~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ntfs/mft.c

Version: ~ [ linux-5.8-rc5 ] ~ [ linux-5.7.8 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.51 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.132 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.188 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.230 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.230 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /**
  2  * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project.
  3  *
  4  * Copyright (c) 2001-2006 Anton Altaparmakov
  5  * Copyright (c) 2002 Richard Russon
  6  *
  7  * This program/include file is free software; you can redistribute it and/or
  8  * modify it under the terms of the GNU General Public License as published
  9  * by the Free Software Foundation; either version 2 of the License, or
 10  * (at your option) any later version.
 11  *
 12  * This program/include file is distributed in the hope that it will be
 13  * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
 14  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 15  * GNU General Public License for more details.
 16  *
 17  * You should have received a copy of the GNU General Public License
 18  * along with this program (in the main directory of the Linux-NTFS
 19  * distribution in the file COPYING); if not, write to the Free Software
 20  * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 21  */
 22 
 23 #include <linux/buffer_head.h>
 24 #include <linux/swap.h>
 25 
 26 #include "attrib.h"
 27 #include "aops.h"
 28 #include "bitmap.h"
 29 #include "debug.h"
 30 #include "dir.h"
 31 #include "lcnalloc.h"
 32 #include "malloc.h"
 33 #include "mft.h"
 34 #include "ntfs.h"
 35 
 36 /**
 37  * map_mft_record_page - map the page in which a specific mft record resides
 38  * @ni:         ntfs inode whose mft record page to map
 39  *
 40  * This maps the page in which the mft record of the ntfs inode @ni is situated
 41  * and returns a pointer to the mft record within the mapped page.
 42  *
 43  * Return value needs to be checked with IS_ERR() and if that is true PTR_ERR()
 44  * contains the negative error code returned.
 45  */
 46 static inline MFT_RECORD *map_mft_record_page(ntfs_inode *ni)
 47 {
 48         loff_t i_size;
 49         ntfs_volume *vol = ni->vol;
 50         struct inode *mft_vi = vol->mft_ino;
 51         struct page *page;
 52         unsigned long index, end_index;
 53         unsigned ofs;
 54 
 55         BUG_ON(ni->page);
 56         /*
 57          * The index into the page cache and the offset within the page cache
 58          * page of the wanted mft record. FIXME: We need to check for
 59          * overflowing the unsigned long, but I don't think we would ever get
 60          * here if the volume was that big...
 61          */
 62         index = (u64)ni->mft_no << vol->mft_record_size_bits >>
 63                         PAGE_CACHE_SHIFT;
 64         ofs = (ni->mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
 65 
 66         i_size = i_size_read(mft_vi);
 67         /* The maximum valid index into the page cache for $MFT's data. */
 68         end_index = i_size >> PAGE_CACHE_SHIFT;
 69 
 70         /* If the wanted index is out of bounds the mft record doesn't exist. */
 71         if (unlikely(index >= end_index)) {
 72                 if (index > end_index || (i_size & ~PAGE_CACHE_MASK) < ofs +
 73                                 vol->mft_record_size) {
 74                         page = ERR_PTR(-ENOENT);
 75                         ntfs_error(vol->sb, "Attemt to read mft record 0x%lx, "
 76                                         "which is beyond the end of the mft.  "
 77                                         "This is probably a bug in the ntfs "
 78                                         "driver.", ni->mft_no);
 79                         goto err_out;
 80                 }
 81         }
 82         /* Read, map, and pin the page. */
 83         page = ntfs_map_page(mft_vi->i_mapping, index);
 84         if (likely(!IS_ERR(page))) {
 85                 /* Catch multi sector transfer fixup errors. */
 86                 if (likely(ntfs_is_mft_recordp((le32*)(page_address(page) +
 87                                 ofs)))) {
 88                         ni->page = page;
 89                         ni->page_ofs = ofs;
 90                         return page_address(page) + ofs;
 91                 }
 92                 ntfs_error(vol->sb, "Mft record 0x%lx is corrupt.  "
 93                                 "Run chkdsk.", ni->mft_no);
 94                 ntfs_unmap_page(page);
 95                 page = ERR_PTR(-EIO);
 96                 NVolSetErrors(vol);
 97         }
 98 err_out:
 99         ni->page = NULL;
100         ni->page_ofs = 0;
101         return (void*)page;
102 }
103 
104 /**
105  * map_mft_record - map, pin and lock an mft record
106  * @ni:         ntfs inode whose MFT record to map
107  *
108  * First, take the mrec_lock mutex.  We might now be sleeping, while waiting
109  * for the mutex if it was already locked by someone else.
110  *
111  * The page of the record is mapped using map_mft_record_page() before being
112  * returned to the caller.
113  *
114  * This in turn uses ntfs_map_page() to get the page containing the wanted mft
115  * record (it in turn calls read_cache_page() which reads it in from disk if
116  * necessary, increments the use count on the page so that it cannot disappear
117  * under us and returns a reference to the page cache page).
118  *
119  * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it
120  * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed
121  * and the post-read mst fixups on each mft record in the page have been
122  * performed, the page gets PG_uptodate set and PG_locked cleared (this is done
123  * in our asynchronous I/O completion handler end_buffer_read_mft_async()).
124  * ntfs_map_page() waits for PG_locked to become clear and checks if
125  * PG_uptodate is set and returns an error code if not. This provides
126  * sufficient protection against races when reading/using the page.
127  *
128  * However there is the write mapping to think about. Doing the above described
129  * checking here will be fine, because when initiating the write we will set
130  * PG_locked and clear PG_uptodate making sure nobody is touching the page
131  * contents. Doing the locking this way means that the commit to disk code in
132  * the page cache code paths is automatically sufficiently locked with us as
133  * we will not touch a page that has been locked or is not uptodate. The only
134  * locking problem then is them locking the page while we are accessing it.
135  *
136  * So that code will end up having to own the mrec_lock of all mft
137  * records/inodes present in the page before I/O can proceed. In that case we
138  * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be
139  * accessing anything without owning the mrec_lock mutex.  But we do need to
140  * use them because of the read_cache_page() invocation and the code becomes so
141  * much simpler this way that it is well worth it.
142  *
143  * The mft record is now ours and we return a pointer to it. You need to check
144  * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return
145  * the error code.
146  *
147  * NOTE: Caller is responsible for setting the mft record dirty before calling
148  * unmap_mft_record(). This is obviously only necessary if the caller really
149  * modified the mft record...
150  * Q: Do we want to recycle one of the VFS inode state bits instead?
151  * A: No, the inode ones mean we want to change the mft record, not we want to
152  * write it out.
153  */
154 MFT_RECORD *map_mft_record(ntfs_inode *ni)
155 {
156         MFT_RECORD *m;
157 
158         ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
159 
160         /* Make sure the ntfs inode doesn't go away. */
161         atomic_inc(&ni->count);
162 
163         /* Serialize access to this mft record. */
164         mutex_lock(&ni->mrec_lock);
165 
166         m = map_mft_record_page(ni);
167         if (likely(!IS_ERR(m)))
168                 return m;
169 
170         mutex_unlock(&ni->mrec_lock);
171         atomic_dec(&ni->count);
172         ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m));
173         return m;
174 }
175 
176 /**
177  * unmap_mft_record_page - unmap the page in which a specific mft record resides
178  * @ni:         ntfs inode whose mft record page to unmap
179  *
180  * This unmaps the page in which the mft record of the ntfs inode @ni is
181  * situated and returns. This is a NOOP if highmem is not configured.
182  *
183  * The unmap happens via ntfs_unmap_page() which in turn decrements the use
184  * count on the page thus releasing it from the pinned state.
185  *
186  * We do not actually unmap the page from memory of course, as that will be
187  * done by the page cache code itself when memory pressure increases or
188  * whatever.
189  */
190 static inline void unmap_mft_record_page(ntfs_inode *ni)
191 {
192         BUG_ON(!ni->page);
193 
194         // TODO: If dirty, blah...
195         ntfs_unmap_page(ni->page);
196         ni->page = NULL;
197         ni->page_ofs = 0;
198         return;
199 }
200 
201 /**
202  * unmap_mft_record - release a mapped mft record
203  * @ni:         ntfs inode whose MFT record to unmap
204  *
205  * We release the page mapping and the mrec_lock mutex which unmaps the mft
206  * record and releases it for others to get hold of. We also release the ntfs
207  * inode by decrementing the ntfs inode reference count.
208  *
209  * NOTE: If caller has modified the mft record, it is imperative to set the mft
210  * record dirty BEFORE calling unmap_mft_record().
211  */
212 void unmap_mft_record(ntfs_inode *ni)
213 {
214         struct page *page = ni->page;
215 
216         BUG_ON(!page);
217 
218         ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
219 
220         unmap_mft_record_page(ni);
221         mutex_unlock(&ni->mrec_lock);
222         atomic_dec(&ni->count);
223         /*
224          * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to
225          * ntfs_clear_extent_inode() in the extent inode case, and to the
226          * caller in the non-extent, yet pure ntfs inode case, to do the actual
227          * tear down of all structures and freeing of all allocated memory.
228          */
229         return;
230 }
231 
232 /**
233  * map_extent_mft_record - load an extent inode and attach it to its base
234  * @base_ni:    base ntfs inode
235  * @mref:       mft reference of the extent inode to load
236  * @ntfs_ino:   on successful return, pointer to the ntfs_inode structure
237  *
238  * Load the extent mft record @mref and attach it to its base inode @base_ni.
239  * Return the mapped extent mft record if IS_ERR(result) is false.  Otherwise
240  * PTR_ERR(result) gives the negative error code.
241  *
242  * On successful return, @ntfs_ino contains a pointer to the ntfs_inode
243  * structure of the mapped extent inode.
244  */
245 MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref,
246                 ntfs_inode **ntfs_ino)
247 {
248         MFT_RECORD *m;
249         ntfs_inode *ni = NULL;
250         ntfs_inode **extent_nis = NULL;
251         int i;
252         unsigned long mft_no = MREF(mref);
253         u16 seq_no = MSEQNO(mref);
254         bool destroy_ni = false;
255 
256         ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).",
257                         mft_no, base_ni->mft_no);
258         /* Make sure the base ntfs inode doesn't go away. */
259         atomic_inc(&base_ni->count);
260         /*
261          * Check if this extent inode has already been added to the base inode,
262          * in which case just return it. If not found, add it to the base
263          * inode before returning it.
264          */
265         mutex_lock(&base_ni->extent_lock);
266         if (base_ni->nr_extents > 0) {
267                 extent_nis = base_ni->ext.extent_ntfs_inos;
268                 for (i = 0; i < base_ni->nr_extents; i++) {
269                         if (mft_no != extent_nis[i]->mft_no)
270                                 continue;
271                         ni = extent_nis[i];
272                         /* Make sure the ntfs inode doesn't go away. */
273                         atomic_inc(&ni->count);
274                         break;
275                 }
276         }
277         if (likely(ni != NULL)) {
278                 mutex_unlock(&base_ni->extent_lock);
279                 atomic_dec(&base_ni->count);
280                 /* We found the record; just have to map and return it. */
281                 m = map_mft_record(ni);
282                 /* map_mft_record() has incremented this on success. */
283                 atomic_dec(&ni->count);
284                 if (likely(!IS_ERR(m))) {
285                         /* Verify the sequence number. */
286                         if (likely(le16_to_cpu(m->sequence_number) == seq_no)) {
287                                 ntfs_debug("Done 1.");
288                                 *ntfs_ino = ni;
289                                 return m;
290                         }
291                         unmap_mft_record(ni);
292                         ntfs_error(base_ni->vol->sb, "Found stale extent mft "
293                                         "reference! Corrupt filesystem. "
294                                         "Run chkdsk.");
295                         return ERR_PTR(-EIO);
296                 }
297 map_err_out:
298                 ntfs_error(base_ni->vol->sb, "Failed to map extent "
299                                 "mft record, error code %ld.", -PTR_ERR(m));
300                 return m;
301         }
302         /* Record wasn't there. Get a new ntfs inode and initialize it. */
303         ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no);
304         if (unlikely(!ni)) {
305                 mutex_unlock(&base_ni->extent_lock);
306                 atomic_dec(&base_ni->count);
307                 return ERR_PTR(-ENOMEM);
308         }
309         ni->vol = base_ni->vol;
310         ni->seq_no = seq_no;
311         ni->nr_extents = -1;
312         ni->ext.base_ntfs_ino = base_ni;
313         /* Now map the record. */
314         m = map_mft_record(ni);
315         if (IS_ERR(m)) {
316                 mutex_unlock(&base_ni->extent_lock);
317                 atomic_dec(&base_ni->count);
318                 ntfs_clear_extent_inode(ni);
319                 goto map_err_out;
320         }
321         /* Verify the sequence number if it is present. */
322         if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) {
323                 ntfs_error(base_ni->vol->sb, "Found stale extent mft "
324                                 "reference! Corrupt filesystem. Run chkdsk.");
325                 destroy_ni = true;
326                 m = ERR_PTR(-EIO);
327                 goto unm_err_out;
328         }
329         /* Attach extent inode to base inode, reallocating memory if needed. */
330         if (!(base_ni->nr_extents & 3)) {
331                 ntfs_inode **tmp;
332                 int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *);
333 
334                 tmp = kmalloc(new_size, GFP_NOFS);
335                 if (unlikely(!tmp)) {
336                         ntfs_error(base_ni->vol->sb, "Failed to allocate "
337                                         "internal buffer.");
338                         destroy_ni = true;
339                         m = ERR_PTR(-ENOMEM);
340                         goto unm_err_out;
341                 }
342                 if (base_ni->nr_extents) {
343                         BUG_ON(!base_ni->ext.extent_ntfs_inos);
344                         memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size -
345                                         4 * sizeof(ntfs_inode *));
346                         kfree(base_ni->ext.extent_ntfs_inos);
347                 }
348                 base_ni->ext.extent_ntfs_inos = tmp;
349         }
350         base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni;
351         mutex_unlock(&base_ni->extent_lock);
352         atomic_dec(&base_ni->count);
353         ntfs_debug("Done 2.");
354         *ntfs_ino = ni;
355         return m;
356 unm_err_out:
357         unmap_mft_record(ni);
358         mutex_unlock(&base_ni->extent_lock);
359         atomic_dec(&base_ni->count);
360         /*
361          * If the extent inode was not attached to the base inode we need to
362          * release it or we will leak memory.
363          */
364         if (destroy_ni)
365                 ntfs_clear_extent_inode(ni);
366         return m;
367 }
368 
369 #ifdef NTFS_RW
370 
371 /**
372  * __mark_mft_record_dirty - set the mft record and the page containing it dirty
373  * @ni:         ntfs inode describing the mapped mft record
374  *
375  * Internal function.  Users should call mark_mft_record_dirty() instead.
376  *
377  * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni,
378  * as well as the page containing the mft record, dirty.  Also, mark the base
379  * vfs inode dirty.  This ensures that any changes to the mft record are
380  * written out to disk.
