent mft
1033                 * record.
1034                 */
1035                ntfs_debug("Base inode 0x%lx is not in icache, write the "
1036                                "extent record.", na.mft_no);
1037                return true;
1038        }
1039        ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no);
1040        /*
1041         * The base inode is in icache.  Check if it has the extent inode
1042         * corresponding to this extent mft record attached.
1043         */
1044        ni = NTFS_I(vi);
1045        mutex_lock(&ni->extent_lock);
1046        if (ni->nr_extents <= 0) {
1047                /*
1048                 * The base inode has no attached extent inodes, write this
1049                 * extent mft record.
1050                 */
1051                mutex_unlock(&ni->extent_lock);
1052                iput(vi);
1053                ntfs_debug("Base inode 0x%lx has no attached extent inodes, "
1054                                "write the extent record.", na.mft_no);
1055                return true;
1056        }
1057        /* Iterate over the attached extent inodes. */
1058        extent_nis = ni->ext.extent_ntfs_inos;
1059        for (eni = NULL, i = 0; i < ni->nr_extents; ++i) {
1060                if (mft_no == extent_nis[i]->mft_no) {
1061                        /*
1062                         * Found the extent inode corresponding to this extent
1063                         * mft record.
1064                         */
1065                        eni = extent_nis[i];
1066                        break;
1067                }
1068        }
1069        /*
1070         * If the extent inode was not attached to the base inode, write this
1071         * extent mft record.
1072         */
1073        if (!eni) {
1074                mutex_unlock(&ni->extent_lock);
1075                iput(vi);
1076                ntfs_debug("Extent inode 0x%lx is not attached to its base "
1077                                "inode 0x%lx, write the extent record.",
1078                                mft_no, na.mft_no);
1079                return true;
1080        }
1081        ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.",
1082                        mft_no, na.mft_no);
1083        /* Take a reference to the extent ntfs inode. */
1084        atomic_inc(&eni->count);
1085        mutex_unlock(&ni->extent_lock);
1086        /*
1087         * Found the extent inode coresponding to this extent mft record.
1088         * Try to take the mft record lock.
1089         */
1090        if (unlikely(!mutex_trylock(&eni->mrec_lock))) {
1091                atomic_dec(&eni->count);
1092                iput(vi);
1093                ntfs_debug("Extent mft record 0x%lx is already locked, do "
1094                                "not write it.", mft_no);
1095                return false;
1096        }
1097        ntfs_debug("Managed to lock extent mft record 0x%lx, write it.",
1098                        mft_no);
1099        if (NInoTestClearDirty(eni))
1100                ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.",
1101                                mft_no);
1102        /*
1103         * The write has to occur while we hold the mft record lock so return
1104         * the locked extent ntfs inode.
1105         */
1106        *locked_ni = eni;
1107        return true;
1108}
1109
1110static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
1111                "chkdsk.";
1112
1113/**
1114 * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name
1115 * @vol:        volume on which to search for a free mft record
1116 * @base_ni:    open base inode if allocating an extent mft record or NULL
1117 *
1118 * Search for a free mft record in the mft bitmap attribute on the ntfs volume
1119 * @vol.
1120 *
1121 * If @base_ni is NULL start the search at the default allocator position.
1122 *
1123 * If @base_ni is not NULL start the search at the mft record after the base
1124 * mft record @base_ni.
1125 *
1126 * Return the free mft record on success and -errno on error.  An error code of
1127 * -ENOSPC means that there are no free mft records in the currently
1128 * initialized mft bitmap.
1129 *
1130 * Locking: Caller must hold vol->mftbmp_lock for writing.
1131 */
1132static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol,
1133                ntfs_inode *base_ni)
1134{
1135        s64 pass_end, ll, data_pos, pass_start, ofs, bit;
1136        unsigned long flags;
1137        struct address_space *mftbmp_mapping;
1138        u8 *buf, *byte;
1139        struct page *page;
1140        unsigned int page_ofs, size;
1141        u8 pass, b;
1142
1143        ntfs_debug("Searching for free mft record in the currently "
1144                        "initialized mft bitmap.");
1145        mftbmp_mapping = vol->mftbmp_ino->i_mapping;
1146        /*
1147         * Set the end of the pass making sure we do not overflow the mft
1148         * bitmap.
1149         */
1150        read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags);
1151        pass_end = NTFS_I(vol->mft_ino)->allocated_size >>
1152                        vol->mft_record_size_bits;
1153        read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags);
1154        read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1155        ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3;
1156        read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1157        if (pass_end > ll)
1158                pass_end = ll;
1159        pass = 1;
1160        if (!base_ni)
1161                data_pos = vol->mft_data_pos;
1162        else
1163                data_pos = base_ni->mft_no + 1;
1164        if (data_pos < 24)
1165                data_pos = 24;
1166        if (data_pos >= pass_end) {
1167                data_pos = 24;
1168                pass = 2;
1169                /* This happens on a freshly formatted volume. */
1170                if (data_pos >= pass_end)
1171                        return -ENOSPC;
1172        }
1173        pass_start = data_pos;
1174        ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, "
1175                        "pass_end 0x%llx, data_pos 0x%llx.", pass,
1176                        (long long)pass_start, (long long)pass_end,
1177                        (long long)data_pos);
1178        /* Loop until a free mft record is found. */
1179        for (; pass <= 2;) {
1180                /* Cap size to pass_end. */
1181                ofs = data_pos >> 3;
1182                page_ofs = ofs & ~PAGE_MASK;
1183                size = PAGE_SIZE - page_ofs;
1184                ll = ((pass_end + 7) >> 3) - ofs;
1185                if (size > ll)
1186                        size = ll;
1187                size <<= 3;
1188                /*
1189                 * If we are still within the active pass, search the next page
1190                 * for a zero bit.
1191                 */
1192                if (size) {
1193                        page = ntfs_map_page(mftbmp_mapping,
1194                                        ofs >> PAGE_SHIFT);
1195                        if (IS_ERR(page)) {
1196                                ntfs_error(vol->sb, "Failed to read mft "
1197                                                "bitmap, aborting.");
1198                                return PTR_ERR(page);
1199                        }
1200                        buf = (u8*)page_address(page) + page_ofs;
1201                        bit = data_pos & 7;
1202                        data_pos &= ~7ull;
1203                        ntfs_debug("Before inner for loop: size 0x%x, "
1204                                        "data_pos 0x%llx, bit 0x%llx", size,
1205                                        (long long)data_pos, (long long)bit);
1206                        for (; bit < size && data_pos + bit < pass_end;
1207                                        bit &= ~7ull, bit += 8) {
1208                                byte = buf + (bit >> 3);
1209                                if (*byte == 0xff)
1210                                        continue;
1211                                b = ffz((unsigned long)*byte);
1212                                if (b < 8 && b >= (bit & 7)) {
1213                                        ll = data_pos + (bit & ~7ull) + b;
1214                                        if (unlikely(ll > (1ll << 32))) {
1215                                                ntfs_unmap_page(page);
1216                                                return -ENOSPC;
1217                                        }
1218                                        *byte |= 1 << b;
1219                                        flush_dcache_page(page);
1220                                        set_page_dirty(page);
1221                                        ntfs_unmap_page(page);
1222                                        ntfs_debug("Done.  (Found and "
1223                                                        "allocated mft record "
1224                                                        "0x%llx.)",
1225                                                        (long long)ll);
1226                                        return ll;
1227                                }
1228                        }
1229                        ntfs_debug("After inner for loop: size 0x%x, "
1230                                        "data_pos 0x%llx, bit 0x%llx", size,
1231                                        (long long)data_pos, (long long)bit);
1232                        data_pos += size;
1233                        ntfs_unmap_page(page);
1234                        /*
1235                         * If the end of the pass has not been reached yet,
1236                         * continue searching the mft bitmap for a zero bit.
1237                         */
1238                        if (data_pos < pass_end)
1239                                continue;
1240                }
1241                /* Do the next pass. */
1242                if (++pass == 2) {
1243                        /*
1244                         * Starting the second pass, in which we scan the first
1245                         * part of the zone which we omitted earlier.
1246                         */
1247                        pass_end = pass_start;
1248                        data_pos = pass_start = 24;
1249                        ntfs_debug("pass %i, pass_start 0x%llx, pass_end "
1250                                        "0x%llx.", pass, (long long)pass_start,
1251                                        (long long)pass_end);
1252                        if (data_pos >= pass_end)
1253                                break;
1254                }
1255        }
1256        /* No free mft records in currently initialized mft bitmap. */
1257        ntfs_debug("Done.  (No free mft records left in currently initialized "
1258                        "mft bitmap.)");
1259        return -ENOSPC;
1260}
1261
1262/**
1263 * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster
1264 * @vol:        volume on which to extend the mft bitmap attribute
1265 *
1266 * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster.
1267 *
1268 * Note: Only changes allocated_size, i.e. does not touch initialized_size or
1269 * data_size.
1270 *
1271 * Return 0 on success and -errno on error.
1272 *
1273 * Locking: - Caller must hold vol->mftbmp_lock for writing.
1274 *          - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for
1275 *            writing and releases it before returning.
1276 *          - This function takes vol->lcnbmp_lock for writing and releases it
1277 *            before returning.
1278 */
1279static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol)
1280{
1281        LCN lcn;
1282        s64 ll;
1283        unsigned long flags;
1284        struct page *page;
1285        ntfs_inode *mft_ni, *mftbmp_ni;
1286        runlist_element *rl, *rl2 = NULL;
1287        ntfs_attr_search_ctx *ctx = NULL;
1288        MFT_RECORD *mrec;
1289        ATTR_RECORD *a = NULL;
1290        int ret, mp_size;
1291        u32 old_alen = 0;
1292        u8 *b, tb;
1293        struct {
1294                u8 added_cluster:1;
1295                u8 added_run:1;
1296                u8 mp_rebuilt:1;
1297        } status = { 0, 0, 0 };
1298
1299        ntfs_debug("Extending mft bitmap allocation.");
1300        mft_ni = NTFS_I(vol->mft_ino);
1301        mftbmp_ni = NTFS_I(vol->mftbmp_ino);
1302        /*
1303         * Determine the last lcn of the mft bitmap.  The allocated size of the
1304         * mft bitmap cannot be zero so we are ok to do this.