381  *
382  * NOTE:  We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES)
383  * on the base vfs inode, because even though file data may have been modified,
384  * it is dirty in the inode meta data rather than the data page cache of the
385  * inode, and thus there are no data pages that need writing out.  Therefore, a
386  * full mark_inode_dirty() is overkill.  A mark_inode_dirty_sync(), on the
387  * other hand, is not sufficient, because ->write_inode needs to be called even
388  * in case of fdatasync. This needs to happen or the file data would not
389  * necessarily hit the device synchronously, even though the vfs inode has the
390  * O_SYNC flag set.  Also, I_DIRTY_DATASYNC simply "feels" better than just
391  * I_DIRTY_SYNC, since the file data has not actually hit the block device yet,
392  * which is not what I_DIRTY_SYNC on its own would suggest.
393  */
394 void __mark_mft_record_dirty(ntfs_inode *ni)
395 {
396         ntfs_inode *base_ni;
397 
398         ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
399         BUG_ON(NInoAttr(ni));
400         mark_ntfs_record_dirty(ni->page, ni->page_ofs);
401         /* Determine the base vfs inode and mark it dirty, too. */
402         mutex_lock(&ni->extent_lock);
403         if (likely(ni->nr_extents >= 0))
404                 base_ni = ni;
405         else
406                 base_ni = ni->ext.base_ntfs_ino;
407         mutex_unlock(&ni->extent_lock);
408         __mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC);
409 }
410 
411 static const char *ntfs_please_email = "Please email "
412                 "linux-ntfs-dev@lists.sourceforge.net and say that you saw "
413                 "this message.  Thank you.";
414 
415 /**
416  * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror
417  * @vol:        ntfs volume on which the mft record to synchronize resides
418  * @mft_no:     mft record number of mft record to synchronize
419  * @m:          mapped, mst protected (extent) mft record to synchronize
420  *
421  * Write the mapped, mst protected (extent) mft record @m with mft record
422  * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol,
423  * bypassing the page cache and the $MFTMirr inode itself.
424  *
425  * This function is only for use at umount time when the mft mirror inode has
426  * already been disposed off.  We BUG() if we are called while the mft mirror
427  * inode is still attached to the volume.
428  *
429  * On success return 0.  On error return -errno.
430  *
431  * NOTE:  This function is not implemented yet as I am not convinced it can
432  * actually be triggered considering the sequence of commits we do in super.c::
433  * ntfs_put_super().  But just in case we provide this place holder as the
434  * alternative would be either to BUG() or to get a NULL pointer dereference
435  * and Oops.
436  */
437 static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol,
438                 const unsigned long mft_no, MFT_RECORD *m)
439 {
440         BUG_ON(vol->mftmirr_ino);
441         ntfs_error(vol->sb, "Umount time mft mirror syncing is not "
442                         "implemented yet.  %s", ntfs_please_email);
443         return -EOPNOTSUPP;
444 }
445 
446 /**
447  * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror
448  * @vol:        ntfs volume on which the mft record to synchronize resides
449  * @mft_no:     mft record number of mft record to synchronize
450  * @m:          mapped, mst protected (extent) mft record to synchronize
451  * @sync:       if true, wait for i/o completion
452  *
453  * Write the mapped, mst protected (extent) mft record @m with mft record
454  * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol.
455  *
456  * On success return 0.  On error return -errno and set the volume errors flag
457  * in the ntfs volume @vol.
458  *
459  * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
460  *
461  * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
462  * schedule i/o via ->writepage or do it via kntfsd or whatever.
463  */
464 int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no,
465                 MFT_RECORD *m, int sync)
466 {
467         struct page *page;
468         unsigned int blocksize = vol->sb->s_blocksize;
469         int max_bhs = vol->mft_record_size / blocksize;
470         struct buffer_head *bhs[max_bhs];
471         struct buffer_head *bh, *head;
472         u8 *kmirr;
473         runlist_element *rl;
474         unsigned int block_start, block_end, m_start, m_end, page_ofs;
475         int i_bhs, nr_bhs, err = 0;
476         unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
477 
478         ntfs_debug("Entering for inode 0x%lx.", mft_no);
479         BUG_ON(!max_bhs);
480         if (unlikely(!vol->mftmirr_ino)) {
481                 /* This could happen during umount... */
482                 err = ntfs_sync_mft_mirror_umount(vol, mft_no, m);
483                 if (likely(!err))
484                         return err;
485                 goto err_out;
486         }
487         /* Get the page containing the mirror copy of the mft record @m. */
488         page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >>
489                         (PAGE_CACHE_SHIFT - vol->mft_record_size_bits));
490         if (IS_ERR(page)) {
491                 ntfs_error(vol->sb, "Failed to map mft mirror page.");
492                 err = PTR_ERR(page);
493                 goto err_out;
494         }
495         lock_page(page);
496         BUG_ON(!PageUptodate(page));
497         ClearPageUptodate(page);
498         /* Offset of the mft mirror record inside the page. */
499         page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
500         /* The address in the page of the mirror copy of the mft record @m. */
501         kmirr = page_address(page) + page_ofs;
502         /* Copy the mst protected mft record to the mirror. */
503         memcpy(kmirr, m, vol->mft_record_size);
504         /* Create uptodate buffers if not present. */
505         if (unlikely(!page_has_buffers(page))) {
506                 struct buffer_head *tail;
507 
508                 bh = head = alloc_page_buffers(page, blocksize, 1);
509                 do {
510                         set_buffer_uptodate(bh);
511                         tail = bh;
512                         bh = bh->b_this_page;
513                 } while (bh);
514                 tail->b_this_page = head;
515                 attach_page_buffers(page, head);
516         }
517         bh = head = page_buffers(page);
518         BUG_ON(!bh);
519         rl = NULL;
520         nr_bhs = 0;
521         block_start = 0;
522         m_start = kmirr - (u8*)page_address(page);
523         m_end = m_start + vol->mft_record_size;
524         do {
525                 block_end = block_start + blocksize;
526                 /* If the buffer is outside the mft record, skip it. */
527                 if (block_end <= m_start)
528                         continue;
529                 if (unlikely(block_start >= m_end))
530                         break;
531                 /* Need to map the buffer if it is not mapped already. */
532                 if (unlikely(!buffer_mapped(bh))) {
533                         VCN vcn;
534                         LCN lcn;
535                         unsigned int vcn_ofs;
536 
537                         bh->b_bdev = vol->sb->s_bdev;
538                         /* Obtain the vcn and offset of the current block. */
539                         vcn = ((VCN)mft_no << vol->mft_record_size_bits) +
540                                         (block_start - m_start);
541                         vcn_ofs = vcn & vol->cluster_size_mask;
542                         vcn >>= vol->cluster_size_bits;
543                         if (!rl) {
544                                 down_read(&NTFS_I(vol->mftmirr_ino)->
545                                                 runlist.lock);
546                                 rl = NTFS_I(vol->mftmirr_ino)->runlist.rl;
547                                 /*
548                                  * $MFTMirr always has the whole of its runlist
549                                  * in memory.
550                                  */
551                                 BUG_ON(!rl);
552                         }
553                         /* Seek to element containing target vcn. */
554                         while (rl->length && rl[1].vcn <= vcn)
555                                 rl++;
556                         lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
557                         /* For $MFTMirr, only lcn >= 0 is a successful remap. */
558                         if (likely(lcn >= 0)) {
559                                 /* Setup buffer head to correct block. */
560                                 bh->b_blocknr = ((lcn <<
561                                                 vol->cluster_size_bits) +
562                                                 vcn_ofs) >> blocksize_bits;
563                                 set_buffer_mapped(bh);
564                         } else {
565                                 bh->b_blocknr = -1;
566                                 ntfs_error(vol->sb, "Cannot write mft mirror "
567                                                 "record 0x%lx because its "
568                                                 "location on disk could not "
569                                                 "be determined (error code "
570                                                 "%lli).", mft_no,
571                                                 (long long)lcn);
572                                 err = -EIO;
573                         }
574                 }
575                 BUG_ON(!buffer_uptodate(bh));
576                 BUG_ON(!nr_bhs && (m_start != block_start));
577                 BUG_ON(nr_bhs >= max_bhs);
578                 bhs[nr_bhs++] = bh;
579                 BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
580         } while (block_start = block_end, (bh = bh->b_this_page) != head);
581         if (unlikely(rl))
582                 up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock);
583         if (likely(!err)) {
584                 /* Lock buffers and start synchronous write i/o on them. */
585                 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
586                         struct buffer_head *tbh = bhs[i_bhs];
587 
588                         if (!trylock_buffer(tbh))
589                                 BUG();
590                         BUG_ON(!buffer_uptodate(tbh));
591                         clear_buffer_dirty(tbh);
592                         get_bh(tbh);
593                         tbh->b_end_io = end_buffer_write_sync;
594                         submit_bh(WRITE, tbh);
595                 }
596                 /* Wait on i/o completion of buffers. */
597                 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
598                         struct buffer_head *tbh = bhs[i_bhs];
599 
600                         wait_on_buffer(tbh);
601                         if (unlikely(!buffer_uptodate(tbh))) {
602                                 err = -EIO;
603                                 /*
604                                  * Set the buffer uptodate so the page and
605                                  * buffer states do not become out of sync.
606                                  */
607                                 set_buffer_uptodate(tbh);
608                         }
609                 }
610         } else /* if (unlikely(err)) */ {
611                 /* Clean the buffers. */
612                 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
613                         clear_buffer_dirty(bhs[i_bhs]);
614         }
615         /* Current state: all buffers are clean, unlocked, and uptodate. */
616         /* Remove the mst protection fixups again. */
617         post_write_mst_fixup((NTFS_RECORD*)kmirr);
618         flush_dcache_page(page);
619         SetPageUptodate(page);
620         unlock_page(page);
621         ntfs_unmap_page(page);
622         if (likely(!err)) {
623                 ntfs_debug("Done.");
624         } else {
625                 ntfs_error(vol->sb, "I/O error while writing mft mirror "
626                                 "record 0x%lx!", mft_no);
627 err_out:
628                 ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error "
629                                 "code %i).  Volume will be left marked dirty "
630                                 "on umount.  Run ntfsfix on the partition "
631                                 "after umounting to correct this.", -err);
632                 NVolSetErrors(vol);
633         }
634         return err;
635 }
636 
637 /**
638  * write_mft_record_nolock - write out a mapped (extent) mft record
639  * @ni:         ntfs inode describing the mapped (extent) mft record
640  * @m:          mapped (extent) mft record to write
641  * @sync:       if true, wait for i/o completion
642  *
643  * Write the mapped (extent) mft record @m described by the (regular or extent)
644  * ntfs inode @ni to backing store.  If the mft record @m has a counterpart in
645  * the mft mirror, that is also updated.
646  *
647  * We only write the mft record if the ntfs inode @ni is dirty and the first
648  * buffer belonging to its mft record is dirty, too.  We ignore the dirty state
649  * of subsequent buffers because we could have raced with
650  * fs/ntfs/aops.c::mark_ntfs_record_dirty().
651  *
652  * On success, clean the mft record and return 0.  On error, leave the mft
653  * record dirty and return -errno.
654  *
655  * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
656  * However, if the mft record has a counterpart in the mft mirror and @sync is
657  * true, we write the mft record, wait for i/o completion, and only then write
658  * the mft mirror copy.  This ensures that if the system crashes either the mft
659  * or the mft mirror will contain a self-consistent mft record @m.  If @sync is
660  * false on the other hand, we start i/o on both and then wait for completion
661  * on them.  This provides a speedup but no longer guarantees that you will end
662  * up with a self-consistent mft record in the case of a crash but if you asked
663  * for asynchronous writing you probably do not care about that anyway.
664  *
665  * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
666  * schedule i/o via ->writepage or do it via kntfsd or whatever.
667  */
668 int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync)
669 {
670         ntfs_volume *vol = ni->vol;
671         struct page *page = ni->page;
672         unsigned int blocksize = vol->sb->s_blocksize;
673         unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
674         int max_bhs = vol->mft_record_size / blocksize;
675         struct buffer_head *bhs[max_bhs];
676         struct buffer_head *bh, *head;
677         runlist_element *rl;
678         unsigned int block_start, block_end, m_start, m_end;
679         int i_bhs, nr_bhs, err = 0;
680 
681         ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
682         BUG_ON(NInoAttr(ni));
683         BUG_ON(!max_bhs);
684         BUG_ON(!PageLocked(page));
685         /*
686          * If the ntfs_inode is clean no need to do anything.  If it is dirty,
687          * mark it as clean now so that it can be redirtied later on if needed.
688          * There is no danger of races since the caller is holding the locks
689          * for the mft record @m and the page it is in.
690          */
691         if (!NInoTestClearDirty(ni))
692                 goto done;
693         bh = head = page_buffers(page);
694         BUG_ON(!bh);
695         rl = NULL;
696         nr_bhs = 0;
697         block_start = 0;
698         m_start = ni->page_ofs;
699         m_end = m_start + vol->mft_record_size;
700         do {
701                 block_end = block_start + blocksize;
702                 /* If the buffer is outside the mft record, skip it. */
703                 if (block_end <= m_start)
704                         continue;
705                 if (unlikely(block_start >= m_end))
706                         break;
707                 /*
708                  * If this block is not the first one in the record, we ignore
709                  * the buffer's dirty state because we could have raced with a
710                  * parallel mark_ntfs_record_dirty().
711                  */
712                 if (block_start == m_start) {
713                         /* This block is the first one in the record. */
714                         if (!buffer_dirty(bh)) {
715                                 BUG_ON(nr_bhs);
716                                 /* Clean records are not written out. */
717                                 break;
718                         }
719                 }
720                 /* Need to map the buffer if it is not mapped already. */
721                 if (unlikely(!buffer_mapped(bh))) {
722                         VCN vcn;
723                         LCN lcn;
724                         unsigned int vcn_ofs;
725 
726                         bh->b_bdev = vol->sb->s_bdev;
727                         /* Obtain the vcn and offset of the current block. */
728                         vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) +
729                                         (block_start - m_start);
730                         vcn_ofs = vcn & vol->cluster_size_mask;
731                         vcn >>= vol->cluster_size_bits;
732                         if (!rl) {
733                                 down_read(&NTFS_I(vol->mft_ino)->runlist.lock);
734                                 rl = NTFS_I(vol->mft_ino)->runlist.rl;
735                                 BUG_ON(!rl);
736                         }
737                         /* Seek to element containing target vcn. */
738                         while (rl->length && rl[1].vcn <= vcn)
739                                 rl++;
740                         lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
741                         /* For $MFT, only lcn >= 0 is a successful remap. */
742                         if (likely(lcn >= 0)) {
743                                 /* Setup buffer head to correct block. */
744                                 bh->b_blocknr = ((lcn <<
745                                                 vol->cluster_size_bits) +
746                                                 vcn_ofs) >> blocksize_bits;
747                                 set_buffer_mapped(bh);
748                         } else {
749                                 bh->b_blocknr = -1;
750                                 ntfs_error(vol->sb, "Cannot write mft record "
751                                                 "0x%lx because its location "
752                                                 "on disk could not be "
753                                                 "determined (error code %lli).",
754                                                 ni->mft_no, (long long)lcn);
755                                 err = -EIO;
756                         }
757                 }
758                 BUG_ON(!buffer_uptodate(bh));
759                 BUG_ON(!nr_bhs && (m_start != block_start));
760                 BUG_ON(nr_bhs >= max_bhs);
761                 bhs[nr_bhs++] = bh;
762                 BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
763         } while (block_start = block_end, (bh = bh->b_this_page) != head);
764         if (unlikely(rl))
765                 up_read(&NTFS_I(vol->mft_ino)->runlist.lock);
766         if (!nr_bhs)
767                 goto done;
768         if (unlikely(err))
769                 goto cleanup_out;
770         /* Apply the mst protection fixups. */
771         err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size);
772         if (err) {
773                 ntfs_error(vol->sb, "Failed to apply mst fixups!");
774                 goto cleanup_out;
775         }
776         flush_dcache_mft_record_page(ni);
777         /* Lock buffers and start synchronous write i/o on them. */
778         for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
779                 struct buffer_head *tbh = bhs[i_bhs];
780 
781                 if (!trylock_buffer(tbh))
782                         BUG();
783                 BUG_ON(!buffer_uptodate(tbh));
784                 clear_buffer_dirty(tbh);
785                 get_bh(tbh);
786                 tbh->b_end_io = end_buffer_write_sync;
787                 submit_bh(WRITE, tbh);
788         }
789         /* Synchronize the mft mirror now if not @sync. */
790         if (!sync && ni->mft_no < vol->mftmirr_size)
791                 ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
792         /* Wait on i/o completion of buffers. */
793         for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
794                 struct buffer_head *tbh = bhs[i_bhs];
795 
796                 wait_on_buffer(tbh);
797                 if (unlikely(!buffer_uptodate(tbh))) {
798                         err = -EIO;
799                         /*
800                          * Set the buffer uptodate so the page and buffer
801                          * states do not become out of sync.
802                          */
803                         if (PageUptodate(page))
804                                 set_buffer_uptodate(tbh);
805                 }
806         }
807         /* If @sync, now synchronize the mft mirror. */
808         if (sync && ni->mft_no < vol->mftmirr_size)
809                 ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
810         /* Remove the mst protection fixups again. */
811         post_write_mst_fixup((NTFS_RECORD*)m);
812         flush_dcache_mft_record_page(ni);
813         if (unlikely(err)) {
814                 /* I/O error during writing.  This is really bad! */
815                 ntfs_error(vol->sb, "I/O error while writing mft record "
816                                 "0x%lx!  Marking base inode as bad.  You "
817                                 "should unmount the volume and run chkdsk.",
818                                 ni->mft_no);
819                 goto err_out;
820         }
821 done:
822         ntfs_debug("Done.");
823         return 0;
824 cleanup_out:
825         /* Clean the buffers. */
826         for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
827                 clear_buffer_dirty(bhs[i_bhs]);
828 err_out:
829         /*
830          * Current state: all buffers are clean, unlocked, and uptodate.
831          * The caller should mark the base inode as bad so that no more i/o
832          * happens.  ->clear_inode() will still be invoked so all extent inodes
833          * and other allocated memory will be freed.
834          */
835         if (err == -ENOMEM) {
836                 ntfs_error(vol->sb, "Not enough memory to write mft record.  "
837                                 "Redirtying so the write is retried later.");
838                 mark_mft_record_dirty(ni);
839                 err = 0;
840         } else
841                 NVolSetErrors(vol);
842         return err;
843 }
844 
845 /**
846  * ntfs_may_write_mft_record - check if an mft record may be written out
847  * @vol:        [IN]  ntfs volume on which the mft record to check resides
848  * @mft_no:     [IN]  mft record number of the mft record to check
849  * @m:          [IN]  mapped mft record to check
850  * @locked_ni:  [OUT] caller has to unlock this ntfs inode if one is returned
851  *
852  * Check if the mapped (base or extent) mft record @m with mft record number
853  * @mft_no belonging to the ntfs volume @vol may be written out.  If necessary
854  * and possible the ntfs inode of the mft record is locked and the base vfs
855  * inode is pinned.  The locked ntfs inode is then returned in @locked_ni.  The
856  * caller is responsible for unlocking the ntfs inode and unpinning the base
857  * vfs inode.
858  *
859  * Return 'true' if the mft record may be written out and 'false' if not.
860  *
861  * The caller has locked the page and cleared the uptodate flag on it which
862  * means that we can safely write out any dirty mft records that do not have
863  * their inodes in icache as determined by ilookup5() as anyone
864  * opening/creating such an inode would block when attempting to map the mft
865  * record in read_cache_page() until we are finished with the write out.
866  *
867  * Here is a description of the tests we perform:
868  *
869  * If the inode is found in icache we know the mft record must be a base mft
870  * record.  If it is dirty, we do not write it and return 'false' as the vfs
871  * inode write paths will result in the access times being updated which would
872  * cause the base mft record to be redirtied and written out again.  (We know
873  * the access time update will modify the base mft record because Windows
874  * chkdsk complains if the standard information attribute is not in the base
875  * mft record.)
876  *
877  * If the inode is in icache and not dirty, we attempt to lock the mft record
878  * and if we find the lock was already taken, it is not safe to write the mft
879  * record and we return 'false'.
880  *
881  * If we manage to obtain the lock we have exclusive access to the mft record,
882  * which also allows us safe writeout of the mft record.  We then set
883  * @locked_ni to the locked ntfs inode and return 'true'.
884  *
885  * Note we cannot just lock the mft record and sleep while waiting for the lock
886  * because this would deadlock due to lock reversal (normally the mft record is
887  * locked before the page is locked but we already have the page locked here
888  * when we try to lock the mft record).
889  *
890  * If the inode is not in icache we need to perform further checks.
891  *
892  * If the mft record is not a FILE record or it is a base mft record, we can
893  * safely write it and return 'true'.
894  *
895  * We now know the mft record is an extent mft record.  We check if the inode
896  * corresponding to its base mft record is in icache and obtain a reference to
897  * it if it is.  If it is not, we can safely write it and return 'true'.
898  *
899  * We now have the base inode for the extent mft record.  We check if it has an
900  * ntfs inode for the extent mft record attached and if not it is safe to write
901  * the extent mft record and we return 'true'.
902  *
903  * The ntfs inode for the extent mft record is attached to the base inode so we
904  * attempt to lock the extent mft record and if we find the lock was already
905  * taken, it is not safe to write the extent mft record and we return 'false'.
906  *
907  * If we manage to obtain the lock we have exclusive access to the extent mft
908  * record, which also allows us safe writeout of the extent mft record.  We
909  * set the ntfs inode of the extent mft record clean and then set @locked_ni to
910  * the now locked ntfs inode and return 'true'.
911  *
912  * Note, the reason for actually writing dirty mft records here and not just
913  * relying on the vfs inode dirty code paths is that we can have mft records
914  * modified without them ever having actual inodes in memory.  Also we can have
915  * dirty mft records with clean ntfs inodes in memory.  None of the described
916  * cases would result in the dirty mft records being written out if we only
917  * relied on the vfs inode dirty code paths.  And these cases can really occur
918  * during allocation of new mft records and in particular when the
919  * initialized_size of the $MFT/$DATA attribute is extended and the new space
920  * is initialized using ntfs_mft_record_format().  The clean inode can then
921  * appear if the mft record is reused for a new inode before it got written
922  * out.
923  */
924 bool ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
925                 const MFT_RECORD *m, ntfs_inode **locked_ni)
926 {
927         struct super_block *sb = vol->sb;
928         struct inode *mft_vi = vol->mft_ino;
929         struct inode *vi;
930         ntfs_inode *ni, *eni, **extent_nis;
931         int i;
932         ntfs_attr na;
933 
934         ntfs_debug("Entering for inode 0x%lx.", mft_no);
935         /*
936          * Normally we do not return a locked inode so set @locked_ni to NULL.
937          */
938         BUG_ON(!locked_ni);
939         *locked_ni = NULL;
940         /*
941          * Check if the inode corresponding to this mft record is in the VFS
942          * inode cache and obtain a reference to it if it is.
943          */
944         ntfs_debug("Looking for inode 0x%lx in icache.", mft_no);
945         na.mft_no = mft_no;
946         na.name = NULL;
947         na.name_len = 0;
948         na.type = AT_UNUSED;
949         /*
950          * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
951          * we get here for it rather often.
952          */
953         if (!mft_no) {
954                 /* Balance the below iput(). */
955                 vi = igrab(mft_vi);
956                 BUG_ON(vi != mft_vi);
957         } else {
958                 /*
959                  * Have to use ilookup5_nowait() since ilookup5() waits for the
960                  * inode lock which causes ntfs to deadlock when a concurrent
961                  * inode write via the inode dirty code paths and the page
962                  * dirty code path of the inode dirty code path when writing
963                  * $MFT occurs.
964                  */
965                 vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
966         }
967         if (vi) {
968                 ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
969                 /* The inode is in icache. */
970                 ni = NTFS_I(vi);
971                 /* Take a reference to the ntfs inode. */
972                 atomic_inc(&ni->count);
973                 /* If the inode is dirty, do not write this record. */
974                 if (NInoDirty(ni)) {
975                         ntfs_debug("Inode 0x%lx is dirty, do not write it.",
976                                         mft_no);
977                         atomic_dec(&ni->count);
978                         iput(vi);
979                         return false;
980                 }
981                 ntfs_debug("Inode 0x%lx is not dirty.", mft_no);
982                 /* The inode is not dirty, try to take the mft record lock. */
983                 if (unlikely(!mutex_trylock(&ni->mrec_lock))) {
984                         ntfs_debug("Mft record 0x%lx is already locked, do "
985                                         "not write it.", mft_no);
986                         atomic_dec(&ni->count);
987                         iput(vi);
988                         return false;
989                 }
990                 ntfs_debug("Managed to lock mft record 0x%lx, write it.",
991                                 mft_no);
992                 /*
993                  * The write has to occur while we hold the mft record lock so
994                  * return the locked ntfs inode.
995                  */
996                 *locked_ni = ni;
997                 return true;
998         }
999         ntfs_debug("Inode 0x%lx is not in icache.", mft_no);
1000         /* The inode is not in icache. */
1001         /* Write the record if it is not a mft record (type "FILE"). */
1002         if (!ntfs_is_mft_record(m->magic)) {
1003                 ntfs_debug("Mft record 0x%lx is not a FILE record, write it.",
1004                                 mft_no);
1005                 return true;
1006         }
1007         /* Write the mft record if it is a base inode. */
1008         if (!m->base_mft_record) {
1009                 ntfs_debug("Mft record 0x%lx is a base record, write it.",
1010                                 mft_no);
1011                 return true;
1012         }
1013         /*
1014          * This is an extent mft record.  Check if the inode corresponding to
1015          * its base mft record is in icache and obtain a reference to it if it
1016          * is.
1017          */
1018         na.mft_no = MREF_LE(m->base_mft_record);
1019         ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
1020                         "inode 0x%lx in icache.", mft_no, na.mft_no);
1021         if (!na.mft_no) {
1022                 /* Balance the below iput(). */
1023                 vi = igrab(mft_vi);
1024                 BUG_ON(vi != mft_vi);
1025         } else
1026                 vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
1027                                 &na);
1028         if (!vi) {
1029                 /*
1030                  * The base inode is not in icache, write this extent mft
1031                  * record.
1032                  */
1033                 ntfs_debug("Base inode 0x%lx is not in icache, write the "
1034                                 "extent record.", na.mft_no);
1035                 return true;
1036         }
1037         ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no);
1038         /*
1039          * The base inode is in icache.  Check if it has the extent inode
1040          * corresponding to this extent mft record attached.
1041          */
1042         ni = NTFS_I(vi);
1043         mutex_lock(&ni->extent_lock);
1044         if (ni->nr_extents <= 0) {
1045                 /*
1046                  * The base inode has no attached extent inodes, write this
1047                  * extent mft record.
1048                  */
1049                 mutex_unlock(&ni->extent_lock);
1050                 iput(vi);
1051                 ntfs_debug("Base inode 0x%lx has no attached extent inodes, "
1052                                 "write the extent record.", na.mft_no);
1053                 return true;
1054         }
1055         /* Iterate over the attached extent inodes. */
1056         extent_nis = ni->ext.extent_ntfs_inos;
1057         for (eni = NULL, i = 0; i < ni->nr_extents; ++i) {
1058                 if (mft_no == extent_nis[i]->mft_no) {
1059                         /*
1060                          * Found the extent inode corresponding to this extent
1061                          * mft record.
1062                          */
1063                         eni = extent_nis[i];
1064                         break;
1065                 }
1066         }
1067         /*
1068          * If the extent inode was not attached to the base inode, write this
1069          * extent mft record.
1070          */
1071         if (!eni) {
1072                 mutex_unlock(&ni->extent_lock);
1073                 iput(vi);
1074                 ntfs_debug("Extent inode 0x%lx is not attached to its base "
1075                                 "inode 0x%lx, write the extent record.",
1076                                 mft_no, na.mft_no);
1077                 return true;
1078         }
1079         ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.",
1080                         mft_no, na.mft_no);
1081         /* Take a reference to the extent ntfs inode. */
1082         atomic_inc(&eni->count);
1083         mutex_unlock(&ni->extent_lock);
1084         /*
1085          * Found the extent inode coresponding to this extent mft record.
1086          * Try to take the mft record lock.
1087          */
1088         if (unlikely(!mutex_trylock(&eni->mrec_lock))) {
1089                 atomic_dec(&eni->count);
1090                 iput(vi);
1091                 ntfs_debug("Extent mft record 0x%lx is already locked, do "
1092                                 "not write it.", mft_no);
1093                 return false;
1094         }
1095         ntfs_debug("Managed to lock extent mft record 0x%lx, write it.",
1096                         mft_no);
1097         if (NInoTestClearDirty(eni))
1098                 ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.",
1099                                 mft_no);
1100         /*
1101          * The write has to occur while we hold the mft record lock so return
1102          * the locked extent ntfs inode.
1103          */
1104         *locked_ni = eni;
1105         return true;
1106 }
1107 
1108 static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
1109                 "chkdsk.";
1110 
1111 /**
1112  * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name
1113  * @vol:        volume on which to search for a free mft record
1114  * @base_ni:    open base inode if allocating an extent mft record or NULL
1115  *
1116  * Search for a free mft record in the mft bitmap attribute on the ntfs volume
1117  * @vol.
1118  *
1119  * If @base_ni is NULL start the search at the default allocator position.
1120  *
1121  * If @base_ni is not NULL start the search at the mft record after the base
1122  * mft record @base_ni.
1123  *
1124  * Return the free mft record on success and -errno on error.  An error code of
1125  * -ENOSPC means that there are no free mft records in the currently
1126  * initialized mft bitmap.
1127  *
1128  * Locking: Caller must hold vol->mftbmp_lock for writing.
1129  */
1130 static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol,
1131                 ntfs_inode *base_ni)
1132 {
1133         s64 pass_end, ll, data_pos, pass_start, ofs, bit;
1134         unsigned long flags;
1135         struct address_space *mftbmp_mapping;
1136         u8 *buf, *byte;
1137         struct page *page;
1138         unsigned int page_ofs, size;
1139         u8 pass, b;
1140 
1141         ntfs_debug("Searching for free mft record in the currently "
1142                         "initialized mft bitmap.");
1143         mftbmp_mapping = vol->mftbmp_ino->i_mapping;
1144         /*
1145          * Set the end of the pass making sure we do not overflow the mft
1146          * bitmap.