1305         */
1306        down_write(&mftbmp_ni->runlist.lock);
1307        read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1308        ll = mftbmp_ni->allocated_size;
1309        read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1310        rl = ntfs_attr_find_vcn_nolock(mftbmp_ni,
1311                        (ll - 1) >> vol->cluster_size_bits, NULL);
1312        if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1313                up_write(&mftbmp_ni->runlist.lock);
1314                ntfs_error(vol->sb, "Failed to determine last allocated "
1315                                "cluster of mft bitmap attribute.");
1316                if (!IS_ERR(rl))
1317                        ret = -EIO;
1318                else
1319                        ret = PTR_ERR(rl);
1320                return ret;
1321        }
1322        lcn = rl->lcn + rl->length;
1323        ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.",
1324                        (long long)lcn);
1325        /*
1326         * Attempt to get the cluster following the last allocated cluster by
1327         * hand as it may be in the MFT zone so the allocator would not give it
1328         * to us.
1329         */
1330        ll = lcn >> 3;
1331        page = ntfs_map_page(vol->lcnbmp_ino->i_mapping,
1332                        ll >> PAGE_SHIFT);
1333        if (IS_ERR(page)) {
1334                up_write(&mftbmp_ni->runlist.lock);
1335                ntfs_error(vol->sb, "Failed to read from lcn bitmap.");
1336                return PTR_ERR(page);
1337        }
1338        b = (u8*)page_address(page) + (ll & ~PAGE_MASK);
1339        tb = 1 << (lcn & 7ull);
1340        down_write(&vol->lcnbmp_lock);
1341        if (*b != 0xff && !(*b & tb)) {
1342                /* Next cluster is free, allocate it. */
1343                *b |= tb;
1344                flush_dcache_page(page);
1345                set_page_dirty(page);
1346                up_write(&vol->lcnbmp_lock);
1347                ntfs_unmap_page(page);
1348                /* Update the mft bitmap runlist. */
1349                rl->length++;
1350                rl[1].vcn++;
1351                status.added_cluster = 1;
1352                ntfs_debug("Appending one cluster to mft bitmap.");
1353        } else {
1354                up_write(&vol->lcnbmp_lock);
1355                ntfs_unmap_page(page);
1356                /* Allocate a cluster from the DATA_ZONE. */
1357                rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE,
1358                                true);
1359                if (IS_ERR(rl2)) {
1360                        up_write(&mftbmp_ni->runlist.lock);
1361                        ntfs_error(vol->sb, "Failed to allocate a cluster for "
1362                                        "the mft bitmap.");
1363                        return PTR_ERR(rl2);
1364                }
1365                rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2);
1366                if (IS_ERR(rl)) {
1367                        up_write(&mftbmp_ni->runlist.lock);
1368                        ntfs_error(vol->sb, "Failed to merge runlists for mft "
1369                                        "bitmap.");
1370                        if (ntfs_cluster_free_from_rl(vol, rl2)) {
1371                                ntfs_error(vol->sb, "Failed to deallocate "
1372                                                "allocated cluster.%s", es);
1373                                NVolSetErrors(vol);
1374                        }
1375                        ntfs_free(rl2);
1376                        return PTR_ERR(rl);
1377                }
1378                mftbmp_ni->runlist.rl = rl;
1379                status.added_run = 1;
1380                ntfs_debug("Adding one run to mft bitmap.");
1381                /* Find the last run in the new runlist. */
1382                for (; rl[1].length; rl++)
1383                        ;
1384        }
1385        /*
1386         * Update the attribute record as well.  Note: @rl is the last
1387         * (non-terminator) runlist element of mft bitmap.
1388         */
1389        mrec = map_mft_record(mft_ni);
1390        if (IS_ERR(mrec)) {
1391                ntfs_error(vol->sb, "Failed to map mft record.");
1392                ret = PTR_ERR(mrec);
1393                goto undo_alloc;
1394        }
1395        ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1396        if (unlikely(!ctx)) {
1397                ntfs_error(vol->sb, "Failed to get search context.");
1398                ret = -ENOMEM;
1399                goto undo_alloc;
1400        }
1401        ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1402                        mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1403                        0, ctx);
1404        if (unlikely(ret)) {
1405                ntfs_error(vol->sb, "Failed to find last attribute extent of "
1406                                "mft bitmap attribute.");
1407                if (ret == -ENOENT)
1408                        ret = -EIO;
1409                goto undo_alloc;
1410        }
1411        a = ctx->attr;
1412        ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1413        /* Search back for the previous last allocated cluster of mft bitmap. */
1414        for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) {
1415                if (ll >= rl2->vcn)
1416                        break;
1417        }
1418        BUG_ON(ll < rl2->vcn);
1419        BUG_ON(ll >= rl2->vcn + rl2->length);
1420        /* Get the size for the new mapping pairs array for this extent. */
1421        mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1422        if (unlikely(mp_size <= 0)) {
1423                ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1424                                "mft bitmap attribute extent.");
1425                ret = mp_size;
1426                if (!ret)
1427                        ret = -EIO;
1428                goto undo_alloc;
1429        }
1430        /* Expand the attribute record if necessary. */
1431        old_alen = le32_to_cpu(a->length);
1432        ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1433                        le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1434        if (unlikely(ret)) {
1435                if (ret != -ENOSPC) {
1436                        ntfs_error(vol->sb, "Failed to resize attribute "
1437                                        "record for mft bitmap attribute.");
1438                        goto undo_alloc;
1439                }
1440                // TODO: Deal with this by moving this extent to a new mft
1441                // record or by starting a new extent in a new mft record or by
1442                // moving other attributes out of this mft record.
1443                // Note: It will need to be a special mft record and if none of
1444                // those are available it gets rather complicated...
1445                ntfs_error(vol->sb, "Not enough space in this mft record to "
1446                                "accommodate extended mft bitmap attribute "
1447                                "extent.  Cannot handle this yet.");
1448                ret = -EOPNOTSUPP;
1449                goto undo_alloc;
1450        }
1451        status.mp_rebuilt = 1;
1452        /* Generate the mapping pairs array directly into the attr record. */
1453        ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1454                        le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1455                        mp_size, rl2, ll, -1, NULL);
1456        if (unlikely(ret)) {
1457                ntfs_error(vol->sb, "Failed to build mapping pairs array for "
1458                                "mft bitmap attribute.");
1459                goto undo_alloc;
1460        }
1461        /* Update the highest_vcn. */
1462        a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1463        /*
1464         * We now have extended the mft bitmap allocated_size by one cluster.
1465         * Reflect this in the ntfs_inode structure and the attribute record.
1466         */
1467        if (a->data.non_resident.lowest_vcn) {
1468                /*
1469                 * We are not in the first attribute extent, switch to it, but
1470                 * first ensure the changes will make it to disk later.
1471                 */
1472                flush_dcache_mft_record_page(ctx->ntfs_ino);
1473                mark_mft_record_dirty(ctx->ntfs_ino);
1474                ntfs_attr_reinit_search_ctx(ctx);
1475                ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1476                                mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL,
1477                                0, ctx);
1478                if (unlikely(ret)) {
1479                        ntfs_error(vol->sb, "Failed to find first attribute "
1480                                        "extent of mft bitmap attribute.");
1481                        goto restore_undo_alloc;
1482                }
1483                a = ctx->attr;
1484        }
1485        write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1486        mftbmp_ni->allocated_size += vol->cluster_size;
1487        a->data.non_resident.allocated_size =
1488                        cpu_to_sle64(mftbmp_ni->allocated_size);
1489        write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1490        /* Ensure the changes make it to disk. */
1491        flush_dcache_mft_record_page(ctx->ntfs_ino);
1492        mark_mft_record_dirty(ctx->ntfs_ino);
1493        ntfs_attr_put_search_ctx(ctx);
1494        unmap_mft_record(mft_ni);
1495        up_write(&mftbmp_ni->runlist.lock);
1496        ntfs_debug("Done.");
1497        return 0;
1498restore_undo_alloc:
1499        ntfs_attr_reinit_search_ctx(ctx);
1500        if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1501                        mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1502                        0, ctx)) {
1503                ntfs_error(vol->sb, "Failed to find last attribute extent of "
1504                                "mft bitmap attribute.%s", es);
1505                write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1506                mftbmp_ni->allocated_size += vol->cluster_size;
1507                write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1508                ntfs_attr_put_search_ctx(ctx);
1509                unmap_mft_record(mft_ni);
1510                up_write(&mftbmp_ni->runlist.lock);
1511                /*
1512                 * The only thing that is now wrong is ->allocated_size of the
1513                 * base attribute extent which chkdsk should be able to fix.