1147          */
1148         read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags);
1149         pass_end = NTFS_I(vol->mft_ino)->allocated_size >>
1150                         vol->mft_record_size_bits;
1151         read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags);
1152         read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1153         ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3;
1154         read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1155         if (pass_end > ll)
1156                 pass_end = ll;
1157         pass = 1;
1158         if (!base_ni)
1159                 data_pos = vol->mft_data_pos;
1160         else
1161                 data_pos = base_ni->mft_no + 1;
1162         if (data_pos < 24)
1163                 data_pos = 24;
1164         if (data_pos >= pass_end) {
1165                 data_pos = 24;
1166                 pass = 2;
1167                 /* This happens on a freshly formatted volume. */
1168                 if (data_pos >= pass_end)
1169                         return -ENOSPC;
1170         }
1171         pass_start = data_pos;
1172         ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, "
1173                         "pass_end 0x%llx, data_pos 0x%llx.", pass,
1174                         (long long)pass_start, (long long)pass_end,
1175                         (long long)data_pos);
1176         /* Loop until a free mft record is found. */
1177         for (; pass <= 2;) {
1178                 /* Cap size to pass_end. */
1179                 ofs = data_pos >> 3;
1180                 page_ofs = ofs & ~PAGE_CACHE_MASK;
1181                 size = PAGE_CACHE_SIZE - page_ofs;
1182                 ll = ((pass_end + 7) >> 3) - ofs;
1183                 if (size > ll)
1184                         size = ll;
1185                 size <<= 3;
1186                 /*
1187                  * If we are still within the active pass, search the next page
1188                  * for a zero bit.
1189                  */
1190                 if (size) {
1191                         page = ntfs_map_page(mftbmp_mapping,
1192                                         ofs >> PAGE_CACHE_SHIFT);
1193                         if (IS_ERR(page)) {
1194                                 ntfs_error(vol->sb, "Failed to read mft "
1195                                                 "bitmap, aborting.");
1196                                 return PTR_ERR(page);
1197                         }
1198                         buf = (u8*)page_address(page) + page_ofs;
1199                         bit = data_pos & 7;
1200                         data_pos &= ~7ull;
1201                         ntfs_debug("Before inner for loop: size 0x%x, "
1202                                         "data_pos 0x%llx, bit 0x%llx", size,
1203                                         (long long)data_pos, (long long)bit);
1204                         for (; bit < size && data_pos + bit < pass_end;
1205                                         bit &= ~7ull, bit += 8) {
1206                                 byte = buf + (bit >> 3);
1207                                 if (*byte == 0xff)
1208                                         continue;
1209                                 b = ffz((unsigned long)*byte);
1210                                 if (b < 8 && b >= (bit & 7)) {
1211                                         ll = data_pos + (bit & ~7ull) + b;
1212                                         if (unlikely(ll > (1ll << 32))) {
1213                                                 ntfs_unmap_page(page);
1214                                                 return -ENOSPC;
1215                                         }
1216                                         *byte |= 1 << b;
1217                                         flush_dcache_page(page);
1218                                         set_page_dirty(page);
1219                                         ntfs_unmap_page(page);
1220                                         ntfs_debug("Done.  (Found and "
1221                                                         "allocated mft record "
1222                                                         "0x%llx.)",
1223                                                         (long long)ll);
1224                                         return ll;
1225                                 }
1226                         }
1227                         ntfs_debug("After inner for loop: size 0x%x, "
1228                                         "data_pos 0x%llx, bit 0x%llx", size,
1229                                         (long long)data_pos, (long long)bit);
1230                         data_pos += size;
1231                         ntfs_unmap_page(page);
1232                         /*
1233                          * If the end of the pass has not been reached yet,
1234                          * continue searching the mft bitmap for a zero bit.
1235                          */
1236                         if (data_pos < pass_end)
1237                                 continue;
1238                 }
1239                 /* Do the next pass. */
1240                 if (++pass == 2) {
1241                         /*
1242                          * Starting the second pass, in which we scan the first
1243                          * part of the zone which we omitted earlier.
1244                          */
1245                         pass_end = pass_start;
1246                         data_pos = pass_start = 24;
1247                         ntfs_debug("pass %i, pass_start 0x%llx, pass_end "
1248                                         "0x%llx.", pass, (long long)pass_start,
1249                                         (long long)pass_end);
1250                         if (data_pos >= pass_end)
1251                                 break;
1252                 }
1253         }
1254         /* No free mft records in currently initialized mft bitmap. */
1255         ntfs_debug("Done.  (No free mft records left in currently initialized "
1256                         "mft bitmap.)");
1257         return -ENOSPC;
1258 }
1259 
1260 /**
1261  * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster
1262  * @vol:        volume on which to extend the mft bitmap attribute
1263  *
1264  * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster.
1265  *
1266  * Note: Only changes allocated_size, i.e. does not touch initialized_size or
1267  * data_size.
1268  *
1269  * Return 0 on success and -errno on error.
1270  *
1271  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1272  *          - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for
1273  *            writing and releases it before returning.
1274  *          - This function takes vol->lcnbmp_lock for writing and releases it
1275  *            before returning.
1276  */
1277 static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol)
1278 {
1279         LCN lcn;
1280         s64 ll;
1281         unsigned long flags;
1282         struct page *page;
1283         ntfs_inode *mft_ni, *mftbmp_ni;
1284         runlist_element *rl, *rl2 = NULL;
1285         ntfs_attr_search_ctx *ctx = NULL;
1286         MFT_RECORD *mrec;
1287         ATTR_RECORD *a = NULL;
1288         int ret, mp_size;
1289         u32 old_alen = 0;
1290         u8 *b, tb;
1291         struct {
1292                 u8 added_cluster:1;
1293                 u8 added_run:1;
1294                 u8 mp_rebuilt:1;
1295         } status = { 0, 0, 0 };
1296 
1297         ntfs_debug("Extending mft bitmap allocation.");
1298         mft_ni = NTFS_I(vol->mft_ino);
1299         mftbmp_ni = NTFS_I(vol->mftbmp_ino);
1300         /*
1301          * Determine the last lcn of the mft bitmap.  The allocated size of the
1302          * mft bitmap cannot be zero so we are ok to do this.
1303          */
1304         down_write(&mftbmp_ni->runlist.lock);
1305         read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1306         ll = mftbmp_ni->allocated_size;
1307         read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1308         rl = ntfs_attr_find_vcn_nolock(mftbmp_ni,
1309                         (ll - 1) >> vol->cluster_size_bits, NULL);
1310         if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1311                 up_write(&mftbmp_ni->runlist.lock);
1312                 ntfs_error(vol->sb, "Failed to determine last allocated "
1313                                 "cluster of mft bitmap attribute.");
1314                 if (!IS_ERR(rl))
1315                         ret = -EIO;
1316                 else
1317                         ret = PTR_ERR(rl);
1318                 return ret;
1319         }
1320         lcn = rl->lcn + rl->length;
1321         ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.",
1322                         (long long)lcn);
1323         /*
1324          * Attempt to get the cluster following the last allocated cluster by
1325          * hand as it may be in the MFT zone so the allocator would not give it
1326          * to us.
1327          */
1328         ll = lcn >> 3;
1329         page = ntfs_map_page(vol->lcnbmp_ino->i_mapping,
1330                         ll >> PAGE_CACHE_SHIFT);
1331         if (IS_ERR(page)) {
1332                 up_write(&mftbmp_ni->runlist.lock);
1333                 ntfs_error(vol->sb, "Failed to read from lcn bitmap.");
1334                 return PTR_ERR(page);
1335         }
1336         b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK);
1337         tb = 1 << (lcn & 7ull);
1338         down_write(&vol->lcnbmp_lock);
1339         if (*b != 0xff && !(*b & tb)) {
1340                 /* Next cluster is free, allocate it. */
1341                 *b |= tb;
1342                 flush_dcache_page(page);
1343                 set_page_dirty(page);
1344                 up_write(&vol->lcnbmp_lock);
1345                 ntfs_unmap_page(page);
1346                 /* Update the mft bitmap runlist. */
1347                 rl->length++;
1348                 rl[1].vcn++;
1349                 status.added_cluster = 1;
1350                 ntfs_debug("Appending one cluster to mft bitmap.");
1351         } else {
1352                 up_write(&vol->lcnbmp_lock);
1353                 ntfs_unmap_page(page);
1354                 /* Allocate a cluster from the DATA_ZONE. */
1355                 rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE,
1356                                 true);
1357                 if (IS_ERR(rl2)) {
1358                         up_write(&mftbmp_ni->runlist.lock);
1359                         ntfs_error(vol->sb, "Failed to allocate a cluster for "
1360                                         "the mft bitmap.");
1361                         return PTR_ERR(rl2);
1362                 }
1363                 rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2);
1364                 if (IS_ERR(rl)) {
1365                         up_write(&mftbmp_ni->runlist.lock);
1366                         ntfs_error(vol->sb, "Failed to merge runlists for mft "
1367                                         "bitmap.");
1368                         if (ntfs_cluster_free_from_rl(vol, rl2)) {
1369                                 ntfs_error(vol->sb, "Failed to dealocate "
1370                                                 "allocated cluster.%s", es);
1371                                 NVolSetErrors(vol);
1372                         }
1373                         ntfs_free(rl2);
1374                         return PTR_ERR(rl);
1375                 }
1376                 mftbmp_ni->runlist.rl = rl;
1377                 status.added_run = 1;
1378                 ntfs_debug("Adding one run to mft bitmap.");
1379                 /* Find the last run in the new runlist. */
1380                 for (; rl[1].length; rl++)
1381                         ;
1382         }
1383         /*
1384          * Update the attribute record as well.  Note: @rl is the last
1385          * (non-terminator) runlist element of mft bitmap.
1386          */
1387         mrec = map_mft_record(mft_ni);
1388         if (IS_ERR(mrec)) {
1389                 ntfs_error(vol->sb, "Failed to map mft record.");
1390                 ret = PTR_ERR(mrec);
1391                 goto undo_alloc;
1392         }
1393         ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1394         if (unlikely(!ctx)) {
1395                 ntfs_error(vol->sb, "Failed to get search context.");
1396                 ret = -ENOMEM;
1397                 goto undo_alloc;
1398         }
1399         ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1400                         mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1401                         0, ctx);
1402         if (unlikely(ret)) {
1403                 ntfs_error(vol->sb, "Failed to find last attribute extent of "
1404                                 "mft bitmap attribute.");
1405                 if (ret == -ENOENT)
1406                         ret = -EIO;
1407                 goto undo_alloc;
1408         }
1409         a = ctx->attr;
1410         ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1411         /* Search back for the previous last allocated cluster of mft bitmap. */
1412         for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) {
1413                 if (ll >= rl2->vcn)
1414                         break;
1415         }
1416         BUG_ON(ll < rl2->vcn);
1417         BUG_ON(ll >= rl2->vcn + rl2->length);
1418         /* Get the size for the new mapping pairs array for this extent. */
1419         mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1420         if (unlikely(mp_size <= 0)) {
1421                 ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1422                                 "mft bitmap attribute extent.");
1423                 ret = mp_size;
1424                 if (!ret)
1425                         ret = -EIO;
1426                 goto undo_alloc;
1427         }
1428         /* Expand the attribute record if necessary. */
1429         old_alen = le32_to_cpu(a->length);
1430         ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1431                         le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1432         if (unlikely(ret)) {
1433                 if (ret != -ENOSPC) {
1434                         ntfs_error(vol->sb, "Failed to resize attribute "
1435                                         "record for mft bitmap attribute.");
1436                         goto undo_alloc;
1437                 }
1438                 // TODO: Deal with this by moving this extent to a new mft
1439                 // record or by starting a new extent in a new mft record or by
1440                 // moving other attributes out of this mft record.
1441                 // Note: It will need to be a special mft record and if none of
1442                 // those are available it gets rather complicated...
1443                 ntfs_error(vol->sb, "Not enough space in this mft record to "
1444                                 "accomodate extended mft bitmap attribute "
1445                                 "extent.  Cannot handle this yet.");
1446                 ret = -EOPNOTSUPP;
1447                 goto undo_alloc;
1448         }
1449         status.mp_rebuilt = 1;
1450         /* Generate the mapping pairs array directly into the attr record. */
1451         ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1452                         le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1453                         mp_size, rl2, ll, -1, NULL);
1454         if (unlikely(ret)) {
1455                 ntfs_error(vol->sb, "Failed to build mapping pairs array for "
1456                                 "mft bitmap attribute.");
1457                 goto undo_alloc;
1458         }
1459         /* Update the highest_vcn. */
1460         a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1461         /*
1462          * We now have extended the mft bitmap allocated_size by one cluster.
1463          * Reflect this in the ntfs_inode structure and the attribute record.
1464          */
1465         if (a->data.non_resident.lowest_vcn) {
1466                 /*
1467                  * We are not in the first attribute extent, switch to it, but
1468                  * first ensure the changes will make it to disk later.
1469                  */
1470                 flush_dcache_mft_record_page(ctx->ntfs_ino);
1471                 mark_mft_record_dirty(ctx->ntfs_ino);
1472                 ntfs_attr_reinit_search_ctx(ctx);
1473                 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1474                                 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL,
1475                                 0, ctx);
1476                 if (unlikely(ret)) {
1477                         ntfs_error(vol->sb, "Failed to find first attribute "
1478                                         "extent of mft bitmap attribute.");
1479                         goto restore_undo_alloc;
1480                 }
1481                 a = ctx->attr;
1482         }
1483         write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1484         mftbmp_ni->allocated_size += vol->cluster_size;
1485         a->data.non_resident.allocated_size =
1486                         cpu_to_sle64(mftbmp_ni->allocated_size);
1487         write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1488         /* Ensure the changes make it to disk. */
1489         flush_dcache_mft_record_page(ctx->ntfs_ino);
1490         mark_mft_record_dirty(ctx->ntfs_ino);
1491         ntfs_attr_put_search_ctx(ctx);
1492         unmap_mft_record(mft_ni);
1493         up_write(&mftbmp_ni->runlist.lock);
1494         ntfs_debug("Done.");
1495         return 0;
1496 restore_undo_alloc:
1497         ntfs_attr_reinit_search_ctx(ctx);
1498         if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1499                         mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1500                         0, ctx)) {
1501                 ntfs_error(vol->sb, "Failed to find last attribute extent of "
1502                                 "mft bitmap attribute.%s", es);
1503                 write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1504                 mftbmp_ni->allocated_size += vol->cluster_size;
1505                 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1506                 ntfs_attr_put_search_ctx(ctx);
1507                 unmap_mft_record(mft_ni);
1508                 up_write(&mftbmp_ni->runlist.lock);
1509                 /*
1510                  * The only thing that is now wrong is ->allocated_size of the
1511                  * base attribute extent which chkdsk should be able to fix.
1512                  */
1513                 NVolSetErrors(vol);
1514                 return ret;
1515         }
1516         a = ctx->attr;
1517         a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2);
1518 undo_alloc:
1519         if (status.added_cluster) {
1520                 /* Truncate the last run in the runlist by one cluster. */
1521                 rl->length--;
1522                 rl[1].vcn--;
1523         } else if (status.added_run) {
1524                 lcn = rl->lcn;
1525                 /* Remove the last run from the runlist. */
1526                 rl->lcn = rl[1].lcn;
1527                 rl->length = 0;
1528         }
1529         /* Deallocate the cluster. */
1530         down_write(&vol->lcnbmp_lock);
1531         if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1532                 ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es);
1533                 NVolSetErrors(vol);
1534         }
1535         up_write(&vol->lcnbmp_lock);
1536         if (status.mp_rebuilt) {
1537                 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1538                                 a->data.non_resident.mapping_pairs_offset),
1539                                 old_alen - le16_to_cpu(
1540                                 a->data.non_resident.mapping_pairs_offset),
1541                                 rl2, ll, -1, NULL)) {
1542                         ntfs_error(vol->sb, "Failed to restore mapping pairs "
1543                                         "array.%s", es);
1544                         NVolSetErrors(vol);
1545                 }
1546                 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1547                         ntfs_error(vol->sb, "Failed to restore attribute "
1548                                         "record.%s", es);
1549                         NVolSetErrors(vol);
1550                 }
1551                 flush_dcache_mft_record_page(ctx->ntfs_ino);
1552                 mark_mft_record_dirty(ctx->ntfs_ino);
1553         }
1554         if (ctx)
1555                 ntfs_attr_put_search_ctx(ctx);
1556         if (!IS_ERR(mrec))
1557                 unmap_mft_record(mft_ni);
1558         up_write(&mftbmp_ni->runlist.lock);
1559         return ret;
1560 }
1561 
1562 /**
1563  * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data
1564  * @vol:        volume on which to extend the mft bitmap attribute
1565  *
1566  * Extend the initialized portion of the mft bitmap attribute on the ntfs
1567  * volume @vol by 8 bytes.
1568  *
1569  * Note:  Only changes initialized_size and data_size, i.e. requires that
1570  * allocated_size is big enough to fit the new initialized_size.
1571  *
1572  * Return 0 on success and -error on error.
1573  *
1574  * Locking: Caller must hold vol->mftbmp_lock for writing.
1575  */
1576 static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol)
1577 {
1578         s64 old_data_size, old_initialized_size;
1579         unsigned long flags;
1580         struct inode *mftbmp_vi;
1581         ntfs_inode *mft_ni, *mftbmp_ni;
1582         ntfs_attr_search_ctx *ctx;
1583         MFT_RECORD *mrec;
1584         ATTR_RECORD *a;
1585         int ret;
1586 
1587         ntfs_debug("Extending mft bitmap initiailized (and data) size.");
1588         mft_ni = NTFS_I(vol->mft_ino);
1589         mftbmp_vi = vol->mftbmp_ino;
1590         mftbmp_ni = NTFS_I(mftbmp_vi);
1591         /* Get the attribute record. */
1592         mrec = map_mft_record(mft_ni);
1593         if (IS_ERR(mrec)) {
1594                 ntfs_error(vol->sb, "Failed to map mft record.");
1595                 return PTR_ERR(mrec);
1596         }
1597         ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1598         if (unlikely(!ctx)) {
1599                 ntfs_error(vol->sb, "Failed to get search context.");
1600                 ret = -ENOMEM;
1601                 goto unm_err_out;
1602         }
1603         ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1604                         mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx);
1605         if (unlikely(ret)) {
1606                 ntfs_error(vol->sb, "Failed to find first attribute extent of "
1607                                 "mft bitmap attribute.");
1608                 if (ret == -ENOENT)
1609                         ret = -EIO;
1610                 goto put_err_out;
1611         }
1612         a = ctx->attr;
1613         write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1614         old_data_size = i_size_read(mftbmp_vi);
1615         old_initialized_size = mftbmp_ni->initialized_size;
1616         /*
1617          * We can simply update the initialized_size before filling the space
1618          * with zeroes because the caller is holding the mft bitmap lock for
1619          * writing which ensures that no one else is trying to access the data.
1620          */
1621         mftbmp_ni->initialized_size += 8;
1622         a->data.non_resident.initialized_size =
1623                         cpu_to_sle64(mftbmp_ni->initialized_size);
1624         if (mftbmp_ni->initialized_size > old_data_size) {
1625                 i_size_write(mftbmp_vi, mftbmp_ni->initialized_size);
1626                 a->data.non_resident.data_size =
1627                                 cpu_to_sle64(mftbmp_ni->initialized_size);
1628         }
1629         write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1630         /* Ensure the changes make it to disk. */
1631         flush_dcache_mft_record_page(ctx->ntfs_ino);
1632         mark_mft_record_dirty(ctx->ntfs_ino);
1633         ntfs_attr_put_search_ctx(ctx);
1634         unmap_mft_record(mft_ni);
1635         /* Initialize the mft bitmap attribute value with zeroes. */
1636         ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0);
1637         if (likely(!ret)) {
1638                 ntfs_debug("Done.  (Wrote eight initialized bytes to mft "
1639                                 "bitmap.");
1640                 return 0;
1641         }
1642         ntfs_error(vol->sb, "Failed to write to mft bitmap.");
1643         /* Try to recover from the error. */
1644         mrec = map_mft_record(mft_ni);
1645         if (IS_ERR(mrec)) {
1646                 ntfs_error(vol->sb, "Failed to map mft record.%s", es);
1647                 NVolSetErrors(vol);
1648                 return ret;
1649         }
1650         ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1651         if (unlikely(!ctx)) {
1652                 ntfs_error(vol->sb, "Failed to get search context.%s", es);
1653                 NVolSetErrors(vol);
1654                 goto unm_err_out;
1655         }
1656         if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1657                         mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) {
1658                 ntfs_error(vol->sb, "Failed to find first attribute extent of "
1659                                 "mft bitmap attribute.%s", es);
1660                 NVolSetErrors(vol);
1661 put_err_out:
1662                 ntfs_attr_put_search_ctx(ctx);
1663 unm_err_out:
1664                 unmap_mft_record(mft_ni);
1665                 goto err_out;
1666         }
1667         a = ctx->attr;
1668         write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1669         mftbmp_ni->initialized_size = old_initialized_size;
1670         a->data.non_resident.initialized_size =
1671                         cpu_to_sle64(old_initialized_size);
1672         if (i_size_read(mftbmp_vi) != old_data_size) {
1673                 i_size_write(mftbmp_vi, old_data_size);
1674                 a->data.non_resident.data_size = cpu_to_sle64(old_data_size);
1675         }
1676         write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1677         flush_dcache_mft_record_page(ctx->ntfs_ino);
1678         mark_mft_record_dirty(ctx->ntfs_ino);
1679         ntfs_attr_put_search_ctx(ctx);
1680         unmap_mft_record(mft_ni);
1681 #ifdef DEBUG
1682         read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1683         ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, "
1684                         "data_size 0x%llx, initialized_size 0x%llx.",
1685                         (long long)mftbmp_ni->allocated_size,
1686                         (long long)i_size_read(mftbmp_vi),
1687                         (long long)mftbmp_ni->initialized_size);
1688         read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1689 #endif /* DEBUG */
1690 err_out:
1691         return ret;
1692 }
1693 
1694 /**
1695  * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute
1696  * @vol:        volume on which to extend the mft data attribute
1697  *
1698  * Extend the mft data attribute on the ntfs volume @vol by 16 mft records
1699  * worth of clusters or if not enough space for this by one mft record worth
1700  * of clusters.
1701  *
1702  * Note:  Only changes allocated_size, i.e. does not touch initialized_size or
1703  * data_size.
1704  *
1705  * Return 0 on success and -errno on error.
1706  *
1707  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1708  *          - This function takes NTFS_I(vol->mft_ino)->runlist.lock for
1709  *            writing and releases it before returning.
1710  *          - This function calls functions which take vol->lcnbmp_lock for
1711  *            writing and release it before returning.
1712  */
1713 static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol)
1714 {
1715         LCN lcn;
1716         VCN old_last_vcn;
1717         s64 min_nr, nr, ll;
1718         unsigned long flags;
1719         ntfs_inode *mft_ni;
1720         runlist_element *rl, *rl2;
1721         ntfs_attr_search_ctx *ctx = NULL;
1722         MFT_RECORD *mrec;
1723         ATTR_RECORD *a = NULL;
1724         int ret, mp_size;
1725         u32 old_alen = 0;
1726         bool mp_rebuilt = false;
1727 
1728         ntfs_debug("Extending mft data allocation.");
1729         mft_ni = NTFS_I(vol->mft_ino);
1730         /*
1731          * Determine the preferred allocation location, i.e. the last lcn of
1732          * the mft data attribute.  The allocated size of the mft data
1733          * attribute cannot be zero so we are ok to do this.
1734          */
1735         down_write(&mft_ni->runlist.lock);
1736         read_lock_irqsave(&mft_ni->size_lock, flags);
1737         ll = mft_ni->allocated_size;
1738         read_unlock_irqrestore(&mft_ni->size_lock, flags);
1739         rl = ntfs_attr_find_vcn_nolock(mft_ni,
1740                         (ll - 1) >> vol->cluster_size_bits, NULL);
1741         if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1742                 up_write(&mft_ni->runlist.lock);
1743                 ntfs_error(vol->sb, "Failed to determine last allocated "
1744                                 "cluster of mft data attribute.");
1745                 if (!IS_ERR(rl))
1746                         ret = -EIO;
1747                 else
1748                         ret = PTR_ERR(rl);
1749                 return ret;
1750         }
1751         lcn = rl->lcn + rl->length;
1752         ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn);
1753         /* Minimum allocation is one mft record worth of clusters. */
1754         min_nr = vol->mft_record_size >> vol->cluster_size_bits;
1755         if (!min_nr)
1756                 min_nr = 1;
1757         /* Want to allocate 16 mft records worth of clusters. */
1758         nr = vol->mft_record_size << 4 >> vol->cluster_size_bits;
1759         if (!nr)
1760                 nr = min_nr;
1761         /* Ensure we do not go above 2^32-1 mft records. */
1762         read_lock_irqsave(&mft_ni->size_lock, flags);
1763         ll = mft_ni->allocated_size;
1764         read_unlock_irqrestore(&mft_ni->size_lock, flags);
1765         if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1766                         vol->mft_record_size_bits >= (1ll << 32))) {
1767                 nr = min_nr;
1768                 if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1769                                 vol->mft_record_size_bits >= (1ll << 32))) {
1770                         ntfs_warning(vol->sb, "Cannot allocate mft record "
1771                                         "because the maximum number of inodes "
1772                                         "(2^32) has already been reached.");
1773                         up_write(&mft_ni->runlist.lock);
1774                         return -ENOSPC;
1775                 }
1776         }
1777         ntfs_debug("Trying mft data allocation with %s cluster count %lli.",
1778                         nr > min_nr ? "default" : "minimal", (long long)nr);
1779         old_last_vcn = rl[1].vcn;
1780         do {
1781                 rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE,
1782                                 true);
1783                 if (likely(!IS_ERR(rl2)))
1784                         break;
1785                 if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) {
1786                         ntfs_error(vol->sb, "Failed to allocate the minimal "
1787                                         "number of clusters (%lli) for the "
1788                                         "mft data attribute.", (long long)nr);
1789                         up_write(&mft_ni->runlist.lock);
1790                         return PTR_ERR(rl2);
1791                 }
1792                 /*
1793                  * There is not enough space to do the allocation, but there
1794                  * might be enough space to do a minimal allocation so try that
1795                  * before failing.
1796                  */
1797                 nr = min_nr;
1798                 ntfs_debug("Retrying mft data allocation with minimal cluster "
1799                                 "count %lli.", (long long)nr);
1800         } while (1);
1801         rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2);
1802         if (IS_ERR(rl)) {
1803                 up_write(&mft_ni->runlist.lock);
1804                 ntfs_error(vol->sb, "Failed to merge runlists for mft data "
1805                                 "attribute.");
1806                 if (ntfs_cluster_free_from_rl(vol, rl2)) {
1807                         ntfs_error(vol->sb, "Failed to dealocate clusters "
1808                                         "from the mft data attribute.%s", es);
1809                         NVolSetErrors(vol);
1810                 }
1811                 ntfs_free(rl2);
1812                 return PTR_ERR(rl);
1813         }
1814         mft_ni->runlist.rl = rl;
1815         ntfs_debug("Allocated %lli clusters.", (long long)nr);
1816         /* Find the last run in the new runlist. */
1817         for (; rl[1].length; rl++)
1818                 ;
1819         /* Update the attribute record as well. */
1820         mrec = map_mft_record(mft_ni);
1821         if (IS_ERR(mrec)) {
1822                 ntfs_error(vol->sb, "Failed to map mft record.");
1823                 ret = PTR_ERR(mrec);
1824                 goto undo_alloc;
1825         }
1826         ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1827         if (unlikely(!ctx)) {
1828                 ntfs_error(vol->sb, "Failed to get search context.");
1829                 ret = -ENOMEM;
1830                 goto undo_alloc;
1831         }
1832         ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1833                         CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx);
1834         if (unlikely(ret)) {
1835                 ntfs_error(vol->sb, "Failed to find last attribute extent of "
1836                                 "mft data attribute.");
1837                 if (ret == -ENOENT)
1838                         ret = -EIO;
1839                 goto undo_alloc;
1840         }
1841         a = ctx->attr;
1842         ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1843         /* Search back for the previous last allocated cluster of mft bitmap. */
1844         for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) {
1845                 if (ll >= rl2->vcn)
1846                         break;
1847         }
1848         BUG_ON(ll < rl2->vcn);
1849         BUG_ON(ll >= rl2->vcn + rl2->length);
1850         /* Get the size for the new mapping pairs array for this extent. */
1851         mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1852         if (unlikely(mp_size <= 0)) {
1853                 ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1854                                 "mft data attribute extent.");
1855                 ret = mp_size;
1856                 if (!ret)
1857                         ret = -EIO;
1858                 goto undo_alloc;
1859         }
1860         /* Expand the attribute record if necessary. */
1861         old_alen = le32_to_cpu(a->length);
1862         ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1863                         le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1864         if (unlikely(ret)) {
1865                 if (ret != -ENOSPC) {
1866                         ntfs_error(vol->sb, "Failed to resize attribute "
1867                                         "record for mft data attribute.");
1868                         goto undo_alloc;
1869                 }
1870                 // TODO: Deal with this by moving this extent to a new mft
1871                 // record or by starting a new extent in a new mft record or by
1872                 // moving other attributes out of this mft record.
1873                 // Note: Use the special reserved mft records and ensure that
1874                 // this extent is not required to find the mft record in
1875                 // question.  If no free special records left we would need to
1876                 // move an existing record away, insert ours in its place, and
1877                 // then place the moved record into the newly allocated space
1878                 // and we would then need to update all references to this mft
1879                 // record appropriately.  This is rather complicated...