1514                 */
1515                NVolSetErrors(vol);
1516                return ret;
1517        }
1518        a = ctx->attr;
1519        a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2);
1520undo_alloc:
1521        if (status.added_cluster) {
1522                /* Truncate the last run in the runlist by one cluster. */
1523                rl->length--;
1524                rl[1].vcn--;
1525        } else if (status.added_run) {
1526                lcn = rl->lcn;
1527                /* Remove the last run from the runlist. */
1528                rl->lcn = rl[1].lcn;
1529                rl->length = 0;
1530        }
1531        /* Deallocate the cluster. */
1532        down_write(&vol->lcnbmp_lock);
1533        if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1534                ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es);
1535                NVolSetErrors(vol);
1536        }
1537        up_write(&vol->lcnbmp_lock);
1538        if (status.mp_rebuilt) {
1539                if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1540                                a->data.non_resident.mapping_pairs_offset),
1541                                old_alen - le16_to_cpu(
1542                                a->data.non_resident.mapping_pairs_offset),
1543                                rl2, ll, -1, NULL)) {
1544                        ntfs_error(vol->sb, "Failed to restore mapping pairs "
1545                                        "array.%s", es);
1546                        NVolSetErrors(vol);
1547                }
1548                if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1549                        ntfs_error(vol->sb, "Failed to restore attribute "
1550                                        "record.%s", es);
1551                        NVolSetErrors(vol);
1552                }
1553                flush_dcache_mft_record_page(ctx->ntfs_ino);
1554                mark_mft_record_dirty(ctx->ntfs_ino);
1555        }
1556        if (ctx)
1557                ntfs_attr_put_search_ctx(ctx);
1558        if (!IS_ERR(mrec))
1559                unmap_mft_record(mft_ni);
1560        up_write(&mftbmp_ni->runlist.lock);
1561        return ret;
1562}
1563
1564/**
1565 * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data
1566 * @vol:        volume on which to extend the mft bitmap attribute
1567 *
1568 * Extend the initialized portion of the mft bitmap attribute on the ntfs
1569 * volume @vol by 8 bytes.
1570 *
1571 * Note:  Only changes initialized_size and data_size, i.e. requires that
1572 * allocated_size is big enough to fit the new initialized_size.
1573 *
1574 * Return 0 on success and -error on error.
1575 *
1576 * Locking: Caller must hold vol->mftbmp_lock for writing.
1577 */
1578static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol)
1579{
1580        s64 old_data_size, old_initialized_size;
1581        unsigned long flags;
1582        struct inode *mftbmp_vi;
1583        ntfs_inode *mft_ni, *mftbmp_ni;
1584        ntfs_attr_search_ctx *ctx;
1585        MFT_RECORD *mrec;
1586        ATTR_RECORD *a;
1587        int ret;
1588
1589        ntfs_debug("Extending mft bitmap initiailized (and data) size.");
1590        mft_ni = NTFS_I(vol->mft_ino);
1591        mftbmp_vi = vol->mftbmp_ino;
1592        mftbmp_ni = NTFS_I(mftbmp_vi);
1593        /* Get the attribute record. */
1594        mrec = map_mft_record(mft_ni);
1595        if (IS_ERR(mrec)) {
1596                ntfs_error(vol->sb, "Failed to map mft record.");
1597                return PTR_ERR(mrec);
1598        }
1599        ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1600        if (unlikely(!ctx)) {
1601                ntfs_error(vol->sb, "Failed to get search context.");
1602                ret = -ENOMEM;
1603                goto unm_err_out;
1604        }
1605        ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1606                        mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx);
1607        if (unlikely(ret)) {
1608                ntfs_error(vol->sb, "Failed to find first attribute extent of "
1609                                "mft bitmap attribute.");
1610                if (ret == -ENOENT)
1611                        ret = -EIO;
1612                goto put_err_out;
1613        }
1614        a = ctx->attr;
1615        write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1616        old_data_size = i_size_read(mftbmp_vi);
1617        old_initialized_size = mftbmp_ni->initialized_size;
1618        /*
1619         * We can simply update the initialized_size before filling the space
1620         * with zeroes because the caller is holding the mft bitmap lock for
1621         * writing which ensures that no one else is trying to access the data.
1622         */
1623        mftbmp_ni->initialized_size += 8;
1624        a->data.non_resident.initialized_size =
1625                        cpu_to_sle64(mftbmp_ni->initialized_size);
1626        if (mftbmp_ni->initialized_size > old_data_size) {
1627                i_size_write(mftbmp_vi, mftbmp_ni->initialized_size);
1628                a->data.non_resident.data_size =
1629                                cpu_to_sle64(mftbmp_ni->initialized_size);
1630        }
1631        write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1632        /* Ensure the changes make it to disk. */
1633        flush_dcache_mft_record_page(ctx->ntfs_ino);
1634        mark_mft_record_dirty(ctx->ntfs_ino);
1635        ntfs_attr_put_search_ctx(ctx);
1636        unmap_mft_record(mft_ni);
1637        /* Initialize the mft bitmap attribute value with zeroes. */
1638        ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0);
1639        if (likely(!ret)) {
1640                ntfs_debug("Done.  (Wrote eight initialized bytes to mft "
1641                                "bitmap.");
1642                return 0;
1643        }
1644        ntfs_error(vol->sb, "Failed to write to mft bitmap.");
1645        /* Try to recover from the error. */
1646        mrec = map_mft_record(mft_ni);
1647        if (IS_ERR(mrec)) {
1648                ntfs_error(vol->sb, "Failed to map mft record.%s", es);
1649                NVolSetErrors(vol);
1650                return ret;
1651        }
1652        ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1653        if (unlikely(!ctx)) {
1654                ntfs_error(vol->sb, "Failed to get search context.%s", es);
1655                NVolSetErrors(vol);
1656                goto unm_err_out;
1657        }
1658        if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1659                        mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) {
1660                ntfs_error(vol->sb, "Failed to find first attribute extent of "
1661                                "mft bitmap attribute.%s", es);
1662                NVolSetErrors(vol);
1663put_err_out:
1664                ntfs_attr_put_search_ctx(ctx);
1665unm_err_out:
1666                unmap_mft_record(mft_ni);
1667                goto err_out;
1668        }
1669        a = ctx->attr;
1670        write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1671        mftbmp_ni->initialized_size = old_initialized_size;
1672        a->data.non_resident.initialized_size =
1673                        cpu_to_sle64(old_initialized_size);
1674        if (i_size_read(mftbmp_vi) != old_data_size) {
1675                i_size_write(mftbmp_vi, old_data_size);
1676                a->data.non_resident.data_size = cpu_to_sle64(old_data_size);
1677        }
1678        write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1679        flush_dcache_mft_record_page(ctx->ntfs_ino);
1680        mark_mft_record_dirty(ctx->ntfs_ino);
1681        ntfs_attr_put_search_ctx(ctx);
1682        unmap_mft_record(mft_ni);
1683#ifdef DEBUG
1684        read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1685        ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, "
1686                        "data_size 0x%llx, initialized_size 0x%llx.",
1687                        (long long)mftbmp_ni->allocated_size,
1688                        (long long)i_size_read(mftbmp_vi),
1689                        (long long)mftbmp_ni->initialized_size);
1690        read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1691#endif /* DEBUG */
1692err_out:
1693        return ret;
1694}
1695
1696/**
1697 * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute
1698 * @vol:        volume on which to extend the mft data attribute
1699 *
1700 * Extend the mft data attribute on the ntfs volume @vol by 16 mft records
1701 * worth of clusters or if not enough space for this by one mft record worth
1702 * of clusters.
1703 *
1704 * Note:  Only changes allocated_size, i.e. does not touch initialized_size or
1705 * data_size.
1706 *
1707 * Return 0 on success and -errno on error.
1708 *
1709 * Locking: - Caller must hold vol->mftbmp_lock for writing.
1710 *          - This function takes NTFS_I(vol->mft_ino)->runlist.lock for
1711 *            writing and releases it before returning.
1712 *          - This function calls functions which take vol->lcnbmp_lock for
1713 *            writing and release it before returning.
1714 */
1715static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol)
1716{
1717        LCN lcn;
1718        VCN old_last_vcn;
1719        s64 min_nr, nr, ll;
1720        unsigned long flags;
1721        ntfs_inode *mft_ni;
1722        runlist_element *rl, *rl2;
1723        ntfs_attr_search_ctx *ctx = NULL;
1724        MFT_RECORD *mrec;
1725        ATTR_RECORD *a = NULL;
1726        int ret, mp_size;
1727        u32 old_alen = 0;
1728        bool mp_rebuilt = false;
1729
1730        ntfs_debug("Extending mft data allocation.");
1731        mft_ni = NTFS_I(vol->mft_ino);
1732        /*
1733         * Determine the preferred allocation location, i.e. the last lcn of
1734         * the mft data attribute.  The allocated size of the mft data
1735         * attribute cannot be zero so we are ok to do this.
1736         */
1737        down_write(&mft_ni->runlist.lock);
1738        read_lock_irqsave(&mft_ni->size_lock, flags);
1739        ll = mft_ni->allocated_size;
1740        read_unlock_irqrestore(&mft_ni->size_lock, flags);
1741        rl = ntfs_attr_find_vcn_nolock(mft_ni,
1742                        (ll - 1) >> vol->cluster_size_bits, NULL);
1743        if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1744                up_write(&mft_ni->runlist.lock);
1745                ntfs_error(vol->sb, "Failed to determine last allocated "
1746                                "cluster of mft data attribute.");
1747                if (!IS_ERR(rl))
1748                        ret = -EIO;
1749                else
1750                        ret = PTR_ERR(rl);
1751                return ret;
1752        }
1753        lcn = rl->lcn + rl->length;
1754        ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn);
1755        /* Minimum allocation is one mft record worth of clusters. */
1756        min_nr = vol->mft_record_size >> vol->cluster_size_bits;
1757        if (!min_nr)
1758                min_nr = 1;
1759        /* Want to allocate 16 mft records worth of clusters. */
1760        nr = vol->mft_record_size << 4 >> vol->cluster_size_bits;
1761        if (!nr)
1762                nr = min_nr;
1763        /* Ensure we do not go above 2^32-1 mft records. */
1764        read_lock_irqsave(&mft_ni->size_lock, flags);
1765        ll = mft_ni->allocated_size;
1766        read_unlock_irqrestore(&mft_ni->size_lock, flags);
1767        if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1768                        vol->mft_record_size_bits >= (1ll << 32))) {
1769                nr = min_nr;
1770                if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1771                                vol->mft_record_size_bits >= (1ll << 32))) {
1772                        ntfs_warning(vol->sb, "Cannot allocate mft record "
1773                                        "because the maximum number of inodes "
1774                                        "(2^32) has already been reached.");
1775                        up_write(&mft_ni->runlist.lock);
1776                        return -ENOSPC;
1777                }
1778        }
1779        ntfs_debug("Trying mft data allocation with %s cluster count %lli.",
1780                        nr > min_nr ? "default" : "minimal", (long long)nr);
1781        old_last_vcn = rl[1].vcn;
1782        do {
1783                rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE,
1784                                true);
1785                if (likely(!IS_ERR(rl2)))
1786                        break;
1787                if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) {
1788                        ntfs_error(vol->sb, "Failed to allocate the minimal "
1789                                        "number of clusters (%lli) for the "
1790                                        "mft data attribute.", (long long)nr);
1791                        up_write(&mft_ni->runlist.lock);
1792                        return PTR_ERR(rl2);
1793                }
1794                /*
1795                 * There is not enough space to do the allocation, but there
1796                 * might be enough space to do a minimal allocation so try that
1797                 * before failing.