1880                 ntfs_error(vol->sb, "Not enough space in this mft record to "
1881                                 "accomodate extended mft data attribute "
1882                                 "extent.  Cannot handle this yet.");
1883                 ret = -EOPNOTSUPP;
1884                 goto undo_alloc;
1885         }
1886         mp_rebuilt = true;
1887         /* Generate the mapping pairs array directly into the attr record. */
1888         ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1889                         le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1890                         mp_size, rl2, ll, -1, NULL);
1891         if (unlikely(ret)) {
1892                 ntfs_error(vol->sb, "Failed to build mapping pairs array of "
1893                                 "mft data attribute.");
1894                 goto undo_alloc;
1895         }
1896         /* Update the highest_vcn. */
1897         a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1898         /*
1899          * We now have extended the mft data allocated_size by nr clusters.
1900          * Reflect this in the ntfs_inode structure and the attribute record.
1901          * @rl is the last (non-terminator) runlist element of mft data
1902          * attribute.
1903          */
1904         if (a->data.non_resident.lowest_vcn) {
1905                 /*
1906                  * We are not in the first attribute extent, switch to it, but
1907                  * first ensure the changes will make it to disk later.
1908                  */
1909                 flush_dcache_mft_record_page(ctx->ntfs_ino);
1910                 mark_mft_record_dirty(ctx->ntfs_ino);
1911                 ntfs_attr_reinit_search_ctx(ctx);
1912                 ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name,
1913                                 mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0,
1914                                 ctx);
1915                 if (unlikely(ret)) {
1916                         ntfs_error(vol->sb, "Failed to find first attribute "
1917                                         "extent of mft data attribute.");
1918                         goto restore_undo_alloc;
1919                 }
1920                 a = ctx->attr;
1921         }
1922         write_lock_irqsave(&mft_ni->size_lock, flags);
1923         mft_ni->allocated_size += nr << vol->cluster_size_bits;
1924         a->data.non_resident.allocated_size =
1925                         cpu_to_sle64(mft_ni->allocated_size);
1926         write_unlock_irqrestore(&mft_ni->size_lock, flags);
1927         /* Ensure the changes make it to disk. */
1928         flush_dcache_mft_record_page(ctx->ntfs_ino);
1929         mark_mft_record_dirty(ctx->ntfs_ino);
1930         ntfs_attr_put_search_ctx(ctx);
1931         unmap_mft_record(mft_ni);
1932         up_write(&mft_ni->runlist.lock);
1933         ntfs_debug("Done.");
1934         return 0;
1935 restore_undo_alloc:
1936         ntfs_attr_reinit_search_ctx(ctx);
1937         if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1938                         CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) {
1939                 ntfs_error(vol->sb, "Failed to find last attribute extent of "
1940                                 "mft data attribute.%s", es);
1941                 write_lock_irqsave(&mft_ni->size_lock, flags);
1942                 mft_ni->allocated_size += nr << vol->cluster_size_bits;
1943                 write_unlock_irqrestore(&mft_ni->size_lock, flags);
1944                 ntfs_attr_put_search_ctx(ctx);
1945                 unmap_mft_record(mft_ni);
1946                 up_write(&mft_ni->runlist.lock);
1947                 /*
1948                  * The only thing that is now wrong is ->allocated_size of the
1949                  * base attribute extent which chkdsk should be able to fix.
1950                  */
1951                 NVolSetErrors(vol);
1952                 return ret;
1953         }
1954         ctx->attr->data.non_resident.highest_vcn =
1955                         cpu_to_sle64(old_last_vcn - 1);
1956 undo_alloc:
1957         if (ntfs_cluster_free(mft_ni, old_last_vcn, -1, ctx) < 0) {
1958                 ntfs_error(vol->sb, "Failed to free clusters from mft data "
1959                                 "attribute.%s", es);
1960                 NVolSetErrors(vol);
1961         }
1962         a = ctx->attr;
1963         if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) {
1964                 ntfs_error(vol->sb, "Failed to truncate mft data attribute "
1965                                 "runlist.%s", es);
1966                 NVolSetErrors(vol);
1967         }
1968         if (mp_rebuilt && !IS_ERR(ctx->mrec)) {
1969                 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1970                                 a->data.non_resident.mapping_pairs_offset),
1971                                 old_alen - le16_to_cpu(
1972                                 a->data.non_resident.mapping_pairs_offset),
1973                                 rl2, ll, -1, NULL)) {
1974                         ntfs_error(vol->sb, "Failed to restore mapping pairs "
1975                                         "array.%s", es);
1976                         NVolSetErrors(vol);
1977                 }
1978                 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1979                         ntfs_error(vol->sb, "Failed to restore attribute "
1980                                         "record.%s", es);
1981                         NVolSetErrors(vol);
1982                 }
1983                 flush_dcache_mft_record_page(ctx->ntfs_ino);
1984                 mark_mft_record_dirty(ctx->ntfs_ino);
1985         } else if (IS_ERR(ctx->mrec)) {
1986                 ntfs_error(vol->sb, "Failed to restore attribute search "
1987                                 "context.%s", es);
1988                 NVolSetErrors(vol);
1989         }
1990         if (ctx)
1991                 ntfs_attr_put_search_ctx(ctx);
1992         if (!IS_ERR(mrec))
1993                 unmap_mft_record(mft_ni);
1994         up_write(&mft_ni->runlist.lock);
1995         return ret;
1996 }
1997 
1998 /**
1999  * ntfs_mft_record_layout - layout an mft record into a memory buffer
2000  * @vol:        volume to which the mft record will belong
2001  * @mft_no:     mft reference specifying the mft record number
2002  * @m:          destination buffer of size >= @vol->mft_record_size bytes
2003  *
2004  * Layout an empty, unused mft record with the mft record number @mft_no into
2005  * the buffer @m.  The volume @vol is needed because the mft record structure
2006  * was modified in NTFS 3.1 so we need to know which volume version this mft
2007  * record will be used on.
2008  *
2009  * Return 0 on success and -errno on error.
2010  */
2011 static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no,
2012                 MFT_RECORD *m)
2013 {
2014         ATTR_RECORD *a;
2015 
2016         ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2017         if (mft_no >= (1ll << 32)) {
2018                 ntfs_error(vol->sb, "Mft record number 0x%llx exceeds "
2019                                 "maximum of 2^32.", (long long)mft_no);
2020                 return -ERANGE;
2021         }
2022         /* Start by clearing the whole mft record to gives us a clean slate. */
2023         memset(m, 0, vol->mft_record_size);
2024         /* Aligned to 2-byte boundary. */
2025         if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver))
2026                 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1);
2027         else {
2028                 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1);
2029                 /*
2030                  * Set the NTFS 3.1+ specific fields while we know that the
2031                  * volume version is 3.1+.
2032                  */
2033                 m->reserved = 0;
2034                 m->mft_record_number = cpu_to_le32((u32)mft_no);
2035         }
2036         m->magic = magic_FILE;
2037         if (vol->mft_record_size >= NTFS_BLOCK_SIZE)
2038                 m->usa_count = cpu_to_le16(vol->mft_record_size /
2039                                 NTFS_BLOCK_SIZE + 1);
2040         else {
2041                 m->usa_count = cpu_to_le16(1);
2042                 ntfs_warning(vol->sb, "Sector size is bigger than mft record "
2043                                 "size.  Setting usa_count to 1.  If chkdsk "
2044                                 "reports this as corruption, please email "
2045                                 "linux-ntfs-dev@lists.sourceforge.net stating "
2046                                 "that you saw this message and that the "
2047                                 "modified filesystem created was corrupt.  "
2048                                 "Thank you.");
2049         }
2050         /* Set the update sequence number to 1. */
2051         *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1);
2052         m->lsn = 0;
2053         m->sequence_number = cpu_to_le16(1);
2054         m->link_count = 0;
2055         /*
2056          * Place the attributes straight after the update sequence array,
2057          * aligned to 8-byte boundary.
2058          */
2059         m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) +
2060                         (le16_to_cpu(m->usa_count) << 1) + 7) & ~7);
2061         m->flags = 0;
2062         /*
2063          * Using attrs_offset plus eight bytes (for the termination attribute).
2064          * attrs_offset is already aligned to 8-byte boundary, so no need to
2065          * align again.
2066          */
2067         m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8);
2068         m->bytes_allocated = cpu_to_le32(vol->mft_record_size);
2069         m->base_mft_record = 0;
2070         m->next_attr_instance = 0;
2071         /* Add the termination attribute. */
2072         a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset));
2073         a->type = AT_END;
2074         a->length = 0;
2075         ntfs_debug("Done.");
2076         return 0;
2077 }
2078 
2079 /**
2080  * ntfs_mft_record_format - format an mft record on an ntfs volume
2081  * @vol:        volume on which to format the mft record
2082  * @mft_no:     mft record number to format
2083  *
2084  * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused
2085  * mft record into the appropriate place of the mft data attribute.  This is
2086  * used when extending the mft data attribute.
2087  *
2088  * Return 0 on success and -errno on error.
2089  */
2090 static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no)
2091 {
2092         loff_t i_size;
2093         struct inode *mft_vi = vol->mft_ino;
2094         struct page *page;
2095         MFT_RECORD *m;
2096         pgoff_t index, end_index;
2097         unsigned int ofs;
2098         int err;
2099 
2100         ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2101         /*
2102          * The index into the page cache and the offset within the page cache
2103          * page of the wanted mft record.
2104          */
2105         index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2106         ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2107         /* The maximum valid index into the page cache for $MFT's data. */
2108         i_size = i_size_read(mft_vi);
2109         end_index = i_size >> PAGE_CACHE_SHIFT;
2110         if (unlikely(index >= end_index)) {
2111                 if (unlikely(index > end_index || ofs + vol->mft_record_size >=
2112                                 (i_size & ~PAGE_CACHE_MASK))) {
2113                         ntfs_error(vol->sb, "Tried to format non-existing mft "
2114                                         "record 0x%llx.", (long long)mft_no);
2115                         return -ENOENT;
2116                 }
2117         }
2118         /* Read, map, and pin the page containing the mft record. */
2119         page = ntfs_map_page(mft_vi->i_mapping, index);
2120         if (IS_ERR(page)) {
2121                 ntfs_error(vol->sb, "Failed to map page containing mft record "
2122                                 "to format 0x%llx.", (long long)mft_no);
2123                 return PTR_ERR(page);
2124         }
2125         lock_page(page);
2126         BUG_ON(!PageUptodate(page));
2127         ClearPageUptodate(page);
2128         m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2129         err = ntfs_mft_record_layout(vol, mft_no, m);
2130         if (unlikely(err)) {
2131                 ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.",
2132                                 (long long)mft_no);
2133                 SetPageUptodate(page);
2134                 unlock_page(page);
2135                 ntfs_unmap_page(page);
2136                 return err;
2137         }
2138         flush_dcache_page(page);
2139         SetPageUptodate(page);
2140         unlock_page(page);
2141         /*
2142          * Make sure the mft record is written out to disk.  We could use
2143          * ilookup5() to check if an inode is in icache and so on but this is
2144          * unnecessary as ntfs_writepage() will write the dirty record anyway.
2145          */
2146         mark_ntfs_record_dirty(page, ofs);
2147         ntfs_unmap_page(page);
2148         ntfs_debug("Done.");
2149         return 0;
2150 }
2151 
2152 /**
2153  * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume
2154  * @vol:        [IN]  volume on which to allocate the mft record
2155  * @mode:       [IN]  mode if want a file or directory, i.e. base inode or 0
2156  * @base_ni:    [IN]  open base inode if allocating an extent mft record or NULL
2157  * @mrec:       [OUT] on successful return this is the mapped mft record
2158  *
2159  * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol.
2160  *
2161  * If @base_ni is NULL make the mft record a base mft record, i.e. a file or
2162  * direvctory inode, and allocate it at the default allocator position.  In
2163  * this case @mode is the file mode as given to us by the caller.  We in
2164  * particular use @mode to distinguish whether a file or a directory is being
2165  * created (S_IFDIR(mode) and S_IFREG(mode), respectively).
2166  *
2167  * If @base_ni is not NULL make the allocated mft record an extent record,
2168  * allocate it starting at the mft record after the base mft record and attach
2169  * the allocated and opened ntfs inode to the base inode @base_ni.  In this
2170  * case @mode must be 0 as it is meaningless for extent inodes.
2171  *
2172  * You need to check the return value with IS_ERR().  If false, the function
2173  * was successful and the return value is the now opened ntfs inode of the
2174  * allocated mft record.  *@mrec is then set to the allocated, mapped, pinned,
2175  * and locked mft record.  If IS_ERR() is true, the function failed and the
2176  * error code is obtained from PTR_ERR(return value).  *@mrec is undefined in
2177  * this case.
2178  *
2179  * Allocation strategy:
2180  *
2181  * To find a free mft record, we scan the mft bitmap for a zero bit.  To
2182  * optimize this we start scanning at the place specified by @base_ni or if
2183  * @base_ni is NULL we start where we last stopped and we perform wrap around
2184  * when we reach the end.  Note, we do not try to allocate mft records below
2185  * number 24 because numbers 0 to 15 are the defined system files anyway and 16
2186  * to 24 are special in that they are used for storing extension mft records
2187  * for the $DATA attribute of $MFT.  This is required to avoid the possibility
2188  * of creating a runlist with a circular dependency which once written to disk
2189  * can never be read in again.  Windows will only use records 16 to 24 for
2190  * normal files if the volume is completely out of space.  We never use them
2191  * which means that when the volume is really out of space we cannot create any
2192  * more files while Windows can still create up to 8 small files.  We can start
2193  * doing this at some later time, it does not matter much for now.
2194  *
2195  * When scanning the mft bitmap, we only search up to the last allocated mft
2196  * record.  If there are no free records left in the range 24 to number of
2197  * allocated mft records, then we extend the $MFT/$DATA attribute in order to
2198  * create free mft records.  We extend the allocated size of $MFT/$DATA by 16
2199  * records at a time or one cluster, if cluster size is above 16kiB.  If there
2200  * is not sufficient space to do this, we try to extend by a single mft record
2201  * or one cluster, if cluster size is above the mft record size.
2202  *
2203  * No matter how many mft records we allocate, we initialize only the first
2204  * allocated mft record, incrementing mft data size and initialized size
2205  * accordingly, open an ntfs_inode for it and return it to the caller, unless
2206  * there are less than 24 mft records, in which case we allocate and initialize
2207  * mft records until we reach record 24 which we consider as the first free mft
2208  * record for use by normal files.
2209  *
2210  * If during any stage we overflow the initialized data in the mft bitmap, we
2211  * extend the initialized size (and data size) by 8 bytes, allocating another
2212  * cluster if required.  The bitmap data size has to be at least equal to the
2213  * number of mft records in the mft, but it can be bigger, in which case the
2214  * superflous bits are padded with zeroes.
2215  *
2216  * Thus, when we return successfully (IS_ERR() is false), we will have:
2217  *      - initialized / extended the mft bitmap if necessary,
2218  *      - initialized / extended the mft data if necessary,
2219  *      - set the bit corresponding to the mft record being allocated in the
2220  *        mft bitmap,
2221  *      - opened an ntfs_inode for the allocated mft record, and we will have
2222  *      - returned the ntfs_inode as well as the allocated mapped, pinned, and
2223  *        locked mft record.
2224  *
2225  * On error, the volume will be left in a consistent state and no record will
2226  * be allocated.  If rolling back a partial operation fails, we may leave some
2227  * inconsistent metadata in which case we set NVolErrors() so the volume is
2228  * left dirty when unmounted.
2229  *
2230  * Note, this function cannot make use of most of the normal functions, like
2231  * for example for attribute resizing, etc, because when the run list overflows
2232  * the base mft record and an attribute list is used, it is very important that
2233  * the extension mft records used to store the $DATA attribute of $MFT can be
2234  * reached without having to read the information contained inside them, as
2235  * this would make it impossible to find them in the first place after the
2236  * volume is unmounted.  $MFT/$BITMAP probably does not need to follow this
2237  * rule because the bitmap is not essential for finding the mft records, but on
2238  * the other hand, handling the bitmap in this special way would make life
2239  * easier because otherwise there might be circular invocations of functions
2240  * when reading the bitmap.
2241  */
2242 ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode,
2243                 ntfs_inode *base_ni, MFT_RECORD **mrec)
2244 {
2245         s64 ll, bit, old_data_initialized, old_data_size;
2246         unsigned long flags;
2247         struct inode *vi;
2248         struct page *page;
2249         ntfs_inode *mft_ni, *mftbmp_ni, *ni;
2250         ntfs_attr_search_ctx *ctx;
2251         MFT_RECORD *m;
2252         ATTR_RECORD *a;
2253         pgoff_t index;
2254         unsigned int ofs;
2255         int err;
2256         le16 seq_no, usn;
2257         bool record_formatted = false;
2258 
2259         if (base_ni) {
2260                 ntfs_debug("Entering (allocating an extent mft record for "
2261                                 "base mft record 0x%llx).",
2262                                 (long long)base_ni->mft_no);
2263                 /* @mode and @base_ni are mutually exclusive. */
2264                 BUG_ON(mode);
2265         } else
2266                 ntfs_debug("Entering (allocating a base mft record).");
2267         if (mode) {
2268                 /* @mode and @base_ni are mutually exclusive. */
2269                 BUG_ON(base_ni);
2270                 /* We only support creation of normal files and directories. */
2271                 if (!S_ISREG(mode) && !S_ISDIR(mode))
2272                         return ERR_PTR(-EOPNOTSUPP);
2273         }
2274         BUG_ON(!mrec);
2275         mft_ni = NTFS_I(vol->mft_ino);
2276         mftbmp_ni = NTFS_I(vol->mftbmp_ino);
2277         down_write(&vol->mftbmp_lock);
2278         bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni);
2279         if (bit >= 0) {
2280                 ntfs_debug("Found and allocated free record (#1), bit 0x%llx.",
2281                                 (long long)bit);
2282                 goto have_alloc_rec;
2283         }
2284         if (bit != -ENOSPC) {
2285                 up_write(&vol->mftbmp_lock);
2286                 return ERR_PTR(bit);
2287         }
2288         /*
2289          * No free mft records left.  If the mft bitmap already covers more
2290          * than the currently used mft records, the next records are all free,
2291          * so we can simply allocate the first unused mft record.
2292          * Note: We also have to make sure that the mft bitmap at least covers
2293          * the first 24 mft records as they are special and whilst they may not
2294          * be in use, we do not allocate from them.
2295          */
2296         read_lock_irqsave(&mft_ni->size_lock, flags);
2297         ll = mft_ni->initialized_size >> vol->mft_record_size_bits;
2298         read_unlock_irqrestore(&mft_ni->size_lock, flags);
2299         read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2300         old_data_initialized = mftbmp_ni->initialized_size;
2301         read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2302         if (old_data_initialized << 3 > ll && old_data_initialized > 3) {
2303                 bit = ll;
2304                 if (bit < 24)
2305                         bit = 24;
2306                 if (unlikely(bit >= (1ll << 32)))
2307                         goto max_err_out;
2308                 ntfs_debug("Found free record (#2), bit 0x%llx.",
2309                                 (long long)bit);
2310                 goto found_free_rec;
2311         }
2312         /*
2313          * The mft bitmap needs to be expanded until it covers the first unused
2314          * mft record that we can allocate.
2315          * Note: The smallest mft record we allocate is mft record 24.
2316          */
2317         bit = old_data_initialized << 3;
2318         if (unlikely(bit >= (1ll << 32)))
2319                 goto max_err_out;
2320         read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2321         old_data_size = mftbmp_ni->allocated_size;
2322         ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, "
2323                         "data_size 0x%llx, initialized_size 0x%llx.",
2324                         (long long)old_data_size,
2325                         (long long)i_size_read(vol->mftbmp_ino),
2326                         (long long)old_data_initialized);
2327         read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2328         if (old_data_initialized + 8 > old_data_size) {
2329                 /* Need to extend bitmap by one more cluster. */
2330                 ntfs_debug("mftbmp: initialized_size + 8 > allocated_size.");
2331                 err = ntfs_mft_bitmap_extend_allocation_nolock(vol);
2332                 if (unlikely(err)) {
2333                         up_write(&vol->mftbmp_lock);
2334                         goto err_out;
2335                 }
2336 #ifdef DEBUG
2337                 read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2338                 ntfs_debug("Status of mftbmp after allocation extension: "
2339                                 "allocated_size 0x%llx, data_size 0x%llx, "
2340                                 "initialized_size 0x%llx.",
2341                                 (long long)mftbmp_ni->allocated_size,
2342                                 (long long)i_size_read(vol->mftbmp_ino),
2343                                 (long long)mftbmp_ni->initialized_size);
2344                 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2345 #endif /* DEBUG */
2346         }
2347         /*
2348          * We now have sufficient allocated space, extend the initialized_size
2349          * as well as the data_size if necessary and fill the new space with
2350          * zeroes.
2351          */
2352         err = ntfs_mft_bitmap_extend_initialized_nolock(vol);
2353         if (unlikely(err)) {
2354                 up_write(&vol->mftbmp_lock);
2355                 goto err_out;
2356         }
2357 #ifdef DEBUG
2358         read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2359         ntfs_debug("Status of mftbmp after initialized extention: "
2360                         "allocated_size 0x%llx, data_size 0x%llx, "
2361                         "initialized_size 0x%llx.",
2362                         (long long)mftbmp_ni->allocated_size,
2363                         (long long)i_size_read(vol->mftbmp_ino),
2364                         (long long)mftbmp_ni->initialized_size);
2365         read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2366 #endif /* DEBUG */
2367         ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit);
2368 found_free_rec:
2369         /* @bit is the found free mft record, allocate it in the mft bitmap. */
2370         ntfs_debug("At found_free_rec.");
2371         err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit);
2372         if (unlikely(err)) {
2373                 ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap.");
2374                 up_write(&vol->mftbmp_lock);
2375                 goto err_out;
2376         }
2377         ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit);
2378 have_alloc_rec:
2379         /*
2380          * The mft bitmap is now uptodate.  Deal with mft data attribute now.
2381          * Note, we keep hold of the mft bitmap lock for writing until all
2382          * modifications to the mft data attribute are complete, too, as they
2383          * will impact decisions for mft bitmap and mft record allocation done
2384          * by a parallel allocation and if the lock is not maintained a
2385          * parallel allocation could allocate the same mft record as this one.
2386          */
2387         ll = (bit + 1) << vol->mft_record_size_bits;
2388         read_lock_irqsave(&mft_ni->size_lock, flags);
2389         old_data_initialized = mft_ni->initialized_size;
2390         read_unlock_irqrestore(&mft_ni->size_lock, flags);
2391         if (ll <= old_data_initialized) {
2392                 ntfs_debug("Allocated mft record already initialized.");
2393                 goto mft_rec_already_initialized;
2394         }
2395         ntfs_debug("Initializing allocated mft record.");
2396         /*
2397          * The mft record is outside the initialized data.  Extend the mft data
2398          * attribute until it covers the allocated record.  The loop is only
2399          * actually traversed more than once when a freshly formatted volume is
2400          * first written to so it optimizes away nicely in the common case.
2401          */
2402         read_lock_irqsave(&mft_ni->size_lock, flags);
2403         ntfs_debug("Status of mft data before extension: "
2404                         "allocated_size 0x%llx, data_size 0x%llx, "
2405                         "initialized_size 0x%llx.",
2406                         (long long)mft_ni->allocated_size,
2407                         (long long)i_size_read(vol->mft_ino),
2408                         (long long)mft_ni->initialized_size);
2409         while (ll > mft_ni->allocated_size) {
2410                 read_unlock_irqrestore(&mft_ni->size_lock, flags);
2411                 err = ntfs_mft_data_extend_allocation_nolock(vol);
2412                 if (unlikely(err)) {
2413                         ntfs_error(vol->sb, "Failed to extend mft data "
2414                                         "allocation.");
2415                         goto undo_mftbmp_alloc_nolock;
2416                 }
2417                 read_lock_irqsave(&mft_ni->size_lock, flags);
2418                 ntfs_debug("Status of mft data after allocation extension: "
2419                                 "allocated_size 0x%llx, data_size 0x%llx, "
2420                                 "initialized_size 0x%llx.",
2421                                 (long long)mft_ni->allocated_size,
2422                                 (long long)i_size_read(vol->mft_ino),
2423                                 (long long)mft_ni->initialized_size);
2424         }
2425         read_unlock_irqrestore(&mft_ni->size_lock, flags);
2426         /*
2427          * Extend mft data initialized size (and data size of course) to reach
2428          * the allocated mft record, formatting the mft records allong the way.
2429          * Note: We only modify the ntfs_inode structure as that is all that is
2430          * needed by ntfs_mft_record_format().  We will update the attribute
2431          * record itself in one fell swoop later on.
2432          */
2433         write_lock_irqsave(&mft_ni->size_lock, flags);
2434         old_data_initialized = mft_ni->initialized_size;
2435         old_data_size = vol->mft_ino->i_size;
2436         while (ll > mft_ni->initialized_size) {
2437                 s64 new_initialized_size, mft_no;
2438                 
2439                 new_initialized_size = mft_ni->initialized_size +
2440                                 vol->mft_record_size;
2441                 mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits;
2442                 if (new_initialized_size > i_size_read(vol->mft_ino))
2443                         i_size_write(vol->mft_ino, new_initialized_size);
2444                 write_unlock_irqrestore(&mft_ni->size_lock, flags);
2445                 ntfs_debug("Initializing mft record 0x%llx.",
2446                                 (long long)mft_no);
2447                 err = ntfs_mft_record_format(vol, mft_no);
2448                 if (unlikely(err)) {
2449                         ntfs_error(vol->sb, "Failed to format mft record.");
2450                         goto undo_data_init;
2451                 }
2452                 write_lock_irqsave(&mft_ni->size_lock, flags);
2453                 mft_ni->initialized_size = new_initialized_size;
2454         }
2455         write_unlock_irqrestore(&mft_ni->size_lock, flags);
2456         record_formatted = true;
2457         /* Update the mft data attribute record to reflect the new sizes. */
2458         m = map_mft_record(mft_ni);
2459         if (IS_ERR(m)) {
2460                 ntfs_error(vol->sb, "Failed to map mft record.");
2461                 err = PTR_ERR(m);
2462                 goto undo_data_init;
2463         }
2464         ctx = ntfs_attr_get_search_ctx(mft_ni, m);
2465         if (unlikely(!ctx)) {
2466                 ntfs_error(vol->sb, "Failed to get search context.");
2467                 err = -ENOMEM;
2468                 unmap_mft_record(mft_ni);
2469                 goto undo_data_init;
2470         }
2471         err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
2472                         CASE_SENSITIVE, 0, NULL, 0, ctx);
2473         if (unlikely(err)) {
2474                 ntfs_error(vol->sb, "Failed to find first attribute extent of "
2475                                 "mft data attribute.");
2476                 ntfs_attr_put_search_ctx(ctx);
2477                 unmap_mft_record(mft_ni);
2478                 goto undo_data_init;
2479         }
2480         a = ctx->attr;
2481         read_lock_irqsave(&mft_ni->size_lock, flags);
2482         a->data.non_resident.initialized_size =
2483                         cpu_to_sle64(mft_ni->initialized_size);
2484         a->data.non_resident.data_size =
2485                         cpu_to_sle64(i_size_read(vol->mft_ino));
2486         read_unlock_irqrestore(&mft_ni->size_lock, flags);
2487         /* Ensure the changes make it to disk. */
2488         flush_dcache_mft_record_page(ctx->ntfs_ino);
2489         mark_mft_record_dirty(ctx->ntfs_ino);
2490         ntfs_attr_put_search_ctx(ctx);
2491         unmap_mft_record(mft_ni);
2492         read_lock_irqsave(&mft_ni->size_lock, flags);
2493         ntfs_debug("Status of mft data after mft record initialization: "
2494                         "allocated_size 0x%llx, data_size 0x%llx, "
2495                         "initialized_size 0x%llx.",
2496                         (long long)mft_ni->allocated_size,
2497                         (long long)i_size_read(vol->mft_ino),
2498                         (long long)mft_ni->initialized_size);
2499         BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size);
2500         BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino));
2501         read_unlock_irqrestore(&mft_ni->size_lock, flags);
2502 mft_rec_already_initialized:
2503         /*
2504          * We can finally drop the mft bitmap lock as the mft data attribute
2505          * has been fully updated.  The only disparity left is that the
2506          * allocated mft record still needs to be marked as in use to match the
2507          * set bit in the mft bitmap but this is actually not a problem since
2508          * this mft record is not referenced from anywhere yet and the fact
2509          * that it is allocated in the mft bitmap means that no-one will try to
2510          * allocate it either.
2511          */
2512         up_write(&vol->mftbmp_lock);
2513         /*
2514          * We now have allocated and initialized the mft record.  Calculate the
2515          * index of and the offset within the page cache page the record is in.
2516          */
2517         index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2518         ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2519         /* Read, map, and pin the page containing the mft record. */
2520         page = ntfs_map_page(vol->mft_ino->i_mapping, index);
2521         if (IS_ERR(page)) {
2522                 ntfs_error(vol->sb, "Failed to map page containing allocated "
2523                                 "mft record 0x%llx.", (long long)bit);
2524                 err = PTR_ERR(page);
2525                 goto undo_mftbmp_alloc;
2526         }
2527         lock_page(page);
2528         BUG_ON(!PageUptodate(page));
2529         ClearPageUptodate(page);
2530         m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2531         /* If we just formatted the mft record no need to do it again. */
2532         if (!record_formatted) {
2533                 /* Sanity check that the mft record is really not in use. */
2534                 if (ntfs_is_file_record(m->magic) &&
2535                                 (m->flags & MFT_RECORD_IN_USE)) {
2536                         ntfs_error(vol->sb, "Mft record 0x%llx was marked "
2537                                         "free in mft bitmap but is marked "
2538                                         "used itself.  Corrupt filesystem.  "
2539                                         "Unmount and run chkdsk.",
2540                                         (long long)bit);
2541                         err = -EIO;
2542                         SetPageUptodate(page);
2543                         unlock_page(page);
2544                         ntfs_unmap_page(page);
2545                         NVolSetErrors(vol);
2546                         goto undo_mftbmp_alloc;
2547                 }
2548                 /*
2549                  * We need to (re-)format the mft record, preserving the
2550                  * sequence number if it is not zero as well as the update
2551                  * sequence number if it is not zero or -1 (0xffff).  This
2552                  * means we do not need to care whether or not something went
2553                  * wrong with the previous mft record.