1798                 */
1799                nr = min_nr;
1800                ntfs_debug("Retrying mft data allocation with minimal cluster "
1801                                "count %lli.", (long long)nr);
1802        } while (1);
1803        rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2);
1804        if (IS_ERR(rl)) {
1805                up_write(&mft_ni->runlist.lock);
1806                ntfs_error(vol->sb, "Failed to merge runlists for mft data "
1807                                "attribute.");
1808                if (ntfs_cluster_free_from_rl(vol, rl2)) {
1809                        ntfs_error(vol->sb, "Failed to deallocate clusters "
1810                                        "from the mft data attribute.%s", es);
1811                        NVolSetErrors(vol);
1812                }
1813                ntfs_free(rl2);
1814                return PTR_ERR(rl);
1815        }
1816        mft_ni->runlist.rl = rl;
1817        ntfs_debug("Allocated %lli clusters.", (long long)nr);
1818        /* Find the last run in the new runlist. */
1819        for (; rl[1].length; rl++)
1820                ;
1821        /* Update the attribute record as well. */
1822        mrec = map_mft_record(mft_ni);
1823        if (IS_ERR(mrec)) {
1824                ntfs_error(vol->sb, "Failed to map mft record.");
1825                ret = PTR_ERR(mrec);
1826                goto undo_alloc;
1827        }
1828        ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1829        if (unlikely(!ctx)) {
1830                ntfs_error(vol->sb, "Failed to get search context.");
1831                ret = -ENOMEM;
1832                goto undo_alloc;
1833        }
1834        ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1835                        CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx);
1836        if (unlikely(ret)) {
1837                ntfs_error(vol->sb, "Failed to find last attribute extent of "
1838                                "mft data attribute.");
1839                if (ret == -ENOENT)
1840                        ret = -EIO;
1841                goto undo_alloc;
1842        }
1843        a = ctx->attr;
1844        ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1845        /* Search back for the previous last allocated cluster of mft bitmap. */
1846        for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) {
1847                if (ll >= rl2->vcn)
1848                        break;
1849        }
1850        BUG_ON(ll < rl2->vcn);
1851        BUG_ON(ll >= rl2->vcn + rl2->length);
1852        /* Get the size for the new mapping pairs array for this extent. */
1853        mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1854        if (unlikely(mp_size <= 0)) {
1855                ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1856                                "mft data attribute extent.");
1857                ret = mp_size;
1858                if (!ret)
1859                        ret = -EIO;
1860                goto undo_alloc;
1861        }
1862        /* Expand the attribute record if necessary. */
1863        old_alen = le32_to_cpu(a->length);
1864        ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1865                        le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1866        if (unlikely(ret)) {
1867                if (ret != -ENOSPC) {
1868                        ntfs_error(vol->sb, "Failed to resize attribute "
1869                                        "record for mft data attribute.");
1870                        goto undo_alloc;
1871                }
1872                // TODO: Deal with this by moving this extent to a new mft
1873                // record or by starting a new extent in a new mft record or by
1874                // moving other attributes out of this mft record.
1875                // Note: Use the special reserved mft records and ensure that
1876                // this extent is not required to find the mft record in
1877                // question.  If no free special records left we would need to
1878                // move an existing record away, insert ours in its place, and
1879                // then place the moved record into the newly allocated space
1880                // and we would then need to update all references to this mft
1881                // record appropriately.  This is rather complicated...
1882                ntfs_error(vol->sb, "Not enough space in this mft record to "
1883                                "accommodate extended mft data attribute "
1884                                "extent.  Cannot handle this yet.");
1885                ret = -EOPNOTSUPP;
1886                goto undo_alloc;
1887        }
1888        mp_rebuilt = true;
1889        /* Generate the mapping pairs array directly into the attr record. */
1890        ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1891                        le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1892                        mp_size, rl2, ll, -1, NULL);
1893        if (unlikely(ret)) {
1894                ntfs_error(vol->sb, "Failed to build mapping pairs array of "
1895                                "mft data attribute.");
1896                goto undo_alloc;
1897        }
1898        /* Update the highest_vcn. */
1899        a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1900        /*
1901         * We now have extended the mft data allocated_size by nr clusters.
1902         * Reflect this in the ntfs_inode structure and the attribute record.
1903         * @rl is the last (non-terminator) runlist element of mft data
1904         * attribute.
1905         */
1906        if (a->data.non_resident.lowest_vcn) {
1907                /*
1908                 * We are not in the first attribute extent, switch to it, but
1909                 * first ensure the changes will make it to disk later.
1910                 */
1911                flush_dcache_mft_record_page(ctx->ntfs_ino);
1912                mark_mft_record_dirty(ctx->ntfs_ino);
1913                ntfs_attr_reinit_search_ctx(ctx);
1914                ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name,
1915                                mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0,
1916                                ctx);
1917                if (unlikely(ret)) {
1918                        ntfs_error(vol->sb, "Failed to find first attribute "
1919                                        "extent of mft data attribute.");
1920                        goto restore_undo_alloc;
1921                }
1922                a = ctx->attr;
1923        }
1924        write_lock_irqsave(&mft_ni->size_lock, flags);
1925        mft_ni->allocated_size += nr << vol->cluster_size_bits;
1926        a->data.non_resident.allocated_size =
1927                        cpu_to_sle64(mft_ni->allocated_size);
1928        write_unlock_irqrestore(&mft_ni->size_lock, flags);
1929        /* Ensure the changes make it to disk. */
1930        flush_dcache_mft_record_page(ctx->ntfs_ino);
1931        mark_mft_record_dirty(ctx->ntfs_ino);
1932        ntfs_attr_put_search_ctx(ctx);
1933        unmap_mft_record(mft_ni);
1934        up_write(&mft_ni->runlist.lock);
1935        ntfs_debug("Done.");
1936        return 0;
1937restore_undo_alloc:
1938        ntfs_attr_reinit_search_ctx(ctx);
1939        if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1940                        CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) {
1941                ntfs_error(vol->sb, "Failed to find last attribute extent of "
1942                                "mft data attribute.%s", es);
1943                write_lock_irqsave(&mft_ni->size_lock, flags);
1944                mft_ni->allocated_size += nr << vol->cluster_size_bits;
1945                write_unlock_irqrestore(&mft_ni->size_lock, flags);
1946                ntfs_attr_put_search_ctx(ctx);
1947                unmap_mft_record(mft_ni);
1948                up_write(&mft_ni->runlist.lock);
1949                /*
1950                 * The only thing that is now wrong is ->allocated_size of the
1951                 * base attribute extent which chkdsk should be able to fix.
1952                 */
1953                NVolSetErrors(vol);
1954                return ret;
1955        }
1956        ctx->attr->data.non_resident.highest_vcn =
1957                        cpu_to_sle64(old_last_vcn - 1);
1958undo_alloc:
1959        if (ntfs_cluster_free(mft_ni, old_last_vcn, -1, ctx) < 0) {
1960                ntfs_error(vol->sb, "Failed to free clusters from mft data "
1961                                "attribute.%s", es);
1962                NVolSetErrors(vol);
1963        }
1964        a = ctx->attr;
1965        if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) {
1966                ntfs_error(vol->sb, "Failed to truncate mft data attribute "
1967                                "runlist.%s", es);
1968                NVolSetErrors(vol);
1969        }
1970        if (mp_rebuilt && !IS_ERR(ctx->mrec)) {
1971                if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1972                                a->data.non_resident.mapping_pairs_offset),
1973                                old_alen - le16_to_cpu(
1974                                a->data.non_resident.mapping_pairs_offset),
1975                                rl2, ll, -1, NULL)) {
1976                        ntfs_error(vol->sb, "Failed to restore mapping pairs "
1977                                        "array.%s", es);
1978                        NVolSetErrors(vol);
1979                }
1980                if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1981                        ntfs_error(vol->sb, "Failed to restore attribute "
1982                                        "record.%s", es);
1983                        NVolSetErrors(vol);
1984                }
1985                flush_dcache_mft_record_page(ctx->ntfs_ino);
1986                mark_mft_record_dirty(ctx->ntfs_ino);
1987        } else if (IS_ERR(ctx->mrec)) {
1988                ntfs_error(vol->sb, "Failed to restore attribute search "
1989                                "context.%s", es);
1990                NVolSetErrors(vol);
1991        }
1992        if (ctx)
1993                ntfs_attr_put_search_ctx(ctx);
1994        if (!IS_ERR(mrec))
1995                unmap_mft_record(mft_ni);
1996        up_write(&mft_ni->runlist.lock);
1997        return ret;
1998}
1999
2000/**
2001 * ntfs_mft_record_layout - layout an mft record into a memory buffer
2002 * @vol:        volume to which the mft record will belong
2003 * @mft_no:     mft reference specifying the mft record number
2004 * @m:          destination buffer of size >= @vol->mft_record_size bytes
2005 *
2006 * Layout an empty, unused mft record with the mft record number @mft_no into
2007 * the buffer @m.  The volume @vol is needed because the mft record structure
2008 * was modified in NTFS 3.1 so we need to know which volume version this mft
2009 * record will be used on.
2010 *
2011 * Return 0 on success and -errno on error.
2012 */
2013static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no,
2014                MFT_RECORD *m)
2015{
2016        ATTR_RECORD *a;
2017
2018        ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2019        if (mft_no >= (1ll << 32)) {
2020                ntfs_error(vol->sb, "Mft record number 0x%llx exceeds "
2021                                "maximum of 2^32.", (long long)mft_no);
2022                return -ERANGE;
2023        }
2024        /* Start by clearing the whole mft record to gives us a clean slate. */
2025        memset(m, 0, vol->mft_record_size);
2026        /* Aligned to 2-byte boundary. */
2027        if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver))
2028                m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1);
2029        else {
2030                m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1);
2031                /*
2032                 * Set the NTFS 3.1+ specific fields while we know that the
2033                 * volume version is 3.1+.
2034                 */
2035                m->reserved = 0;
2036                m->mft_record_number = cpu_to_le32((u32)mft_no);
2037        }
2038        m->magic = magic_FILE;
2039        if (vol->mft_record_size >= NTFS_BLOCK_SIZE)
2040                m->usa_count = cpu_to_le16(vol->mft_record_size /
2041                                NTFS_BLOCK_SIZE + 1);
2042        else {
2043                m->usa_count = cpu_to_le16(1);
2044                ntfs_warning(vol->sb, "Sector size is bigger than mft record "
2045                                "size.  Setting usa_count to 1.  If chkdsk "
2046                                "reports this as corruption, please email "
2047                                "linux-ntfs-dev@lists.sourceforge.net stating "
2048                                "that you saw this message and that the "
2049                                "modified filesystem created was corrupt.  "
2050                                "Thank you.");
2051        }
2052        /* Set the update sequence number to 1. */
2053        *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1);
2054        m->lsn = 0;
2055        m->sequence_number = cpu_to_le16(1);
2056        m->link_count = 0;
2057        /*
2058         * Place the attributes straight after the update sequence array,
2059         * aligned to 8-byte boundary.
2060         */
2061        m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) +
2062                        (le16_to_cpu(m->usa_count) << 1) + 7) & ~7);
2063        m->flags = 0;
2064        /*
2065         * Using attrs_offset plus eight bytes (for the termination attribute).
2066         * attrs_offset is already aligned to 8-byte boundary, so no need to
2067         * align again.
2068         */
2069        m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8);
2070        m->bytes_allocated = cpu_to_le32(vol->mft_record_size);
2071        m->base_mft_record = 0;
2072        m->next_attr_instance = 0;
2073        /* Add the termination attribute. */
2074        a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset));
2075        a->type = AT_END;
2076        a->length = 0;
2077        ntfs_debug("Done.");
2078        return 0;
2079}
2080
2081/**
2082 * ntfs_mft_record_format - format an mft record on an ntfs volume
2083 * @vol:        volume on which to format the mft record
2084 * @mft_no:     mft record number to format
2085 *
2086 * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused
2087 * mft record into the appropriate place of the mft data attribute.  This is
2088 * used when extending the mft data attribute.
2089 *
2090 * Return 0 on success and -errno on error.
2091 */
2092static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no)
2093{
2094        loff_t i_size;
2095        struct inode *mft_vi = vol->mft_ino;
2096        struct page *page;
2097        MFT_RECORD *m;
2098        pgoff_t index, end_index;
2099        unsigned int ofs;
2100        int err;
2101
2102        ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2103        /*
2104         * The index into the page cache and the offset within the page cache
2105         * page of the wanted mft record.
2106         */
2107        index = mft_no << vol->mft_record_size_bits >> PAGE_SHIFT;
2108        ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_MASK;
2109        /* The maximum valid index into the page cache for $MFT's data. */
2110        i_size = i_size_read(mft_vi);
2111        end_index = i_size >> PAGE_SHIFT;
2112        if (unlikely(index >= end_index)) {
2113                if (unlikely(index > end_index || ofs + vol->mft_record_size >=
2114                                (i_size & ~PAGE_MASK))) {
2115                        ntfs_error(vol->sb, "Tried to format non-existing mft "
2116                                        "record 0x%llx.", (long long)mft_no);
2117                        return -ENOENT;
2118                }
2119        }
2120        /* Read, map, and pin the page containing the mft record. */
2121        page = ntfs_map_page(mft_vi->i_mapping, index);
2122        if (IS_ERR(page)) {
2123                ntfs_error(vol->sb, "Failed to map page containing mft record "
2124                                "to format 0x%llx.", (long long)mft_no);
2125                return PTR_ERR(page);
2126        }
2127        lock_page(page);
2128        BUG_ON(!PageUptodate(page));
2129        ClearPageUptodate(page);
2130        m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2131        err = ntfs_mft_record_layout(vol, mft_no, m);
2132        if (unlikely(err)) {
2133                ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.",
2134                                (long long)mft_no);
2135                SetPageUptodate(page);
2136                unlock_page(page);
2137                ntfs_unmap_page(page);
2138                return err;
2139        }
2140        flush_dcache_page(page);
2141        SetPageUptodate(page);
2142        unlock_page(page);
2143        /*
2144         * Make sure the mft record is written out to disk.  We could use
2145         * ilookup5() to check if an inode is in icache and so on but this is
2146         * unnecessary as ntfs_writepage() will write the dirty record anyway.
2147         */
2148        mark_ntfs_record_dirty(page, ofs);
2149        ntfs_unmap_page(page);
2150        ntfs_debug("Done.");
2151        return 0;
2152}
2153
2154/**
2155 * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume
2156 * @vol:        [IN]  volume on which to allocate the mft record
2157 * @mode:       [IN]  mode if want a file or directory, i.e. base inode or 0
2158 * @base_ni:    [IN]  open base inode if allocating an extent mft record or NULL
2159 * @mrec:       [OUT] on successful return this is the mapped mft record
2160 *
2161 * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol.
2162 *
2163 * If @base_ni is NULL make the mft record a base mft record, i.e. a file or
2164 * direvctory inode, and allocate it at the default allocator position.  In
2165 * this case @mode is the file mode as given to us by the caller.  We in
2166 * particular use @mode to distinguish whether a file or a directory is being
2167 * created (S_IFDIR(mode) and S_IFREG(mode), respectively).
2168 *
2169 * If @base_ni is not NULL make the allocated mft record an extent record,
2170 * allocate it starting at the mft record after the base mft record and attach
2171 * the allocated and opened ntfs inode to the base inode @base_ni.  In this
2172 * case @mode must be 0 as it is meaningless for extent inodes.
2173 *
2174 * You need to check the return value with IS_ERR().  If false, the function
2175 * was successful and the return value is the now opened ntfs inode of the
2176 * allocated mft record.  *@mrec is then set to the allocated, mapped, pinned,
2177 * and locked mft record.  If IS_ERR() is true, the function failed and the
2178 * error code is obtained from PTR_ERR(return value).  *@mrec is undefined in
2179 * this case.
2180 *
2181 * Allocation strategy:
2182 *
2183 * To find a free mft record, we scan the mft bitmap for a zero bit.  To
2184 * optimize this we start scanning at the place specified by @base_ni or if
2185 * @base_ni is NULL we start where we last stopped and we perform wrap around
2186 * when we reach the end.  Note, we do not try to allocate mft records below
2187 * number 24 because numbers 0 to 15 are the defined system files anyway and 16
2188 * to 24 are special in that they are used for storing extension mft records
2189 * for the $DATA attribute of $MFT.  This is required to avoid the possibility
2190 * of creating a runlist with a circular dependency which once written to disk
2191 * can never be read in again.  Windows will only use records 16 to 24 for
2192 * normal files if the volume is completely out of space.  We never use them
2193 * which means that when the volume is really out of space we cannot create any
2194 * more files while Windows can still create up to 8 small files.  We can start
2195 * doing this at some later time, it does not matter much for now.
2196 *
2197 * When scanning the mft bitmap, we only search up to the last allocated mft
2198 * record.  If there are no free records left in the range 24 to number of
2199 * allocated mft records, then we extend the $MFT/$DATA attribute in order to
2200 * create free mft records.  We extend the allocated size of $MFT/$DATA by 16
2201 * records at a time or one cluster, if cluster size is above 16kiB.  If there
2202 * is not sufficient space to do this, we try to extend by a single mft record
2203 * or one cluster, if cluster size is above the mft record size.
2204 *
2205 * No matter how many mft records we allocate, we initialize only the first
2206 * allocated mft record, incrementing mft data size and initialized size
2207 * accordingly, open an ntfs_inode for it and return it to the caller, unless
2208 * there are less than 24 mft records, in which case we allocate and initialize
2209 * mft records until we reach record 24 which we consider as the first free mft
2210 * record for use by normal files.
2211 *
2212 * If during any stage we overflow the initialized data in the mft bitmap, we
2213 * extend the initialized size (and data size) by 8 bytes, allocating another
2214 * cluster if required.  The bitmap data size has to be at least equal to the
2215 * number of mft records in the mft, but it can be bigger, in which case the
2216 * superflous bits are padded with zeroes.
2217 *
2218 * Thus, when we return successfully (IS_ERR() is false), we will have:
2219 *      - initialized / extended the mft bitmap if necessary,
2220 *      - initialized / extended the mft data if necessary,
2221 *      - set the bit corresponding to the mft record being allocated in the
2222 *        mft bitmap,
2223 *      - opened an ntfs_inode for the allocated mft record, and we will have
2224 *      - returned the ntfs_inode as well as the allocated mapped, pinned, and
2225 *        locked mft record.
2226 *
2227 * On error, the volume will be left in a consistent state and no record will
2228 * be allocated.  If rolling back a partial operation fails, we may leave some
2229 * inconsistent metadata in which case we set NVolErrors() so the volume is
2230 * left dirty when unmounted.
2231 *
2232 * Note, this function cannot make use of most of the normal functions, like
2233 * for example for attribute resizing, etc, because when the run list overflows
2234 * the base mft record and an attribute list is used, it is very important that
2235 * the extension mft records used to store the $DATA attribute of $MFT can be
2236 * reached without having to read the information contained inside them, as
2237 * this would make it impossible to find them in the first place after the
2238 * volume is unmounted.  $MFT/$BITMAP probably does not need to follow this
2239 * rule because the bitmap is not essential for finding the mft records, but on
2240 * the other hand, handling the bitmap in this special way would make life
2241 * easier because otherwise there might be circular invocations of functions
2242 * when reading the bitmap.
2243 */
2244ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode,
2245                ntfs_inode *base_ni, MFT_RECORD **mrec)
2246{
2247        s64 ll, bit, old_data_initialized, old_data_size;
2248        unsigned long flags;
2249        struct inode *vi;
2250        struct page *page;
2251        ntfs_inode *mft_ni, *mftbmp_ni, *ni;
2252        ntfs_attr_search_ctx *ctx;
2253        MFT_RECORD *m;
2254        ATTR_RECORD *a;
2255        pgoff_t index;
2256        unsigned int ofs;
2257        int err;
2258        le16 seq_no, usn;
2259        bool record_formatted = false;
2260
2261        if (base_ni) {
2262                ntfs_debug("Entering (allocating an extent mft record for "
2263                                "base mft record 0x%llx).",
2264                                (long long)base_ni->mft_no);
2265                /* @mode and @base_ni are mutually exclusive. */
2266                BUG_ON(mode);
2267        } else
2268                ntfs_debug("Entering (allocating a base mft record).");
2269        if (mode) {
2270                /* @mode and @base_ni are mutually exclusive. */
2271                BUG_ON(base_ni);
2272                /* We only support creation of normal files and directories. */
2273                if (!S_ISREG(mode) && !S_ISDIR(mode))
2274                        return ERR_PTR(-EOPNOTSUPP);
2275        }
2276        BUG_ON(!mrec);
2277        mft_ni = NTFS_I(vol->mft_ino);
2278        mftbmp_ni = NTFS_I(vol->mftbmp_ino);
2279        down_write(&vol->mftbmp_lock);
2280        bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni);
2281        if (bit >= 0) {
2282                ntfs_debug("Found and allocated free record (#1), bit 0x%llx.",
2283                                (long long)bit);
2284                goto have_alloc_rec;
2285        }
2286        if (bit != -ENOSPC) {
2287                up_write(&vol->mftbmp_lock);
2288                return ERR_PTR(bit);
2289        }
2290        /*
2291         * No free mft records left.  If the mft bitmap already covers more
2292         * than the currently used mft records, the next records are all free,
2293         * so we can simply allocate the first unused mft record.
2294         * Note: We also have to make sure that the mft bitmap at least covers
2295         * the first 24 mft records as they are special and whilst they may not
2296         * be in use, we do not allocate from them.
2297         */
2298        read_lock_irqsave(&mft_ni->size_lock, flags);
2299        ll = mft_ni->initialized_size >> vol->mft_record_size_bits;
2300        read_unlock_irqrestore(&mft_ni->size_lock, flags);
2301        read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2302        old_data_initialized = mftbmp_ni->initialized_size;
2303        read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2304        if (old_data_initialized << 3 > ll && old_data_initialized > 3) {
2305                bit = ll;
2306                if (bit < 24)
2307                        bit = 24;
2308                if (unlikely(bit >= (1ll << 32)))
2309                        goto max_err_out;
2310                ntfs_debug("Found free record (#2), bit 0x%llx.",
2311                                (long long)bit);
2312                goto found_free_rec;
2313        }
2314        /*
2315         * The mft bitmap needs to be expanded until it covers the first unused
2316         * mft record that we can allocate.
2317         * Note: The smallest mft record we allocate is mft record 24.
2318         */
2319        bit = old_data_initialized << 3;
2320        if (unlikely(bit >= (1ll << 32)))
2321                goto max_err_out;
2322        read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2323        old_data_size = mftbmp_ni->allocated_size;
2324        ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, "
2325                        "data_size 0x%llx, initialized_size 0x%llx.",
2326                        (long long)old_data_size,
2327                        (long long)i_size_read(vol->mftbmp_ino),
2328                        (long long)old_data_initialized);
2329        read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2330        if (old_data_initialized + 8 > old_data_size) {
2331                /* Need to extend bitmap by one more cluster. */
2332                ntfs_debug("mftbmp: initialized_size + 8 > allocated_size.");
2333                err = ntfs_mft_bitmap_extend_allocation_nolock(vol);
2334                if (unlikely(err)) {
2335                        up_write(&vol->mftbmp_lock);
2336                        goto err_out;
2337                }
2338#ifdef DEBUG
2339                read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2340                ntfs_debug("Status of mftbmp after allocation extension: "
2341                                "allocated_size 0x%llx, data_size 0x%llx, "
2342                                "initialized_size 0x%llx.",
2343                                (long long)mftbmp_ni->allocated_size,
2344                                (long long)i_size_read(vol->mftbmp_ino),
2345                                (long long)mftbmp_ni->initialized_size);
2346                read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2347#endif /* DEBUG */
2348        }
2349        /*
2350         * We now have sufficient allocated space, extend the initialized_size
2351         * as well as the data_size if necessary and fill the new space with
2352         * zeroes.
2353         */
2354        err = ntfs_mft_bitmap_extend_initialized_nolock(vol);
2355        if (unlikely(err)) {
2356                up_write(&vol->mftbmp_lock);
2357                goto err_out;
2358        }
2359#ifdef DEBUG
2360        read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2361        ntfs_debug("Status of mftbmp after initialized extension: "
2362                        "allocated_size 0x%llx, data_size 0x%llx, "
2363                        "initialized_size 0x%llx.",
2364                        (long long)mftbmp_ni->allocated_size,
2365                        (long long)i_size_read(vol->mftbmp_ino),
2366                        (long long)mftbmp_ni->initialized_size);
2367        read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2368#endif /* DEBUG */
2369        ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit);
2370found_free_rec:
2371        /* @bit is the found free mft record, allocate it in the mft bitmap. */
2372        ntfs_debug("At found_free_rec.");
2373        err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit);
2374        if (unlikely(err)) {
2375                ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap.");
2376                up_write(&vol->mftbmp_lock);
2377                goto err_out;
2378        }
2379        ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit);
2380have_alloc_rec:
2381        /*
2382         * The mft bitmap is now uptodate.  Deal with mft data attribute now.
2383         * Note, we keep hold of the mft bitmap lock for writing until all
2384         * modifications to the mft data attribute are complete, too, as they
2385         * will impact decisions for mft bitmap and mft record allocation done
2386         * by a parallel allocation and if the lock is not maintained a
2387         * parallel allocation could allocate the same mft record as this one.
2388         */
2389        ll = (bit + 1) << vol->mft_record_size_bits;
2390        read_lock_irqsave(&mft_ni->size_lock, flags);
2391        old_data_initialized = mft_ni->initialized_size;
2392        read_unlock_irqrestore(&mft_ni->size_lock, flags);
2393        if (ll <= old_data_initialized) {
2394                ntfs_debug("Allocated mft record already initialized.");
2395                goto mft_rec_already_initialized;
2396        }
2397        ntfs_debug("Initializing allocated mft record.");
2398        /*
2399         * The mft record is outside the initialized data.  Extend the mft data
2400         * attribute until it covers the allocated record.  The loop is only
2401         * actually traversed more than once when a freshly formatted volume is
2402         * first written to so it optimizes away nicely in the common case.
2403         */
2404        read_lock_irqsave(&mft_ni->size_lock, flags);
2405        ntfs_debug("Status of mft data before extension: "
2406                        "allocated_size 0x%llx, data_size 0x%llx, "
2407                        "initialized_size 0x%llx.",
2408                        (long long)mft_ni->allocated_size,
2409                        (long long)i_size_read(vol->mft_ino),
2410                        (long long)mft_ni->initialized_size);
2411        while (ll > mft_ni->allocated_size) {
2412                read_unlock_irqrestore(&mft_ni->size_lock, flags);
2413                err = ntfs_mft_data_extend_allocation_nolock(vol);
2414                if (unlikely(err)) {
2415                        ntfs_error(vol->sb, "Failed to extend mft data "
2416                                        "allocation.");
2417                        goto undo_mftbmp_alloc_nolock;
2418                }
2419                read_lock_irqsave(&mft_ni->size_lock, flags);
2420                ntfs_debug("Status of mft data after allocation extension: "
2421                                "allocated_size 0x%llx, data_size 0x%llx, "
2422                                "initialized_size 0x%llx.",
2423                                (long long)mft_ni->allocated_size,
2424                                (long long)i_size_read(vol->mft_ino),
2425                                (long long)mft_ni->initialized_size);
2426        }
2427        read_unlock_irqrestore(&mft_ni->size_lock, flags);
2428        /*
2429         * Extend mft data initialized size (and data size of course) to reach
2430         * the allocated mft record, formatting the mft records allong the way.
2431         * Note: We only modify the ntfs_inode structure as that is all that is
2432         * needed by ntfs_mft_record_format().  We will update the attribute
2433         * record itself in one fell swoop later on.
2434         */
2435        write_lock_irqsave(&mft_ni->size_lock, flags);
2436        old_data_initialized = mft_ni->initialized_size;
2437        old_data_size = vol->mft_ino->i_size;
2438        while (ll > mft_ni->initialized_size) {
2439                s64 new_initialized_size, mft_no;
2440                
2441                new_initialized_size = mft_ni->initialized_size +
2442                                vol->mft_record_size;
2443                mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits;
2444                if (new_initialized_size > i_size_read(vol->mft_ino))
2445                        i_size_write(vol->mft_ino, new_initialized_size);
2446                write_unlock_irqrestore(&mft_ni->size_lock, flags);
2447                ntfs_debug("Initializing mft record 0x%llx.",
2448                                (long long)mft_no);
2449                err = ntfs_mft_record_format(vol, mft_no);
2450                if (unlikely(err)) {
2451                        ntfs_error(vol->sb, "Failed to format mft record.");
2452                        goto undo_data_init;
2453                }
2454                write_lock_irqsave(&mft_ni->size_lock, flags);
2455                mft_ni->initialized_size = new_initialized_size;
2456        }
2457        write_unlock_irqrestore(&mft_ni->size_lock, flags);
2458        record_formatted = true;
2459        /* Update the mft data attribute record to reflect the new sizes. */
2460        m = map_mft_record(mft_ni);
2461        if (IS_ERR(m)) {
2462                ntfs_error(vol->sb, "Failed to map mft record.");
2463                err = PTR_ERR(m);
2464                goto undo_data_init;
2465        }
2466        ctx = ntfs_attr_get_search_ctx(mft_ni, m);
2467        if (unlikely(!ctx)) {
2468                ntfs_error(vol->sb, "Failed to get search context.");
2469                err = -ENOMEM;
2470                unmap_mft_record(mft_ni);
2471                goto undo_data_init;
2472        }
2473        err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
2474                        CASE_SENSITIVE, 0, NULL, 0, ctx);
2475        if (unlikely(err)) {
2476                ntfs_error(vol->sb, "Failed to find first attribute extent of "
2477                                "mft data attribute.");
2478                ntfs_attr_put_search_ctx(ctx);
2479                unmap_mft_record(mft_ni);
2480                goto undo_data_init;
2481        }
2482        a = ctx->attr;
2483        read_lock_irqsave(&mft_ni->size_lock, flags);
2484        a->data.non_resident.initialized_size =
2485                        cpu_to_sle64(mft_ni->initialized_size);
2486        a->data.non_resident.data_size =
2487                        cpu_to_sle64(i_size_read(vol->mft_ino));
2488        read_unlock_irqrestore(&mft_ni->size_lock, flags);
2489        /* Ensure the changes make it to disk. */
2490        flush_dcache_mft_record_page(ctx->ntfs_ino);
2491        mark_mft_record_dirty(ctx->ntfs_ino);
2492        ntfs_attr_put_search_ctx(ctx);
2493        unmap_mft_record(mft_ni);
2494        read_lock_irqsave(&mft_ni->size_lock, flags);
2495        ntfs_debug("Status of mft data after mft record initialization: "
2496                        "allocated_size 0x%llx, data_size 0x%llx, "
2497                        "initialized_size 0x%llx.",
2498                        (long long)mft_ni->allocated_size,
2499                        (long long)i_size_read(vol->mft_ino),
2500                        (long long)mft_ni->initialized_size);
2501        BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size);
2502        BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino));
2503        read_unlock_irqrestore(&mft_ni->size_lock, flags);
2504mft_rec_already_initialized:
2505        /*
2506         * We can finally drop the mft bitmap lock as the mft data attribute
2507         * has been fully updated.  The only disparity left is that the
2508         * allocated mft record still needs to be marked as in use to match the
2509         * set bit in the mft bitmap but this is actually not a problem since
2510         * this mft record is not referenced from anywhere yet and the fact
2511         * that it is allocated in the mft bitmap means that no-one will try to
2512         * allocate it either.
2513         */
2514        up_write(&vol->mftbmp_lock);
2515        /*
2516         * We now have allocated and initialized the mft record.  Calculate the
2517         * index of and the offset within the page cache page the record is in.
2518         */
2519        index = bit << vol->mft_record_size_bits >> PAGE_SHIFT;
2520        ofs = (bit << vol->mft_record_size_bits) & ~PAGE_MASK;
2521        /* Read, map, and pin the page containing the mft record. */
2522        page = ntfs_map_page(vol->mft_ino->i_mapping, index);
2523        if (IS_ERR(page)) {
2524                ntfs_error(vol->sb, "Failed to map page containing allocated "
2525                                "mft record 0x%llx.", (long long)bit);
2526                err = PTR_ERR(page);
2527                goto undo_mftbmp_alloc;
2528        }
2529        lock_page(page);
2530        BUG_ON(!PageUptodate(page));
2531        ClearPageUptodate(page);
2532        m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2533        /* If we just formatted the mft record no need to do it again. */
2534        if (!record_formatted) {
2535                /* Sanity check that the mft record is really not in use. */
2536                if (ntfs_is_file_record(m->magic) &&
2537                                (m->flags & MFT_RECORD_IN_USE)) {
2538                        ntfs_error(vol->sb, "Mft record 0x%llx was marked "
2539                                        "free in mft bitmap but is marked "
2540                                        "used itself.  Corrupt filesystem.  "
2541                                        "Unmount and run chkdsk.",
2542                                        (long long)bit);
2543                        err = -EIO;
2544                        SetPageUptodate(page);
2545                        unlock_page(page);
2546                        ntfs_unmap_page(page);
2547                        NVolSetErrors(vol);
2548                        goto undo_mftbmp_alloc;
2549                }
2550                /*
2551                 * We need to (re-)format the mft record, preserving the
2552                 * sequence number if it is not zero as well as the update
2553                 * sequence number if it is not zero or -1 (0xffff).  This
2554                 * means we do not need to care whether or not something went
2555                 * wrong with the previous mft record.
2556                 */
2557                seq_no = m->sequence_number;
2558                usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs));
2559                err = ntfs_mft_record_layout(vol, bit, m);
2560                if (unlikely(err)) {
2561                        ntfs_error(vol->sb, "Failed to layout allocated mft "
2562                                        "record 0x%llx.", (long long)bit);
2563                        SetPageUptodate(page);
2564                        unlock_page(page);
2565                        ntfs_unmap_page(page);
2566                        goto undo_mftbmp_alloc;
2567                }
2568                if (seq_no)
2569                        m->sequence_number = seq_no;
2570                if (usn && le16_to_cpu(usn) != 0xffff)
2571                        *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn;
2572        }
2573        /* Set the mft record itself in use. */
2574        m->flags |= MFT_RECORD_IN_USE;
2575        if (S_ISDIR(mode))
2576                m->flags |= MFT_RECORD_IS_DIRECTORY;
2577        flush_dcache_page(page);
2578        SetPageUptodate(page);
2579        if (base_ni) {
2580                MFT_RECORD *m_tmp;
2581
2582                /*
2583                 * Setup the base mft record in the extent mft record.  This
2584                 * completes initialization of the allocated extent mft record
2585                 * and we can simply use it with map_extent_mft_record().
2586                 */
2587                m->base_mft_record = MK_LE_MREF(base_ni->mft_no,
2588                                base_ni->seq_no);
2589                /*
2590                 * Allocate an extent inode structure for the new mft record,
2591                 * attach it to the base inode @base_ni and map, pin, and lock
2592                 * its, i.e. the allocated, mft record.
2593                 */
2594                m_tmp = map_extent_mft_record(base_ni, bit, &ni);
2595                if (IS_ERR(m_tmp)) {
2596                        ntfs_error(vol->sb, "Failed to map allocated extent "
2597                                        "mft record 0x%llx.", (long long)bit);
2598                        err = PTR_ERR(m_tmp);
2599                        /* Set the mft record itself not in use. */
2600                        m->flags &= cpu_to_le16(
2601                                        ~le16_to_cpu(MFT_RECORD_IN_USE));
2602                        flush_dcache_page(page);
2603                        /* Make sure the mft record is written out to disk. */
2604                        mark_ntfs_record_dirty(page, ofs);
2605                        unlock_page(page);
2606                        ntfs_unmap_page(page);
2607                        goto undo_mftbmp_alloc;
2608                }
2609                BUG_ON(m != m_tmp);
2610                /*
2611                 * Make sure the allocated mft record is written out to disk.
2612                 * No need to set the inode dirty because the caller is going
2613                 * to do that anyway after finishing with the new extent mft
2614                 * record (e.g. at a minimum a new attribute will be added to
2615                 * the mft record.
2616                 */
2617                mark_ntfs_record_dirty(page, ofs);
2618                unlock_page(page);
2619                /*
2620                 * Need to unmap the page since map_extent_mft_record() mapped
2621                 * it as well so we have it mapped twice at the moment.
2622                 */
2623                ntfs_unmap_page(page);
2624        } else {
2625                /*
2626                 * Allocate a new VFS inode and set it up.  NOTE: @vi->i_nlink
2627                 * is set to 1 but the mft record->link_count is 0.  The caller
2628                 * needs to bear this in mind.
2629                 */
2630                vi = new_inode(vol->sb);
2631                if (unlikely(!vi)) {
2632                        err = -ENOMEM;
2633                        /* Set the mft record itself not in use. */
2634                        m->flags &= cpu_to_le16(
2635                                        ~le16_to_cpu(MFT_RECORD_IN_USE));
2636                        flush_dcache_page(page);
2637                        /* Make sure the mft record is written out to disk. */
2638                        mark_ntfs_record_dirty(page, ofs);
2639                        unlock_page(page);
2640                        ntfs_unmap_page(page);
2641                        goto undo_mftbmp_alloc;
2642                }
2643                vi->i_ino = bit;
2644                /*
2645                 * This is for checking whether an inode has changed w.r.t. a
2646                 * file so that the file can be updated if necessary (compare
2647                 * with f_version).
2648                 */
2649                vi->i_version = 1;
2650
2651                /* The owner and group come from the ntfs volume. */
2652                vi->i_uid = vol->uid;
2653                vi->i_gid = vol->gid;
2654
2655                /* Initialize the ntfs specific part of @vi. */
2656                ntfs_init_big_inode(vi);
2657                ni = NTFS_I(vi);
2658                /*
2659                 * Set the appropriate mode, attribute type, and name.  For
2660                 * directories, also setup the index values to the defaults.
2661                 */
2662                if (S_ISDIR(mode)) {
2663                        vi->i_mode = S_IFDIR | S_IRWXUGO;
2664                        vi->i_mode &= ~vol->dmask;
2665
2666                        NInoSetMstProtected(ni);
2667                        ni->type = AT_INDEX_ALLOCATION;
2668                        ni->name = I30;
2669                        ni->name_len = 4;
2670
2671                        ni->itype.index.block_size = 4096;
2672                        ni->itype.index.block_size_bits = ntfs_ffs(4096) - 1;
2673                        ni->itype.index.collation_rule = COLLATION_FILE_NAME;
2674                        if (vol->cluster_size <= ni->itype.index.block_size) {
2675                                ni->itype.index.vcn_size = vol->cluster_size;
2676                                ni->itype.index.vcn_size_bits =
2677                                                vol->cluster_size_bits;
2678                        } else {
2679                                ni->itype.index.vcn_size = vol->sector_size;
2680                                ni->itype.index.vcn_size_bits =
2681                                                vol->sector_size_bits;
2682                        }
2683                } else {
2684                        vi->i_mode = S_IFREG | S_IRWXUGO;
2685                        vi->i_mode &= ~vol->fmask;
2686
2687                        ni->type = AT_DATA;
2688                        ni->name = NULL;
2689                        ni->name_len = 0;
2690                }
2691                if (IS_RDONLY(vi))
2692                        vi->i_mode &= ~S_IWUGO;
2693
2694                /* Set the inode times to the current time. */
2695                vi->i_atime = vi->i_mtime = vi->i_ctime =
2696                        current_time(vi);
2697                /*
2698                 * Set the file size to 0, the ntfs inode sizes are set to 0 by
2699                 * the call to ntfs_init_big_inode() below.
2700                 */
2701                vi->i_size = 0;
2702                vi->i_blocks = 0;
2703
2704                /* Set the sequence number. */
2705                vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
2706                /*
2707                 * Manually map, pin, and lock the mft record as we already
2708                 * have its page mapped and it is very easy to do.
2709                 */
2710                atomic_inc(&ni->count);
2711                mutex_lock(&ni->mrec_lock);
2712                ni->page = page;
2713                ni->page_ofs = ofs;
2714                /*
2715                 * Make sure the allocated mft record is written out to disk.
2716                 * NOTE: We do not set the ntfs inode dirty because this would
2717                 * fail in ntfs_write_inode() because the inode does not have a
2718                 * standard information attribute yet.  Also, there is no need
2719                 * to set the inode dirty because the caller is going to do
2720                 * that anyway after finishing with the new mft record (e.g. at
2721                 * a minimum some new attributes will be added to the mft
2722                 * record.
2723                 */
2724                mark_ntfs_record_dirty(page, ofs);
2725                unlock_page(page);
2726
2727                /* Add the inode to the inode hash for the superblock. */
2728                insert_inode_hash(vi);
2729
2730                /* Update the default mft allocation position. */
2731                vol->mft_data_pos = bit + 1;
2732        }
2733        /*
2734         * Return the opened, allocated inode of the allocated mft record as
2735         * well as the mapped, pinned, and locked mft record.
2736         */
2737        ntfs_debug("Returning opened, allocated %sinode 0x%llx.",
2738                        base_ni ? "extent " : "", (long long)bit);
2739        *mrec = m;
2740        return ni;
2741undo_data_init:
2742        write_lock_irqsave(&mft_ni->size_lock, flags);
2743        mft_ni->initialized_size = old_data_initialized;
2744        i_size_write(vol->mft_ino, old_data_size);
2745        write_unlock_irqrestore(&mft_ni->size_lock, flags);
2746        goto undo_mftbmp_alloc_nolock;
2747undo_mftbmp_alloc:
2748        down_write(&vol->mftbmp_lock);
2749undo_mftbmp_alloc_nolock:
2750        if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) {
2751                ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2752                NVolSetErrors(vol);
2753        }
2754        up_write(&vol->mftbmp_lock);
2755err_out:
2756        return ERR_PTR(err);
2757max_err_out:
2758        ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum "
2759                        "number of inodes (2^32) has already been reached.");
2760        up_write(&vol->mftbmp_lock);
2761        return ERR_PTR(-ENOSPC);
2762}
2763
2764/**
2765 * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume
2766 * @ni:         ntfs inode of the mapped extent mft record to free
2767 * @m:          mapped extent mft record of the ntfs inode @ni
2768 *
2769 * Free the mapped extent mft record @m of the extent ntfs inode @ni.
2770 *
2771 * Note that this function unmaps the mft record and closes and destroys @ni
2772 * internally and hence you cannot use either @ni nor @m any more after this
2773 * function returns success.
2774 *
2775 * On success return 0 and on error return -errno.  @ni and @m are still valid
2776 * in this case and have not been freed.
2777 *
2778 * For some errors an error message is displayed and the success code 0 is
2779 * returned and the volume is then left dirty on umount.  This makes sense in
2780 * case we could not rollback the changes that were already done since the
2781 * caller no longer wants to reference this mft record so it does not matter to
2782 * the caller if something is wrong with it as long as it is properly detached
2783 * from the base inode.
2784 */
2785int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m)
2786{
2787        unsigned long mft_no = ni->mft_no;
2788        ntfs_volume *vol = ni->vol;
2789        ntfs_inode *base_ni;
2790        ntfs_inode **extent_nis;
2791        int i, err;
2792        le16 old_seq_no;
2793        u16 seq_no;
2794        
2795        BUG_ON(NInoAttr(ni));
2796        BUG_ON(ni->nr_extents != -1);
2797
2798        mutex_lock(&ni->extent_lock);
2799        base_ni = ni->ext.base_ntfs_ino;
2800        mutex_unlock(&ni->extent_lock);
2801
2802        BUG_ON(base_ni->nr_extents <= 0);
2803
2804        ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n",
2805                        mft_no, base_ni->mft_no);
2806
2807        mutex_lock(&base_ni->extent_lock);
2808
2809        /* Make sure we are holding the only reference to the extent inode. */
2810        if (atomic_read(&ni->count) > 2) {
2811                ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, "
2812                                "not freeing.", base_ni->mft_no);
2813                mutex_unlock(&base_ni->extent_lock);
2814                return -EBUSY;
2815        }
2816
2817        /* Dissociate the ntfs inode from the base inode. */
2818        extent_nis = base_ni->ext.extent_ntfs_inos;
2819        err = -ENOENT;
2820        for (i = 0; i < base_ni->nr_extents; i++) {
2821                if (ni != extent_nis[i])
2822                        continue;
2823                extent_nis += i;
2824                base_ni->nr_extents--;
2825                memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) *
2826                                sizeof(ntfs_inode*));
2827                err = 0;
2828                break;
2829        }
2830
2831        mutex_unlock(&base_ni->extent_lock);
2832
2833        if (unlikely(err)) {
2834                ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to "
2835                                "its base inode 0x%lx.", mft_no,
2836                                base_ni->mft_no);
2837                BUG();
2838        }
2839
2840        /*
2841         * The extent inode is no longer attached to the base inode so no one
2842         * can get a reference to it any more.
2843         */
2844
2845        /* Mark the mft record as not in use. */
2846        m->flags &= ~MFT_RECORD_IN_USE;
2847
2848        /* Increment the sequence number, skipping zero, if it is not zero. */
2849        old_seq_no = m->sequence_number;
2850        seq_no = le16_to_cpu(old_seq_no);
2851        if (seq_no == 0xffff)
2852                seq_no = 1;
2853        else if (seq_no)
2854                seq_no++;
2855        m->sequence_number = cpu_to_le16(seq_no);
2856
2857        /*
2858         * Set the ntfs inode dirty and write it out.  We do not need to worry
2859         * about the base inode here since whatever caused the extent mft
2860         * record to be freed is guaranteed to do it already.
2861         */
2862        NInoSetDirty(ni);
2863        err = write_mft_record(ni, m, 0);
2864        if (unlikely(err)) {
2865                ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not "
2866                                "freeing.", mft_no);
2867                goto rollback;
2868        }
2869rollback_error:
2870        /* Unmap and throw away the now freed extent inode. */
2871        unmap_extent_mft_record(ni);
2872        ntfs_clear_extent_inode(ni);
2873
2874        /* Clear the bit in the $MFT/$BITMAP corresponding to this record. */
2875        down_write(&vol->mftbmp_lock);
2876        err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no);
2877        up_write(&vol->mftbmp_lock);
2878        if (unlikely(err)) {
2879                /*
2880                 * The extent inode is gone but we failed to deallocate it in
2881                 * the mft bitmap.  Just emit a warning and leave the volume
2882                 * dirty on umount.
2883                 */
2884                ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2885                NVolSetErrors(vol);
2886        }
2887        return 0;
2888rollback:
2889        /* Rollback what we did... */
2890        mutex_lock(&base_ni->extent_lock);
2891        extent_nis = base_ni->ext.extent_ntfs_inos;
2892        if (!(base_ni->nr_extents & 3)) {
2893                int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*);
2894
2895                extent_nis = kmalloc(new_size, GFP_NOFS);
2896                if (unlikely(!extent_nis)) {
2897                        ntfs_error(vol->sb, "Failed to allocate internal "
2898                                        "buffer during rollback.%s", es);
2899                        mutex_unlock(&base_ni->extent_lock);
2900                        NVolSetErrors(vol);
2901                        goto rollback_error;
2902                }
2903                if (base_ni->nr_extents) {
2904                        BUG_ON(!base_ni->ext.extent_ntfs_inos);
2905                        memcpy(extent_nis, base_ni->ext.extent_ntfs_inos,
2906                                        new_size - 4 * sizeof(ntfs_inode*));
2907                        kfree(base_ni->ext.extent_ntfs_inos);
2908                }
2909                base_ni->ext.extent_ntfs_inos = extent_nis;
2910        }
2911        m->flags |= MFT_RECORD_IN_USE;
2912        m->sequence_number = old_seq_no;
2913        extent_nis[base_ni->nr_extents++] = ni;
2914        mutex_unlock(&base_ni->extent_lock);
2915        mark_mft_record_dirty(ni);
2916        return err;
2917}
2918#endif /* NTFS_RW */
2919