2554                  */
2555                 seq_no = m->sequence_number;
2556                 usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs));
2557                 err = ntfs_mft_record_layout(vol, bit, m);
2558                 if (unlikely(err)) {
2559                         ntfs_error(vol->sb, "Failed to layout allocated mft "
2560                                         "record 0x%llx.", (long long)bit);
2561                         SetPageUptodate(page);
2562                         unlock_page(page);
2563                         ntfs_unmap_page(page);
2564                         goto undo_mftbmp_alloc;
2565                 }
2566                 if (seq_no)
2567                         m->sequence_number = seq_no;
2568                 if (usn && le16_to_cpu(usn) != 0xffff)
2569                         *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn;
2570         }
2571         /* Set the mft record itself in use. */
2572         m->flags |= MFT_RECORD_IN_USE;
2573         if (S_ISDIR(mode))
2574                 m->flags |= MFT_RECORD_IS_DIRECTORY;
2575         flush_dcache_page(page);
2576         SetPageUptodate(page);
2577         if (base_ni) {
2578                 /*
2579                  * Setup the base mft record in the extent mft record.  This
2580                  * completes initialization of the allocated extent mft record
2581                  * and we can simply use it with map_extent_mft_record().
2582                  */
2583                 m->base_mft_record = MK_LE_MREF(base_ni->mft_no,
2584                                 base_ni->seq_no);
2585                 /*
2586                  * Allocate an extent inode structure for the new mft record,
2587                  * attach it to the base inode @base_ni and map, pin, and lock
2588                  * its, i.e. the allocated, mft record.
2589                  */
2590                 m = map_extent_mft_record(base_ni, bit, &ni);
2591                 if (IS_ERR(m)) {
2592                         ntfs_error(vol->sb, "Failed to map allocated extent "
2593                                         "mft record 0x%llx.", (long long)bit);
2594                         err = PTR_ERR(m);
2595                         /* Set the mft record itself not in use. */
2596                         m->flags &= cpu_to_le16(
2597                                         ~le16_to_cpu(MFT_RECORD_IN_USE));
2598                         flush_dcache_page(page);
2599                         /* Make sure the mft record is written out to disk. */
2600                         mark_ntfs_record_dirty(page, ofs);
2601                         unlock_page(page);
2602                         ntfs_unmap_page(page);
2603                         goto undo_mftbmp_alloc;
2604                 }
2605                 /*
2606                  * Make sure the allocated mft record is written out to disk.
2607                  * No need to set the inode dirty because the caller is going
2608                  * to do that anyway after finishing with the new extent mft
2609                  * record (e.g. at a minimum a new attribute will be added to
2610                  * the mft record.
2611                  */
2612                 mark_ntfs_record_dirty(page, ofs);
2613                 unlock_page(page);
2614                 /*
2615                  * Need to unmap the page since map_extent_mft_record() mapped
2616                  * it as well so we have it mapped twice at the moment.
2617                  */
2618                 ntfs_unmap_page(page);
2619         } else {
2620                 /*
2621                  * Allocate a new VFS inode and set it up.  NOTE: @vi->i_nlink
2622                  * is set to 1 but the mft record->link_count is 0.  The caller
2623                  * needs to bear this in mind.
2624                  */
2625                 vi = new_inode(vol->sb);
2626                 if (unlikely(!vi)) {
2627                         err = -ENOMEM;
2628                         /* Set the mft record itself not in use. */
2629                         m->flags &= cpu_to_le16(
2630                                         ~le16_to_cpu(MFT_RECORD_IN_USE));
2631                         flush_dcache_page(page);
2632                         /* Make sure the mft record is written out to disk. */
2633                         mark_ntfs_record_dirty(page, ofs);
2634                         unlock_page(page);
2635                         ntfs_unmap_page(page);
2636                         goto undo_mftbmp_alloc;
2637                 }
2638                 vi->i_ino = bit;
2639                 /*
2640                  * This is for checking whether an inode has changed w.r.t. a
2641                  * file so that the file can be updated if necessary (compare
2642                  * with f_version).
2643                  */
2644                 vi->i_version = 1;
2645 
2646                 /* The owner and group come from the ntfs volume. */
2647                 vi->i_uid = vol->uid;
2648                 vi->i_gid = vol->gid;
2649 
2650                 /* Initialize the ntfs specific part of @vi. */
2651                 ntfs_init_big_inode(vi);
2652                 ni = NTFS_I(vi);
2653                 /*
2654                  * Set the appropriate mode, attribute type, and name.  For
2655                  * directories, also setup the index values to the defaults.
2656                  */
2657                 if (S_ISDIR(mode)) {
2658                         vi->i_mode = S_IFDIR | S_IRWXUGO;
2659                         vi->i_mode &= ~vol->dmask;
2660 
2661                         NInoSetMstProtected(ni);
2662                         ni->type = AT_INDEX_ALLOCATION;
2663                         ni->name = I30;
2664                         ni->name_len = 4;
2665 
2666                         ni->itype.index.block_size = 4096;
2667                         ni->itype.index.block_size_bits = ntfs_ffs(4096) - 1;
2668                         ni->itype.index.collation_rule = COLLATION_FILE_NAME;
2669                         if (vol->cluster_size <= ni->itype.index.block_size) {
2670                                 ni->itype.index.vcn_size = vol->cluster_size;
2671                                 ni->itype.index.vcn_size_bits =
2672                                                 vol->cluster_size_bits;
2673                         } else {
2674                                 ni->itype.index.vcn_size = vol->sector_size;
2675                                 ni->itype.index.vcn_size_bits =
2676                                                 vol->sector_size_bits;
2677                         }
2678                 } else {
2679                         vi->i_mode = S_IFREG | S_IRWXUGO;
2680                         vi->i_mode &= ~vol->fmask;
2681 
2682                         ni->type = AT_DATA;
2683                         ni->name = NULL;
2684                         ni->name_len = 0;
2685                 }
2686                 if (IS_RDONLY(vi))
2687                         vi->i_mode &= ~S_IWUGO;
2688 
2689                 /* Set the inode times to the current time. */
2690                 vi->i_atime = vi->i_mtime = vi->i_ctime =
2691                         current_fs_time(vi->i_sb);
2692                 /*
2693                  * Set the file size to 0, the ntfs inode sizes are set to 0 by
2694                  * the call to ntfs_init_big_inode() below.
2695                  */
2696                 vi->i_size = 0;
2697                 vi->i_blocks = 0;
2698 
2699                 /* Set the sequence number. */
2700                 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
2701                 /*
2702                  * Manually map, pin, and lock the mft record as we already
2703                  * have its page mapped and it is very easy to do.
2704                  */
2705                 atomic_inc(&ni->count);
2706                 mutex_lock(&ni->mrec_lock);
2707                 ni->page = page;
2708                 ni->page_ofs = ofs;
2709                 /*
2710                  * Make sure the allocated mft record is written out to disk.
2711                  * NOTE: We do not set the ntfs inode dirty because this would
2712                  * fail in ntfs_write_inode() because the inode does not have a
2713                  * standard information attribute yet.  Also, there is no need
2714                  * to set the inode dirty because the caller is going to do
2715                  * that anyway after finishing with the new mft record (e.g. at
2716                  * a minimum some new attributes will be added to the mft
2717                  * record.
2718                  */
2719                 mark_ntfs_record_dirty(page, ofs);
2720                 unlock_page(page);
2721 
2722                 /* Add the inode to the inode hash for the superblock. */
2723                 insert_inode_hash(vi);
2724 
2725                 /* Update the default mft allocation position. */
2726                 vol->mft_data_pos = bit + 1;
2727         }
2728         /*
2729          * Return the opened, allocated inode of the allocated mft record as
2730          * well as the mapped, pinned, and locked mft record.
2731          */
2732         ntfs_debug("Returning opened, allocated %sinode 0x%llx.",
2733                         base_ni ? "extent " : "", (long long)bit);
2734         *mrec = m;
2735         return ni;
2736 undo_data_init:
2737         write_lock_irqsave(&mft_ni->size_lock, flags);
2738         mft_ni->initialized_size = old_data_initialized;
2739         i_size_write(vol->mft_ino, old_data_size);
2740         write_unlock_irqrestore(&mft_ni->size_lock, flags);
2741         goto undo_mftbmp_alloc_nolock;
2742 undo_mftbmp_alloc:
2743         down_write(&vol->mftbmp_lock);
2744 undo_mftbmp_alloc_nolock:
2745         if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) {
2746                 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2747                 NVolSetErrors(vol);
2748         }
2749         up_write(&vol->mftbmp_lock);
2750 err_out:
2751         return ERR_PTR(err);
2752 max_err_out:
2753         ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum "
2754                         "number of inodes (2^32) has already been reached.");
2755         up_write(&vol->mftbmp_lock);
2756         return ERR_PTR(-ENOSPC);
2757 }
2758 
2759 /**
2760  * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume
2761  * @ni:         ntfs inode of the mapped extent mft record to free
2762  * @m:          mapped extent mft record of the ntfs inode @ni
2763  *
2764  * Free the mapped extent mft record @m of the extent ntfs inode @ni.
2765  *
2766  * Note that this function unmaps the mft record and closes and destroys @ni
2767  * internally and hence you cannot use either @ni nor @m any more after this
2768  * function returns success.
2769  *
2770  * On success return 0 and on error return -errno.  @ni and @m are still valid
2771  * in this case and have not been freed.
2772  *
2773  * For some errors an error message is displayed and the success code 0 is
2774  * returned and the volume is then left dirty on umount.  This makes sense in
2775  * case we could not rollback the changes that were already done since the
2776  * caller no longer wants to reference this mft record so it does not matter to
2777  * the caller if something is wrong with it as long as it is properly detached
2778  * from the base inode.
2779  */
2780 int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m)
2781 {
2782         unsigned long mft_no = ni->mft_no;
2783         ntfs_volume *vol = ni->vol;
2784         ntfs_inode *base_ni;
2785         ntfs_inode **extent_nis;
2786         int i, err;
2787         le16 old_seq_no;
2788         u16 seq_no;
2789         
2790         BUG_ON(NInoAttr(ni));
2791         BUG_ON(ni->nr_extents != -1);
2792 
2793         mutex_lock(&ni->extent_lock);
2794         base_ni = ni->ext.base_ntfs_ino;
2795         mutex_unlock(&ni->extent_lock);
2796 
2797         BUG_ON(base_ni->nr_extents <= 0);
2798 
2799         ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n",
2800                         mft_no, base_ni->mft_no);
2801 
2802         mutex_lock(&base_ni->extent_lock);
2803 
2804         /* Make sure we are holding the only reference to the extent inode. */
2805         if (atomic_read(&ni->count) > 2) {
2806                 ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, "
2807                                 "not freeing.", base_ni->mft_no);
2808                 mutex_unlock(&base_ni->extent_lock);
2809                 return -EBUSY;
2810         }
2811 
2812         /* Dissociate the ntfs inode from the base inode. */
2813         extent_nis = base_ni->ext.extent_ntfs_inos;
2814         err = -ENOENT;
2815         for (i = 0; i < base_ni->nr_extents; i++) {
2816                 if (ni != extent_nis[i])
2817                         continue;
2818                 extent_nis += i;
2819                 base_ni->nr_extents--;
2820                 memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) *
2821                                 sizeof(ntfs_inode*));
2822                 err = 0;
2823                 break;
2824         }
2825 
2826         mutex_unlock(&base_ni->extent_lock);
2827 
2828         if (unlikely(err)) {
2829                 ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to "
2830                                 "its base inode 0x%lx.", mft_no,
2831                                 base_ni->mft_no);
2832                 BUG();
2833         }
2834 
2835         /*
2836          * The extent inode is no longer attached to the base inode so no one
2837          * can get a reference to it any more.
2838          */
2839 
2840         /* Mark the mft record as not in use. */
2841         m->flags &= ~MFT_RECORD_IN_USE;
2842 
2843         /* Increment the sequence number, skipping zero, if it is not zero. */
2844         old_seq_no = m->sequence_number;
2845         seq_no = le16_to_cpu(old_seq_no);
2846         if (seq_no == 0xffff)
2847                 seq_no = 1;
2848         else if (seq_no)
2849                 seq_no++;
2850         m->sequence_number = cpu_to_le16(seq_no);
2851 
2852         /*
2853          * Set the ntfs inode dirty and write it out.  We do not need to worry
2854          * about the base inode here since whatever caused the extent mft
2855          * record to be freed is guaranteed to do it already.
2856          */
2857         NInoSetDirty(ni);
2858         err = write_mft_record(ni, m, 0);
2859         if (unlikely(err)) {
2860                 ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not "
2861                                 "freeing.", mft_no);
2862                 goto rollback;
2863         }
2864 rollback_error:
2865         /* Unmap and throw away the now freed extent inode. */
2866         unmap_extent_mft_record(ni);
2867         ntfs_clear_extent_inode(ni);
2868 
2869         /* Clear the bit in the $MFT/$BITMAP corresponding to this record. */
2870         down_write(&vol->mftbmp_lock);
2871         err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no);
2872         up_write(&vol->mftbmp_lock);
2873         if (unlikely(err)) {
2874                 /*
2875                  * The extent inode is gone but we failed to deallocate it in
2876                  * the mft bitmap.  Just emit a warning and leave the volume
2877                  * dirty on umount.
2878                  */
2879                 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2880                 NVolSetErrors(vol);
2881         }
2882         return 0;
2883 rollback:
2884         /* Rollback what we did... */
2885         mutex_lock(&base_ni->extent_lock);
2886         extent_nis = base_ni->ext.extent_ntfs_inos;
2887         if (!(base_ni->nr_extents & 3)) {
2888                 int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*);
2889 
2890                 extent_nis = kmalloc(new_size, GFP_NOFS);
2891                 if (unlikely(!extent_nis)) {
2892                         ntfs_error(vol->sb, "Failed to allocate internal "
2893                                         "buffer during rollback.%s", es);
2894                         mutex_unlock(&base_ni->extent_lock);
2895                         NVolSetErrors(vol);
2896                         goto rollback_error;
2897                 }
2898                 if (base_ni->nr_extents) {
2899                         BUG_ON(!base_ni->ext.extent_ntfs_inos);
2900                         memcpy(extent_nis, base_ni->ext.extent_ntfs_inos,
2901                                         new_size - 4 * sizeof(ntfs_inode*));
2902                         kfree(base_ni->ext.extent_ntfs_inos);
2903                 }
2904                 base_ni->ext.extent_ntfs_inos = extent_nis;
2905         }
2906         m->flags |= MFT_RECORD_IN_USE;
2907         m->sequence_number = old_seq_no;
2908         extent_nis[base_ni->nr_extents++] = ni;
2909         mutex_unlock(&base_ni->extent_lock);
2910         mark_mft_record_dirty(ni);
2911         return err;
2912 }
2913 #endif /* NTFS_RW */
2914 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp