xref: /openbmc/linux/fs/ntfs/mft.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /**
2  * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project.
3  *
4  * Copyright (c) 2001-2005 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 	}
97 err_out:
98 	ni->page = NULL;
99 	ni->page_ofs = 0;
100 	return (void*)page;
101 }
102 
103 /**
104  * map_mft_record - map, pin and lock an mft record
105  * @ni:		ntfs inode whose MFT record to map
106  *
107  * First, take the mrec_lock semaphore. We might now be sleeping, while waiting
108  * for the semaphore if it was already locked by someone else.
109  *
110  * The page of the record is mapped using map_mft_record_page() before being
111  * returned to the caller.
112  *
113  * This in turn uses ntfs_map_page() to get the page containing the wanted mft
114  * record (it in turn calls read_cache_page() which reads it in from disk if
115  * necessary, increments the use count on the page so that it cannot disappear
116  * under us and returns a reference to the page cache page).
117  *
118  * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it
119  * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed
120  * and the post-read mst fixups on each mft record in the page have been
121  * performed, the page gets PG_uptodate set and PG_locked cleared (this is done
122  * in our asynchronous I/O completion handler end_buffer_read_mft_async()).
123  * ntfs_map_page() waits for PG_locked to become clear and checks if
124  * PG_uptodate is set and returns an error code if not. This provides
125  * sufficient protection against races when reading/using the page.
126  *
127  * However there is the write mapping to think about. Doing the above described
128  * checking here will be fine, because when initiating the write we will set
129  * PG_locked and clear PG_uptodate making sure nobody is touching the page
130  * contents. Doing the locking this way means that the commit to disk code in
131  * the page cache code paths is automatically sufficiently locked with us as
132  * we will not touch a page that has been locked or is not uptodate. The only
133  * locking problem then is them locking the page while we are accessing it.
134  *
135  * So that code will end up having to own the mrec_lock of all mft
136  * records/inodes present in the page before I/O can proceed. In that case we
137  * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be
138  * accessing anything without owning the mrec_lock semaphore. But we do need
139  * to use them because of the read_cache_page() invocation and the code becomes
140  * so much simpler this way that it is well worth it.
141  *
142  * The mft record is now ours and we return a pointer to it. You need to check
143  * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return
144  * the error code.
145  *
146  * NOTE: Caller is responsible for setting the mft record dirty before calling
147  * unmap_mft_record(). This is obviously only necessary if the caller really
148  * modified the mft record...
149  * Q: Do we want to recycle one of the VFS inode state bits instead?
150  * A: No, the inode ones mean we want to change the mft record, not we want to
151  * write it out.
152  */
153 MFT_RECORD *map_mft_record(ntfs_inode *ni)
154 {
155 	MFT_RECORD *m;
156 
157 	ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
158 
159 	/* Make sure the ntfs inode doesn't go away. */
160 	atomic_inc(&ni->count);
161 
162 	/* Serialize access to this mft record. */
163 	down(&ni->mrec_lock);
164 
165 	m = map_mft_record_page(ni);
166 	if (likely(!IS_ERR(m)))
167 		return m;
168 
169 	up(&ni->mrec_lock);
170 	atomic_dec(&ni->count);
171 	ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m));
172 	return m;
173 }
174 
175 /**
176  * unmap_mft_record_page - unmap the page in which a specific mft record resides
177  * @ni:		ntfs inode whose mft record page to unmap
178  *
179  * This unmaps the page in which the mft record of the ntfs inode @ni is
180  * situated and returns. This is a NOOP if highmem is not configured.
181  *
182  * The unmap happens via ntfs_unmap_page() which in turn decrements the use
183  * count on the page thus releasing it from the pinned state.
184  *
185  * We do not actually unmap the page from memory of course, as that will be
186  * done by the page cache code itself when memory pressure increases or
187  * whatever.
188  */
189 static inline void unmap_mft_record_page(ntfs_inode *ni)
190 {
191 	BUG_ON(!ni->page);
192 
193 	// TODO: If dirty, blah...
194 	ntfs_unmap_page(ni->page);
195 	ni->page = NULL;
196 	ni->page_ofs = 0;
197 	return;
198 }
199 
200 /**
201  * unmap_mft_record - release a mapped mft record
202  * @ni:		ntfs inode whose MFT record to unmap
203  *
204  * We release the page mapping and the mrec_lock mutex which unmaps the mft
205  * record and releases it for others to get hold of. We also release the ntfs
206  * inode by decrementing the ntfs inode reference count.
207  *
208  * NOTE: If caller has modified the mft record, it is imperative to set the mft
209  * record dirty BEFORE calling unmap_mft_record().
210  */
211 void unmap_mft_record(ntfs_inode *ni)
212 {
213 	struct page *page = ni->page;
214 
215 	BUG_ON(!page);
216 
217 	ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
218 
219 	unmap_mft_record_page(ni);
220 	up(&ni->mrec_lock);
221 	atomic_dec(&ni->count);
222 	/*
223 	 * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to
224 	 * ntfs_clear_extent_inode() in the extent inode case, and to the
225 	 * caller in the non-extent, yet pure ntfs inode case, to do the actual
226 	 * tear down of all structures and freeing of all allocated memory.
227 	 */
228 	return;
229 }
230 
231 /**
232  * map_extent_mft_record - load an extent inode and attach it to its base
233  * @base_ni:	base ntfs inode
234  * @mref:	mft reference of the extent inode to load
235  * @ntfs_ino:	on successful return, pointer to the ntfs_inode structure
236  *
237  * Load the extent mft record @mref and attach it to its base inode @base_ni.
238  * Return the mapped extent mft record if IS_ERR(result) is false.  Otherwise
239  * PTR_ERR(result) gives the negative error code.
240  *
241  * On successful return, @ntfs_ino contains a pointer to the ntfs_inode
242  * structure of the mapped extent inode.
243  */
244 MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref,
245 		ntfs_inode **ntfs_ino)
246 {
247 	MFT_RECORD *m;
248 	ntfs_inode *ni = NULL;
249 	ntfs_inode **extent_nis = NULL;
250 	int i;
251 	unsigned long mft_no = MREF(mref);
252 	u16 seq_no = MSEQNO(mref);
253 	BOOL destroy_ni = FALSE;
254 
255 	ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).",
256 			mft_no, base_ni->mft_no);
257 	/* Make sure the base ntfs inode doesn't go away. */
258 	atomic_inc(&base_ni->count);
259 	/*
260 	 * Check if this extent inode has already been added to the base inode,
261 	 * in which case just return it. If not found, add it to the base
262 	 * inode before returning it.
263 	 */
264 	down(&base_ni->extent_lock);
265 	if (base_ni->nr_extents > 0) {
266 		extent_nis = base_ni->ext.extent_ntfs_inos;
267 		for (i = 0; i < base_ni->nr_extents; i++) {
268 			if (mft_no != extent_nis[i]->mft_no)
269 				continue;
270 			ni = extent_nis[i];
271 			/* Make sure the ntfs inode doesn't go away. */
272 			atomic_inc(&ni->count);
273 			break;
274 		}
275 	}
276 	if (likely(ni != NULL)) {
277 		up(&base_ni->extent_lock);
278 		atomic_dec(&base_ni->count);
279 		/* We found the record; just have to map and return it. */
280 		m = map_mft_record(ni);
281 		/* map_mft_record() has incremented this on success. */
282 		atomic_dec(&ni->count);
283 		if (likely(!IS_ERR(m))) {
284 			/* Verify the sequence number. */
285 			if (likely(le16_to_cpu(m->sequence_number) == seq_no)) {
286 				ntfs_debug("Done 1.");
287 				*ntfs_ino = ni;
288 				return m;
289 			}
290 			unmap_mft_record(ni);
291 			ntfs_error(base_ni->vol->sb, "Found stale extent mft "
292 					"reference! Corrupt filesystem. "
293 					"Run chkdsk.");
294 			return ERR_PTR(-EIO);
295 		}
296 map_err_out:
297 		ntfs_error(base_ni->vol->sb, "Failed to map extent "
298 				"mft record, error code %ld.", -PTR_ERR(m));
299 		return m;
300 	}
301 	/* Record wasn't there. Get a new ntfs inode and initialize it. */
302 	ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no);
303 	if (unlikely(!ni)) {
304 		up(&base_ni->extent_lock);
305 		atomic_dec(&base_ni->count);
306 		return ERR_PTR(-ENOMEM);
307 	}
308 	ni->vol = base_ni->vol;
309 	ni->seq_no = seq_no;
310 	ni->nr_extents = -1;
311 	ni->ext.base_ntfs_ino = base_ni;
312 	/* Now map the record. */
313 	m = map_mft_record(ni);
314 	if (IS_ERR(m)) {
315 		up(&base_ni->extent_lock);
316 		atomic_dec(&base_ni->count);
317 		ntfs_clear_extent_inode(ni);
318 		goto map_err_out;
319 	}
320 	/* Verify the sequence number if it is present. */
321 	if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) {
322 		ntfs_error(base_ni->vol->sb, "Found stale extent mft "
323 				"reference! Corrupt filesystem. Run chkdsk.");
324 		destroy_ni = TRUE;
325 		m = ERR_PTR(-EIO);
326 		goto unm_err_out;
327 	}
328 	/* Attach extent inode to base inode, reallocating memory if needed. */
329 	if (!(base_ni->nr_extents & 3)) {
330 		ntfs_inode **tmp;
331 		int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *);
332 
333 		tmp = (ntfs_inode **)kmalloc(new_size, GFP_NOFS);
334 		if (unlikely(!tmp)) {
335 			ntfs_error(base_ni->vol->sb, "Failed to allocate "
336 					"internal buffer.");
337 			destroy_ni = TRUE;
338 			m = ERR_PTR(-ENOMEM);
339 			goto unm_err_out;
340 		}
341 		if (base_ni->nr_extents) {
342 			BUG_ON(!base_ni->ext.extent_ntfs_inos);
343 			memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size -
344 					4 * sizeof(ntfs_inode *));
345 			kfree(base_ni->ext.extent_ntfs_inos);
346 		}
347 		base_ni->ext.extent_ntfs_inos = tmp;
348 	}
349 	base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni;
350 	up(&base_ni->extent_lock);
351 	atomic_dec(&base_ni->count);
352 	ntfs_debug("Done 2.");
353 	*ntfs_ino = ni;
354 	return m;
355 unm_err_out:
356 	unmap_mft_record(ni);
357 	up(&base_ni->extent_lock);
358 	atomic_dec(&base_ni->count);
359 	/*
360 	 * If the extent inode was not attached to the base inode we need to
361 	 * release it or we will leak memory.
362 	 */
363 	if (destroy_ni)
364 		ntfs_clear_extent_inode(ni);
365 	return m;
366 }
367 
368 #ifdef NTFS_RW
369 
370 /**
371  * __mark_mft_record_dirty - set the mft record and the page containing it dirty
372  * @ni:		ntfs inode describing the mapped mft record
373  *
374  * Internal function.  Users should call mark_mft_record_dirty() instead.
375  *
376  * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni,
377  * as well as the page containing the mft record, dirty.  Also, mark the base
378  * vfs inode dirty.  This ensures that any changes to the mft record are
379  * written out to disk.
380  *
381  * NOTE:  We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES)
382  * on the base vfs inode, because even though file data may have been modified,
383  * it is dirty in the inode meta data rather than the data page cache of the
384  * inode, and thus there are no data pages that need writing out.  Therefore, a
385  * full mark_inode_dirty() is overkill.  A mark_inode_dirty_sync(), on the
386  * other hand, is not sufficient, because I_DIRTY_DATASYNC needs to be set to
387  * ensure ->write_inode is called from generic_osync_inode() and this needs to
388  * happen or the file data would not necessarily hit the device synchronously,
389  * even though the vfs inode has the O_SYNC flag set.  Also, I_DIRTY_DATASYNC
390  * simply "feels" better than just I_DIRTY_SYNC, since the file data has not
391  * actually hit the block device yet, which is not what I_DIRTY_SYNC on its own
392  * 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 	down(&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 	up(&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->mftmirr_ino->i_blkbits;
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 (unlikely(test_set_buffer_locked(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.  The caller should call make_bad_inode() on
654  * the base inode to ensure no more access happens to this inode.  We do not do
655  * it here as the caller may want to finish writing other extent mft records
656  * first to minimize on-disk metadata inconsistencies.
657  *
658  * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
659  * However, if the mft record has a counterpart in the mft mirror and @sync is
660  * true, we write the mft record, wait for i/o completion, and only then write
661  * the mft mirror copy.  This ensures that if the system crashes either the mft
662  * or the mft mirror will contain a self-consistent mft record @m.  If @sync is
663  * false on the other hand, we start i/o on both and then wait for completion
664  * on them.  This provides a speedup but no longer guarantees that you will end
665  * up with a self-consistent mft record in the case of a crash but if you asked
666  * for asynchronous writing you probably do not care about that anyway.
667  *
668  * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
669  * schedule i/o via ->writepage or do it via kntfsd or whatever.
670  */
671 int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync)
672 {
673 	ntfs_volume *vol = ni->vol;
674 	struct page *page = ni->page;
675 	unsigned char blocksize_bits = vol->mft_ino->i_blkbits;
676 	unsigned int blocksize = 1 << blocksize_bits;
677 	int max_bhs = vol->mft_record_size / blocksize;
678 	struct buffer_head *bhs[max_bhs];
679 	struct buffer_head *bh, *head;
680 	runlist_element *rl;
681 	unsigned int block_start, block_end, m_start, m_end;
682 	int i_bhs, nr_bhs, err = 0;
683 
684 	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
685 	BUG_ON(NInoAttr(ni));
686 	BUG_ON(!max_bhs);
687 	BUG_ON(!PageLocked(page));
688 	/*
689 	 * If the ntfs_inode is clean no need to do anything.  If it is dirty,
690 	 * mark it as clean now so that it can be redirtied later on if needed.
691 	 * There is no danger of races since the caller is holding the locks
692 	 * for the mft record @m and the page it is in.
693 	 */
694 	if (!NInoTestClearDirty(ni))
695 		goto done;
696 	bh = head = page_buffers(page);
697 	BUG_ON(!bh);
698 	rl = NULL;
699 	nr_bhs = 0;
700 	block_start = 0;
701 	m_start = ni->page_ofs;
702 	m_end = m_start + vol->mft_record_size;
703 	do {
704 		block_end = block_start + blocksize;
705 		/* If the buffer is outside the mft record, skip it. */
706 		if (block_end <= m_start)
707 			continue;
708 		if (unlikely(block_start >= m_end))
709 			break;
710 		/*
711 		 * If this block is not the first one in the record, we ignore
712 		 * the buffer's dirty state because we could have raced with a
713 		 * parallel mark_ntfs_record_dirty().
714 		 */
715 		if (block_start == m_start) {
716 			/* This block is the first one in the record. */
717 			if (!buffer_dirty(bh)) {
718 				BUG_ON(nr_bhs);
719 				/* Clean records are not written out. */
720 				break;
721 			}
722 		}
723 		/* Need to map the buffer if it is not mapped already. */
724 		if (unlikely(!buffer_mapped(bh))) {
725 			VCN vcn;
726 			LCN lcn;
727 			unsigned int vcn_ofs;
728 
729 			bh->b_bdev = vol->sb->s_bdev;
730 			/* Obtain the vcn and offset of the current block. */
731 			vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) +
732 					(block_start - m_start);
733 			vcn_ofs = vcn & vol->cluster_size_mask;
734 			vcn >>= vol->cluster_size_bits;
735 			if (!rl) {
736 				down_read(&NTFS_I(vol->mft_ino)->runlist.lock);
737 				rl = NTFS_I(vol->mft_ino)->runlist.rl;
738 				BUG_ON(!rl);
739 			}
740 			/* Seek to element containing target vcn. */
741 			while (rl->length && rl[1].vcn <= vcn)
742 				rl++;
743 			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
744 			/* For $MFT, only lcn >= 0 is a successful remap. */
745 			if (likely(lcn >= 0)) {
746 				/* Setup buffer head to correct block. */
747 				bh->b_blocknr = ((lcn <<
748 						vol->cluster_size_bits) +
749 						vcn_ofs) >> blocksize_bits;
750 				set_buffer_mapped(bh);
751 			} else {
752 				bh->b_blocknr = -1;
753 				ntfs_error(vol->sb, "Cannot write mft record "
754 						"0x%lx because its location "
755 						"on disk could not be "
756 						"determined (error code %lli).",
757 						ni->mft_no, (long long)lcn);
758 				err = -EIO;
759 			}
760 		}
761 		BUG_ON(!buffer_uptodate(bh));
762 		BUG_ON(!nr_bhs && (m_start != block_start));
763 		BUG_ON(nr_bhs >= max_bhs);
764 		bhs[nr_bhs++] = bh;
765 		BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
766 	} while (block_start = block_end, (bh = bh->b_this_page) != head);
767 	if (unlikely(rl))
768 		up_read(&NTFS_I(vol->mft_ino)->runlist.lock);
769 	if (!nr_bhs)
770 		goto done;
771 	if (unlikely(err))
772 		goto cleanup_out;
773 	/* Apply the mst protection fixups. */
774 	err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size);
775 	if (err) {
776 		ntfs_error(vol->sb, "Failed to apply mst fixups!");
777 		goto cleanup_out;
778 	}
779 	flush_dcache_mft_record_page(ni);
780 	/* Lock buffers and start synchronous write i/o on them. */
781 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
782 		struct buffer_head *tbh = bhs[i_bhs];
783 
784 		if (unlikely(test_set_buffer_locked(tbh)))
785 			BUG();
786 		BUG_ON(!buffer_uptodate(tbh));
787 		clear_buffer_dirty(tbh);
788 		get_bh(tbh);
789 		tbh->b_end_io = end_buffer_write_sync;
790 		submit_bh(WRITE, tbh);
791 	}
792 	/* Synchronize the mft mirror now if not @sync. */
793 	if (!sync && ni->mft_no < vol->mftmirr_size)
794 		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
795 	/* Wait on i/o completion of buffers. */
796 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
797 		struct buffer_head *tbh = bhs[i_bhs];
798 
799 		wait_on_buffer(tbh);
800 		if (unlikely(!buffer_uptodate(tbh))) {
801 			err = -EIO;
802 			/*
803 			 * Set the buffer uptodate so the page and buffer
804 			 * states do not become out of sync.
805 			 */
806 			if (PageUptodate(page))
807 				set_buffer_uptodate(tbh);
808 		}
809 	}
810 	/* If @sync, now synchronize the mft mirror. */
811 	if (sync && ni->mft_no < vol->mftmirr_size)
812 		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
813 	/* Remove the mst protection fixups again. */
814 	post_write_mst_fixup((NTFS_RECORD*)m);
815 	flush_dcache_mft_record_page(ni);
816 	if (unlikely(err)) {
817 		/* I/O error during writing.  This is really bad! */
818 		ntfs_error(vol->sb, "I/O error while writing mft record "
819 				"0x%lx!  Marking base inode as bad.  You "
820 				"should unmount the volume and run chkdsk.",
821 				ni->mft_no);
822 		goto err_out;
823 	}
824 done:
825 	ntfs_debug("Done.");
826 	return 0;
827 cleanup_out:
828 	/* Clean the buffers. */
829 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
830 		clear_buffer_dirty(bhs[i_bhs]);
831 err_out:
832 	/*
833 	 * Current state: all buffers are clean, unlocked, and uptodate.
834 	 * The caller should mark the base inode as bad so that no more i/o
835 	 * happens.  ->clear_inode() will still be invoked so all extent inodes
836 	 * and other allocated memory will be freed.
837 	 */
838 	if (err == -ENOMEM) {
839 		ntfs_error(vol->sb, "Not enough memory to write mft record.  "
840 				"Redirtying so the write is retried later.");
841 		mark_mft_record_dirty(ni);
842 		err = 0;
843 	} else
844 		NVolSetErrors(vol);
845 	return err;
846 }
847 
848 /**
849  * ntfs_may_write_mft_record - check if an mft record may be written out
850  * @vol:	[IN]  ntfs volume on which the mft record to check resides
851  * @mft_no:	[IN]  mft record number of the mft record to check
852  * @m:		[IN]  mapped mft record to check
853  * @locked_ni:	[OUT] caller has to unlock this ntfs inode if one is returned
854  *
855  * Check if the mapped (base or extent) mft record @m with mft record number
856  * @mft_no belonging to the ntfs volume @vol may be written out.  If necessary
857  * and possible the ntfs inode of the mft record is locked and the base vfs
858  * inode is pinned.  The locked ntfs inode is then returned in @locked_ni.  The
859  * caller is responsible for unlocking the ntfs inode and unpinning the base
860  * vfs inode.
861  *
862  * Return TRUE if the mft record may be written out and FALSE if not.
863  *
864  * The caller has locked the page and cleared the uptodate flag on it which
865  * means that we can safely write out any dirty mft records that do not have
866  * their inodes in icache as determined by ilookup5() as anyone
867  * opening/creating such an inode would block when attempting to map the mft
868  * record in read_cache_page() until we are finished with the write out.
869  *
870  * Here is a description of the tests we perform:
871  *
872  * If the inode is found in icache we know the mft record must be a base mft
873  * record.  If it is dirty, we do not write it and return FALSE as the vfs
874  * inode write paths will result in the access times being updated which would
875  * cause the base mft record to be redirtied and written out again.  (We know
876  * the access time update will modify the base mft record because Windows
877  * chkdsk complains if the standard information attribute is not in the base
878  * mft record.)
879  *
880  * If the inode is in icache and not dirty, we attempt to lock the mft record
881  * and if we find the lock was already taken, it is not safe to write the mft
882  * record and we return FALSE.
883  *
884  * If we manage to obtain the lock we have exclusive access to the mft record,
885  * which also allows us safe writeout of the mft record.  We then set
886  * @locked_ni to the locked ntfs inode and return TRUE.
887  *
888  * Note we cannot just lock the mft record and sleep while waiting for the lock
889  * because this would deadlock due to lock reversal (normally the mft record is
890  * locked before the page is locked but we already have the page locked here
891  * when we try to lock the mft record).
892  *
893  * If the inode is not in icache we need to perform further checks.
894  *
895  * If the mft record is not a FILE record or it is a base mft record, we can
896  * safely write it and return TRUE.
897  *
898  * We now know the mft record is an extent mft record.  We check if the inode
899  * corresponding to its base mft record is in icache and obtain a reference to
900  * it if it is.  If it is not, we can safely write it and return TRUE.
901  *
902  * We now have the base inode for the extent mft record.  We check if it has an
903  * ntfs inode for the extent mft record attached and if not it is safe to write
904  * the extent mft record and we return TRUE.
905  *
906  * The ntfs inode for the extent mft record is attached to the base inode so we
907  * attempt to lock the extent mft record and if we find the lock was already
908  * taken, it is not safe to write the extent mft record and we return FALSE.
909  *
910  * If we manage to obtain the lock we have exclusive access to the extent mft
911  * record, which also allows us safe writeout of the extent mft record.  We
912  * set the ntfs inode of the extent mft record clean and then set @locked_ni to
913  * the now locked ntfs inode and return TRUE.
914  *
915  * Note, the reason for actually writing dirty mft records here and not just
916  * relying on the vfs inode dirty code paths is that we can have mft records
917  * modified without them ever having actual inodes in memory.  Also we can have
918  * dirty mft records with clean ntfs inodes in memory.  None of the described
919  * cases would result in the dirty mft records being written out if we only
920  * relied on the vfs inode dirty code paths.  And these cases can really occur
921  * during allocation of new mft records and in particular when the
922  * initialized_size of the $MFT/$DATA attribute is extended and the new space
923  * is initialized using ntfs_mft_record_format().  The clean inode can then
924  * appear if the mft record is reused for a new inode before it got written
925  * out.
926  */
927 BOOL ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
928 		const MFT_RECORD *m, ntfs_inode **locked_ni)
929 {
930 	struct super_block *sb = vol->sb;
931 	struct inode *mft_vi = vol->mft_ino;
932 	struct inode *vi;
933 	ntfs_inode *ni, *eni, **extent_nis;
934 	int i;
935 	ntfs_attr na;
936 
937 	ntfs_debug("Entering for inode 0x%lx.", mft_no);
938 	/*
939 	 * Normally we do not return a locked inode so set @locked_ni to NULL.
940 	 */
941 	BUG_ON(!locked_ni);
942 	*locked_ni = NULL;
943 	/*
944 	 * Check if the inode corresponding to this mft record is in the VFS
945 	 * inode cache and obtain a reference to it if it is.
946 	 */
947 	ntfs_debug("Looking for inode 0x%lx in icache.", mft_no);
948 	na.mft_no = mft_no;
949 	na.name = NULL;
950 	na.name_len = 0;
951 	na.type = AT_UNUSED;
952 	/*
953 	 * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
954 	 * we get here for it rather often.
955 	 */
956 	if (!mft_no) {
957 		/* Balance the below iput(). */
958 		vi = igrab(mft_vi);
959 		BUG_ON(vi != mft_vi);
960 	} else {
961 		/*
962 		 * Have to use ilookup5_nowait() since ilookup5() waits for the
963 		 * inode lock which causes ntfs to deadlock when a concurrent
964 		 * inode write via the inode dirty code paths and the page
965 		 * dirty code path of the inode dirty code path when writing
966 		 * $MFT occurs.
967 		 */
968 		vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
969 	}
970 	if (vi) {
971 		ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
972 		/* The inode is in icache. */
973 		ni = NTFS_I(vi);
974 		/* Take a reference to the ntfs inode. */
975 		atomic_inc(&ni->count);
976 		/* If the inode is dirty, do not write this record. */
977 		if (NInoDirty(ni)) {
978 			ntfs_debug("Inode 0x%lx is dirty, do not write it.",
979 					mft_no);
980 			atomic_dec(&ni->count);
981 			iput(vi);
982 			return FALSE;
983 		}
984 		ntfs_debug("Inode 0x%lx is not dirty.", mft_no);
985 		/* The inode is not dirty, try to take the mft record lock. */
986 		if (unlikely(down_trylock(&ni->mrec_lock))) {
987 			ntfs_debug("Mft record 0x%lx is already locked, do "
988 					"not write it.", mft_no);
989 			atomic_dec(&ni->count);
990 			iput(vi);
991 			return FALSE;
992 		}
993 		ntfs_debug("Managed to lock mft record 0x%lx, write it.",
994 				mft_no);
995 		/*
996 		 * The write has to occur while we hold the mft record lock so
997 		 * return the locked ntfs inode.
998 		 */
999 		*locked_ni = ni;
1000 		return TRUE;
1001 	}
1002 	ntfs_debug("Inode 0x%lx is not in icache.", mft_no);
1003 	/* The inode is not in icache. */
1004 	/* Write the record if it is not a mft record (type "FILE"). */
1005 	if (!ntfs_is_mft_record(m->magic)) {
1006 		ntfs_debug("Mft record 0x%lx is not a FILE record, write it.",
1007 				mft_no);
1008 		return TRUE;
1009 	}
1010 	/* Write the mft record if it is a base inode. */
1011 	if (!m->base_mft_record) {
1012 		ntfs_debug("Mft record 0x%lx is a base record, write it.",
1013 				mft_no);
1014 		return TRUE;
1015 	}
1016 	/*
1017 	 * This is an extent mft record.  Check if the inode corresponding to
1018 	 * its base mft record is in icache and obtain a reference to it if it
1019 	 * is.
1020 	 */
1021 	na.mft_no = MREF_LE(m->base_mft_record);
1022 	ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
1023 			"inode 0x%lx in icache.", mft_no, na.mft_no);
1024 	if (!na.mft_no) {
1025 		/* Balance the below iput(). */
1026 		vi = igrab(mft_vi);
1027 		BUG_ON(vi != mft_vi);
1028 	} else
1029 		vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
1030 				&na);
1031 	if (!vi) {
1032 		/*
1033 		 * The base inode is not in icache, write this extent mft
1034 		 * record.
1035 		 */
1036 		ntfs_debug("Base inode 0x%lx is not in icache, write the "
1037 				"extent record.", na.mft_no);
1038 		return TRUE;
1039 	}
1040 	ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no);
1041 	/*
1042 	 * The base inode is in icache.  Check if it has the extent inode
1043 	 * corresponding to this extent mft record attached.
1044 	 */
1045 	ni = NTFS_I(vi);
1046 	down(&ni->extent_lock);
1047 	if (ni->nr_extents <= 0) {
1048 		/*
1049 		 * The base inode has no attached extent inodes, write this
1050 		 * extent mft record.
1051 		 */
1052 		up(&ni->extent_lock);
1053 		iput(vi);
1054 		ntfs_debug("Base inode 0x%lx has no attached extent inodes, "
1055 				"write the extent record.", na.mft_no);
1056 		return TRUE;
1057 	}
1058 	/* Iterate over the attached extent inodes. */
1059 	extent_nis = ni->ext.extent_ntfs_inos;
1060 	for (eni = NULL, i = 0; i < ni->nr_extents; ++i) {
1061 		if (mft_no == extent_nis[i]->mft_no) {
1062 			/*
1063 			 * Found the extent inode corresponding to this extent
1064 			 * mft record.
1065 			 */
1066 			eni = extent_nis[i];
1067 			break;
1068 		}
1069 	}
1070 	/*
1071 	 * If the extent inode was not attached to the base inode, write this
1072 	 * extent mft record.
1073 	 */
1074 	if (!eni) {
1075 		up(&ni->extent_lock);
1076 		iput(vi);
1077 		ntfs_debug("Extent inode 0x%lx is not attached to its base "
1078 				"inode 0x%lx, write the extent record.",
1079 				mft_no, na.mft_no);
1080 		return TRUE;
1081 	}
1082 	ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.",
1083 			mft_no, na.mft_no);
1084 	/* Take a reference to the extent ntfs inode. */
1085 	atomic_inc(&eni->count);
1086 	up(&ni->extent_lock);
1087 	/*
1088 	 * Found the extent inode coresponding to this extent mft record.
1089 	 * Try to take the mft record lock.
1090 	 */
1091 	if (unlikely(down_trylock(&eni->mrec_lock))) {
1092 		atomic_dec(&eni->count);
1093 		iput(vi);
1094 		ntfs_debug("Extent mft record 0x%lx is already locked, do "
1095 				"not write it.", mft_no);
1096 		return FALSE;
1097 	}
1098 	ntfs_debug("Managed to lock extent mft record 0x%lx, write it.",
1099 			mft_no);
1100 	if (NInoTestClearDirty(eni))
1101 		ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.",
1102 				mft_no);
1103 	/*
1104 	 * The write has to occur while we hold the mft record lock so return
1105 	 * the locked extent ntfs inode.
1106 	 */
1107 	*locked_ni = eni;
1108 	return TRUE;
1109 }
1110 
1111 static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
1112 		"chkdsk.";
1113 
1114 /**
1115  * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name
1116  * @vol:	volume on which to search for a free mft record
1117  * @base_ni:	open base inode if allocating an extent mft record or NULL
1118  *
1119  * Search for a free mft record in the mft bitmap attribute on the ntfs volume
1120  * @vol.
1121  *
1122  * If @base_ni is NULL start the search at the default allocator position.
1123  *
1124  * If @base_ni is not NULL start the search at the mft record after the base
1125  * mft record @base_ni.
1126  *
1127  * Return the free mft record on success and -errno on error.  An error code of
1128  * -ENOSPC means that there are no free mft records in the currently
1129  * initialized mft bitmap.
1130  *
1131  * Locking: Caller must hold vol->mftbmp_lock for writing.
1132  */
1133 static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol,
1134 		ntfs_inode *base_ni)
1135 {
1136 	s64 pass_end, ll, data_pos, pass_start, ofs, bit;
1137 	unsigned long flags;
1138 	struct address_space *mftbmp_mapping;
1139 	u8 *buf, *byte;
1140 	struct page *page;
1141 	unsigned int page_ofs, size;
1142 	u8 pass, b;
1143 
1144 	ntfs_debug("Searching for free mft record in the currently "
1145 			"initialized mft bitmap.");
1146 	mftbmp_mapping = vol->mftbmp_ino->i_mapping;
1147 	/*
1148 	 * Set the end of the pass making sure we do not overflow the mft
1149 	 * bitmap.
1150 	 */
1151 	read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags);
1152 	pass_end = NTFS_I(vol->mft_ino)->allocated_size >>
1153 			vol->mft_record_size_bits;
1154 	read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags);
1155 	read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1156 	ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3;
1157 	read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1158 	if (pass_end > ll)
1159 		pass_end = ll;
1160 	pass = 1;
1161 	if (!base_ni)
1162 		data_pos = vol->mft_data_pos;
1163 	else
1164 		data_pos = base_ni->mft_no + 1;
1165 	if (data_pos < 24)
1166 		data_pos = 24;
1167 	if (data_pos >= pass_end) {
1168 		data_pos = 24;
1169 		pass = 2;
1170 		/* This happens on a freshly formatted volume. */
1171 		if (data_pos >= pass_end)
1172 			return -ENOSPC;
1173 	}
1174 	pass_start = data_pos;
1175 	ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, "
1176 			"pass_end 0x%llx, data_pos 0x%llx.", pass,
1177 			(long long)pass_start, (long long)pass_end,
1178 			(long long)data_pos);
1179 	/* Loop until a free mft record is found. */
1180 	for (; pass <= 2;) {
1181 		/* Cap size to pass_end. */
1182 		ofs = data_pos >> 3;
1183 		page_ofs = ofs & ~PAGE_CACHE_MASK;
1184 		size = PAGE_CACHE_SIZE - page_ofs;
1185 		ll = ((pass_end + 7) >> 3) - ofs;
1186 		if (size > ll)
1187 			size = ll;
1188 		size <<= 3;
1189 		/*
1190 		 * If we are still within the active pass, search the next page
1191 		 * for a zero bit.
1192 		 */
1193 		if (size) {
1194 			page = ntfs_map_page(mftbmp_mapping,
1195 					ofs >> PAGE_CACHE_SHIFT);
1196 			if (unlikely(IS_ERR(page))) {
1197 				ntfs_error(vol->sb, "Failed to read mft "
1198 						"bitmap, aborting.");
1199 				return PTR_ERR(page);
1200 			}
1201 			buf = (u8*)page_address(page) + page_ofs;
1202 			bit = data_pos & 7;
1203 			data_pos &= ~7ull;
1204 			ntfs_debug("Before inner for loop: size 0x%x, "
1205 					"data_pos 0x%llx, bit 0x%llx", size,
1206 					(long long)data_pos, (long long)bit);
1207 			for (; bit < size && data_pos + bit < pass_end;
1208 					bit &= ~7ull, bit += 8) {
1209 				byte = buf + (bit >> 3);
1210 				if (*byte == 0xff)
1211 					continue;
1212 				b = ffz((unsigned long)*byte);
1213 				if (b < 8 && b >= (bit & 7)) {
1214 					ll = data_pos + (bit & ~7ull) + b;
1215 					if (unlikely(ll > (1ll << 32))) {
1216 						ntfs_unmap_page(page);
1217 						return -ENOSPC;
1218 					}
1219 					*byte |= 1 << b;
1220 					flush_dcache_page(page);
1221 					set_page_dirty(page);
1222 					ntfs_unmap_page(page);
1223 					ntfs_debug("Done.  (Found and "
1224 							"allocated mft record "
1225 							"0x%llx.)",
1226 							(long long)ll);
1227 					return ll;
1228 				}
1229 			}
1230 			ntfs_debug("After inner for loop: size 0x%x, "
1231 					"data_pos 0x%llx, bit 0x%llx", size,
1232 					(long long)data_pos, (long long)bit);
1233 			data_pos += size;
1234 			ntfs_unmap_page(page);
1235 			/*
1236 			 * If the end of the pass has not been reached yet,
1237 			 * continue searching the mft bitmap for a zero bit.
1238 			 */
1239 			if (data_pos < pass_end)
1240 				continue;
1241 		}
1242 		/* Do the next pass. */
1243 		if (++pass == 2) {
1244 			/*
1245 			 * Starting the second pass, in which we scan the first
1246 			 * part of the zone which we omitted earlier.
1247 			 */
1248 			pass_end = pass_start;
1249 			data_pos = pass_start = 24;
1250 			ntfs_debug("pass %i, pass_start 0x%llx, pass_end "
1251 					"0x%llx.", pass, (long long)pass_start,
1252 					(long long)pass_end);
1253 			if (data_pos >= pass_end)
1254 				break;
1255 		}
1256 	}
1257 	/* No free mft records in currently initialized mft bitmap. */
1258 	ntfs_debug("Done.  (No free mft records left in currently initialized "
1259 			"mft bitmap.)");
1260 	return -ENOSPC;
1261 }
1262 
1263 /**
1264  * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster
1265  * @vol:	volume on which to extend the mft bitmap attribute
1266  *
1267  * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster.
1268  *
1269  * Note: Only changes allocated_size, i.e. does not touch initialized_size or
1270  * data_size.
1271  *
1272  * Return 0 on success and -errno on error.
1273  *
1274  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1275  *	    - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for
1276  *	      writing and releases it before returning.
1277  *	    - This function takes vol->lcnbmp_lock for writing and releases it
1278  *	      before returning.
1279  */
1280 static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol)
1281 {
1282 	LCN lcn;
1283 	s64 ll;
1284 	unsigned long flags;
1285 	struct page *page;
1286 	ntfs_inode *mft_ni, *mftbmp_ni;
1287 	runlist_element *rl, *rl2 = NULL;
1288 	ntfs_attr_search_ctx *ctx = NULL;
1289 	MFT_RECORD *mrec;
1290 	ATTR_RECORD *a = NULL;
1291 	int ret, mp_size;
1292 	u32 old_alen = 0;
1293 	u8 *b, tb;
1294 	struct {
1295 		u8 added_cluster:1;
1296 		u8 added_run:1;
1297 		u8 mp_rebuilt:1;
1298 	} status = { 0, 0, 0 };
1299 
1300 	ntfs_debug("Extending mft bitmap allocation.");
1301 	mft_ni = NTFS_I(vol->mft_ino);
1302 	mftbmp_ni = NTFS_I(vol->mftbmp_ino);
1303 	/*
1304 	 * Determine the last lcn of the mft bitmap.  The allocated size of the
1305 	 * mft bitmap cannot be zero so we are ok to do this.
1306 	 */
1307 	down_write(&mftbmp_ni->runlist.lock);
1308 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1309 	ll = mftbmp_ni->allocated_size;
1310 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1311 	rl = ntfs_attr_find_vcn_nolock(mftbmp_ni,
1312 			(ll - 1) >> vol->cluster_size_bits, NULL);
1313 	if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1314 		up_write(&mftbmp_ni->runlist.lock);
1315 		ntfs_error(vol->sb, "Failed to determine last allocated "
1316 				"cluster of mft bitmap attribute.");
1317 		if (!IS_ERR(rl))
1318 			ret = -EIO;
1319 		else
1320 			ret = PTR_ERR(rl);
1321 		return ret;
1322 	}
1323 	lcn = rl->lcn + rl->length;
1324 	ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.",
1325 			(long long)lcn);
1326 	/*
1327 	 * Attempt to get the cluster following the last allocated cluster by
1328 	 * hand as it may be in the MFT zone so the allocator would not give it
1329 	 * to us.
1330 	 */
1331 	ll = lcn >> 3;
1332 	page = ntfs_map_page(vol->lcnbmp_ino->i_mapping,
1333 			ll >> PAGE_CACHE_SHIFT);
1334 	if (IS_ERR(page)) {
1335 		up_write(&mftbmp_ni->runlist.lock);
1336 		ntfs_error(vol->sb, "Failed to read from lcn bitmap.");
1337 		return PTR_ERR(page);
1338 	}
1339 	b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK);
1340 	tb = 1 << (lcn & 7ull);
1341 	down_write(&vol->lcnbmp_lock);
1342 	if (*b != 0xff && !(*b & tb)) {
1343 		/* Next cluster is free, allocate it. */
1344 		*b |= tb;
1345 		flush_dcache_page(page);
1346 		set_page_dirty(page);
1347 		up_write(&vol->lcnbmp_lock);
1348 		ntfs_unmap_page(page);
1349 		/* Update the mft bitmap runlist. */
1350 		rl->length++;
1351 		rl[1].vcn++;
1352 		status.added_cluster = 1;
1353 		ntfs_debug("Appending one cluster to mft bitmap.");
1354 	} else {
1355 		up_write(&vol->lcnbmp_lock);
1356 		ntfs_unmap_page(page);
1357 		/* Allocate a cluster from the DATA_ZONE. */
1358 		rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE,
1359 				TRUE);
1360 		if (IS_ERR(rl2)) {
1361 			up_write(&mftbmp_ni->runlist.lock);
1362 			ntfs_error(vol->sb, "Failed to allocate a cluster for "
1363 					"the mft bitmap.");
1364 			return PTR_ERR(rl2);
1365 		}
1366 		rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2);
1367 		if (IS_ERR(rl)) {
1368 			up_write(&mftbmp_ni->runlist.lock);
1369 			ntfs_error(vol->sb, "Failed to merge runlists for mft "
1370 					"bitmap.");
1371 			if (ntfs_cluster_free_from_rl(vol, rl2)) {
1372 				ntfs_error(vol->sb, "Failed to dealocate "
1373 						"allocated cluster.%s", es);
1374 				NVolSetErrors(vol);
1375 			}
1376 			ntfs_free(rl2);
1377 			return PTR_ERR(rl);
1378 		}
1379 		mftbmp_ni->runlist.rl = rl;
1380 		status.added_run = 1;
1381 		ntfs_debug("Adding one run to mft bitmap.");
1382 		/* Find the last run in the new runlist. */
1383 		for (; rl[1].length; rl++)
1384 			;
1385 	}
1386 	/*
1387 	 * Update the attribute record as well.  Note: @rl is the last
1388 	 * (non-terminator) runlist element of mft bitmap.
1389 	 */
1390 	mrec = map_mft_record(mft_ni);
1391 	if (IS_ERR(mrec)) {
1392 		ntfs_error(vol->sb, "Failed to map mft record.");
1393 		ret = PTR_ERR(mrec);
1394 		goto undo_alloc;
1395 	}
1396 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1397 	if (unlikely(!ctx)) {
1398 		ntfs_error(vol->sb, "Failed to get search context.");
1399 		ret = -ENOMEM;
1400 		goto undo_alloc;
1401 	}
1402 	ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1403 			mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1404 			0, ctx);
1405 	if (unlikely(ret)) {
1406 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1407 				"mft bitmap attribute.");
1408 		if (ret == -ENOENT)
1409 			ret = -EIO;
1410 		goto undo_alloc;
1411 	}
1412 	a = ctx->attr;
1413 	ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1414 	/* Search back for the previous last allocated cluster of mft bitmap. */
1415 	for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) {
1416 		if (ll >= rl2->vcn)
1417 			break;
1418 	}
1419 	BUG_ON(ll < rl2->vcn);
1420 	BUG_ON(ll >= rl2->vcn + rl2->length);
1421 	/* Get the size for the new mapping pairs array for this extent. */
1422 	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1423 	if (unlikely(mp_size <= 0)) {
1424 		ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1425 				"mft bitmap attribute extent.");
1426 		ret = mp_size;
1427 		if (!ret)
1428 			ret = -EIO;
1429 		goto undo_alloc;
1430 	}
1431 	/* Expand the attribute record if necessary. */
1432 	old_alen = le32_to_cpu(a->length);
1433 	ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1434 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1435 	if (unlikely(ret)) {
1436 		if (ret != -ENOSPC) {
1437 			ntfs_error(vol->sb, "Failed to resize attribute "
1438 					"record for mft bitmap attribute.");
1439 			goto undo_alloc;
1440 		}
1441 		// TODO: Deal with this by moving this extent to a new mft
1442 		// record or by starting a new extent in a new mft record or by
1443 		// moving other attributes out of this mft record.
1444 		// Note: It will need to be a special mft record and if none of
1445 		// those are available it gets rather complicated...
1446 		ntfs_error(vol->sb, "Not enough space in this mft record to "
1447 				"accomodate extended mft bitmap attribute "
1448 				"extent.  Cannot handle this yet.");
1449 		ret = -EOPNOTSUPP;
1450 		goto undo_alloc;
1451 	}
1452 	status.mp_rebuilt = 1;
1453 	/* Generate the mapping pairs array directly into the attr record. */
1454 	ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1455 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1456 			mp_size, rl2, ll, -1, NULL);
1457 	if (unlikely(ret)) {
1458 		ntfs_error(vol->sb, "Failed to build mapping pairs array for "
1459 				"mft bitmap attribute.");
1460 		goto undo_alloc;
1461 	}
1462 	/* Update the highest_vcn. */
1463 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1464 	/*
1465 	 * We now have extended the mft bitmap allocated_size by one cluster.
1466 	 * Reflect this in the ntfs_inode structure and the attribute record.
1467 	 */
1468 	if (a->data.non_resident.lowest_vcn) {
1469 		/*
1470 		 * We are not in the first attribute extent, switch to it, but
1471 		 * first ensure the changes will make it to disk later.
1472 		 */
1473 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1474 		mark_mft_record_dirty(ctx->ntfs_ino);
1475 		ntfs_attr_reinit_search_ctx(ctx);
1476 		ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1477 				mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL,
1478 				0, ctx);
1479 		if (unlikely(ret)) {
1480 			ntfs_error(vol->sb, "Failed to find first attribute "
1481 					"extent of mft bitmap attribute.");
1482 			goto restore_undo_alloc;
1483 		}
1484 		a = ctx->attr;
1485 	}
1486 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1487 	mftbmp_ni->allocated_size += vol->cluster_size;
1488 	a->data.non_resident.allocated_size =
1489 			cpu_to_sle64(mftbmp_ni->allocated_size);
1490 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1491 	/* Ensure the changes make it to disk. */
1492 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1493 	mark_mft_record_dirty(ctx->ntfs_ino);
1494 	ntfs_attr_put_search_ctx(ctx);
1495 	unmap_mft_record(mft_ni);
1496 	up_write(&mftbmp_ni->runlist.lock);
1497 	ntfs_debug("Done.");
1498 	return 0;
1499 restore_undo_alloc:
1500 	ntfs_attr_reinit_search_ctx(ctx);
1501 	if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1502 			mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1503 			0, ctx)) {
1504 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1505 				"mft bitmap attribute.%s", es);
1506 		write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1507 		mftbmp_ni->allocated_size += vol->cluster_size;
1508 		write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1509 		ntfs_attr_put_search_ctx(ctx);
1510 		unmap_mft_record(mft_ni);
1511 		up_write(&mftbmp_ni->runlist.lock);
1512 		/*
1513 		 * The only thing that is now wrong is ->allocated_size of the
1514 		 * base attribute extent which chkdsk should be able to fix.
1515 		 */
1516 		NVolSetErrors(vol);
1517 		return ret;
1518 	}
1519 	a = ctx->attr;
1520 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2);
1521 undo_alloc:
1522 	if (status.added_cluster) {
1523 		/* Truncate the last run in the runlist by one cluster. */
1524 		rl->length--;
1525 		rl[1].vcn--;
1526 	} else if (status.added_run) {
1527 		lcn = rl->lcn;
1528 		/* Remove the last run from the runlist. */
1529 		rl->lcn = rl[1].lcn;
1530 		rl->length = 0;
1531 	}
1532 	/* Deallocate the cluster. */
1533 	down_write(&vol->lcnbmp_lock);
1534 	if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1535 		ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es);
1536 		NVolSetErrors(vol);
1537 	}
1538 	up_write(&vol->lcnbmp_lock);
1539 	if (status.mp_rebuilt) {
1540 		if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1541 				a->data.non_resident.mapping_pairs_offset),
1542 				old_alen - le16_to_cpu(
1543 				a->data.non_resident.mapping_pairs_offset),
1544 				rl2, ll, -1, NULL)) {
1545 			ntfs_error(vol->sb, "Failed to restore mapping pairs "
1546 					"array.%s", es);
1547 			NVolSetErrors(vol);
1548 		}
1549 		if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1550 			ntfs_error(vol->sb, "Failed to restore attribute "
1551 					"record.%s", es);
1552 			NVolSetErrors(vol);
1553 		}
1554 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1555 		mark_mft_record_dirty(ctx->ntfs_ino);
1556 	}
1557 	if (ctx)
1558 		ntfs_attr_put_search_ctx(ctx);
1559 	if (!IS_ERR(mrec))
1560 		unmap_mft_record(mft_ni);
1561 	up_write(&mftbmp_ni->runlist.lock);
1562 	return ret;
1563 }
1564 
1565 /**
1566  * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data
1567  * @vol:	volume on which to extend the mft bitmap attribute
1568  *
1569  * Extend the initialized portion of the mft bitmap attribute on the ntfs
1570  * volume @vol by 8 bytes.
1571  *
1572  * Note:  Only changes initialized_size and data_size, i.e. requires that
1573  * allocated_size is big enough to fit the new initialized_size.
1574  *
1575  * Return 0 on success and -error on error.
1576  *
1577  * Locking: Caller must hold vol->mftbmp_lock for writing.
1578  */
1579 static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol)
1580 {
1581 	s64 old_data_size, old_initialized_size;
1582 	unsigned long flags;
1583 	struct inode *mftbmp_vi;
1584 	ntfs_inode *mft_ni, *mftbmp_ni;
1585 	ntfs_attr_search_ctx *ctx;
1586 	MFT_RECORD *mrec;
1587 	ATTR_RECORD *a;
1588 	int ret;
1589 
1590 	ntfs_debug("Extending mft bitmap initiailized (and data) size.");
1591 	mft_ni = NTFS_I(vol->mft_ino);
1592 	mftbmp_vi = vol->mftbmp_ino;
1593 	mftbmp_ni = NTFS_I(mftbmp_vi);
1594 	/* Get the attribute record. */
1595 	mrec = map_mft_record(mft_ni);
1596 	if (IS_ERR(mrec)) {
1597 		ntfs_error(vol->sb, "Failed to map mft record.");
1598 		return PTR_ERR(mrec);
1599 	}
1600 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1601 	if (unlikely(!ctx)) {
1602 		ntfs_error(vol->sb, "Failed to get search context.");
1603 		ret = -ENOMEM;
1604 		goto unm_err_out;
1605 	}
1606 	ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1607 			mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx);
1608 	if (unlikely(ret)) {
1609 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
1610 				"mft bitmap attribute.");
1611 		if (ret == -ENOENT)
1612 			ret = -EIO;
1613 		goto put_err_out;
1614 	}
1615 	a = ctx->attr;
1616 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1617 	old_data_size = i_size_read(mftbmp_vi);
1618 	old_initialized_size = mftbmp_ni->initialized_size;
1619 	/*
1620 	 * We can simply update the initialized_size before filling the space
1621 	 * with zeroes because the caller is holding the mft bitmap lock for
1622 	 * writing which ensures that no one else is trying to access the data.
1623 	 */
1624 	mftbmp_ni->initialized_size += 8;
1625 	a->data.non_resident.initialized_size =
1626 			cpu_to_sle64(mftbmp_ni->initialized_size);
1627 	if (mftbmp_ni->initialized_size > old_data_size) {
1628 		i_size_write(mftbmp_vi, mftbmp_ni->initialized_size);
1629 		a->data.non_resident.data_size =
1630 				cpu_to_sle64(mftbmp_ni->initialized_size);
1631 	}
1632 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1633 	/* Ensure the changes make it to disk. */
1634 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1635 	mark_mft_record_dirty(ctx->ntfs_ino);
1636 	ntfs_attr_put_search_ctx(ctx);
1637 	unmap_mft_record(mft_ni);
1638 	/* Initialize the mft bitmap attribute value with zeroes. */
1639 	ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0);
1640 	if (likely(!ret)) {
1641 		ntfs_debug("Done.  (Wrote eight initialized bytes to mft "
1642 				"bitmap.");
1643 		return 0;
1644 	}
1645 	ntfs_error(vol->sb, "Failed to write to mft bitmap.");
1646 	/* Try to recover from the error. */
1647 	mrec = map_mft_record(mft_ni);
1648 	if (IS_ERR(mrec)) {
1649 		ntfs_error(vol->sb, "Failed to map mft record.%s", es);
1650 		NVolSetErrors(vol);
1651 		return ret;
1652 	}
1653 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1654 	if (unlikely(!ctx)) {
1655 		ntfs_error(vol->sb, "Failed to get search context.%s", es);
1656 		NVolSetErrors(vol);
1657 		goto unm_err_out;
1658 	}
1659 	if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1660 			mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) {
1661 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
1662 				"mft bitmap attribute.%s", es);
1663 		NVolSetErrors(vol);
1664 put_err_out:
1665 		ntfs_attr_put_search_ctx(ctx);
1666 unm_err_out:
1667 		unmap_mft_record(mft_ni);
1668 		goto err_out;
1669 	}
1670 	a = ctx->attr;
1671 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1672 	mftbmp_ni->initialized_size = old_initialized_size;
1673 	a->data.non_resident.initialized_size =
1674 			cpu_to_sle64(old_initialized_size);
1675 	if (i_size_read(mftbmp_vi) != old_data_size) {
1676 		i_size_write(mftbmp_vi, old_data_size);
1677 		a->data.non_resident.data_size = cpu_to_sle64(old_data_size);
1678 	}
1679 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1680 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1681 	mark_mft_record_dirty(ctx->ntfs_ino);
1682 	ntfs_attr_put_search_ctx(ctx);
1683 	unmap_mft_record(mft_ni);
1684 #ifdef DEBUG
1685 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1686 	ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, "
1687 			"data_size 0x%llx, initialized_size 0x%llx.",
1688 			(long long)mftbmp_ni->allocated_size,
1689 			(long long)i_size_read(mftbmp_vi),
1690 			(long long)mftbmp_ni->initialized_size);
1691 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1692 #endif /* DEBUG */
1693 err_out:
1694 	return ret;
1695 }
1696 
1697 /**
1698  * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute
1699  * @vol:	volume on which to extend the mft data attribute
1700  *
1701  * Extend the mft data attribute on the ntfs volume @vol by 16 mft records
1702  * worth of clusters or if not enough space for this by one mft record worth
1703  * of clusters.
1704  *
1705  * Note:  Only changes allocated_size, i.e. does not touch initialized_size or
1706  * data_size.
1707  *
1708  * Return 0 on success and -errno on error.
1709  *
1710  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1711  *	    - This function takes NTFS_I(vol->mft_ino)->runlist.lock for
1712  *	      writing and releases it before returning.
1713  *	    - This function calls functions which take vol->lcnbmp_lock for
1714  *	      writing and release it before returning.
1715  */
1716 static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol)
1717 {
1718 	LCN lcn;
1719 	VCN old_last_vcn;
1720 	s64 min_nr, nr, ll;
1721 	unsigned long flags;
1722 	ntfs_inode *mft_ni;
1723 	runlist_element *rl, *rl2;
1724 	ntfs_attr_search_ctx *ctx = NULL;
1725 	MFT_RECORD *mrec;
1726 	ATTR_RECORD *a = NULL;
1727 	int ret, mp_size;
1728 	u32 old_alen = 0;
1729 	BOOL mp_rebuilt = FALSE;
1730 
1731 	ntfs_debug("Extending mft data allocation.");
1732 	mft_ni = NTFS_I(vol->mft_ino);
1733 	/*
1734 	 * Determine the preferred allocation location, i.e. the last lcn of
1735 	 * the mft data attribute.  The allocated size of the mft data
1736 	 * attribute cannot be zero so we are ok to do this.
1737 	 */
1738 	down_write(&mft_ni->runlist.lock);
1739 	read_lock_irqsave(&mft_ni->size_lock, flags);
1740 	ll = mft_ni->allocated_size;
1741 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
1742 	rl = ntfs_attr_find_vcn_nolock(mft_ni,
1743 			(ll - 1) >> vol->cluster_size_bits, NULL);
1744 	if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1745 		up_write(&mft_ni->runlist.lock);
1746 		ntfs_error(vol->sb, "Failed to determine last allocated "
1747 				"cluster of mft data attribute.");
1748 		if (!IS_ERR(rl))
1749 			ret = -EIO;
1750 		else
1751 			ret = PTR_ERR(rl);
1752 		return ret;
1753 	}
1754 	lcn = rl->lcn + rl->length;
1755 	ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn);
1756 	/* Minimum allocation is one mft record worth of clusters. */
1757 	min_nr = vol->mft_record_size >> vol->cluster_size_bits;
1758 	if (!min_nr)
1759 		min_nr = 1;
1760 	/* Want to allocate 16 mft records worth of clusters. */
1761 	nr = vol->mft_record_size << 4 >> vol->cluster_size_bits;
1762 	if (!nr)
1763 		nr = min_nr;
1764 	/* Ensure we do not go above 2^32-1 mft records. */
1765 	read_lock_irqsave(&mft_ni->size_lock, flags);
1766 	ll = mft_ni->allocated_size;
1767 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
1768 	if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1769 			vol->mft_record_size_bits >= (1ll << 32))) {
1770 		nr = min_nr;
1771 		if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1772 				vol->mft_record_size_bits >= (1ll << 32))) {
1773 			ntfs_warning(vol->sb, "Cannot allocate mft record "
1774 					"because the maximum number of inodes "
1775 					"(2^32) has already been reached.");
1776 			up_write(&mft_ni->runlist.lock);
1777 			return -ENOSPC;
1778 		}
1779 	}
1780 	ntfs_debug("Trying mft data allocation with %s cluster count %lli.",
1781 			nr > min_nr ? "default" : "minimal", (long long)nr);
1782 	old_last_vcn = rl[1].vcn;
1783 	do {
1784 		rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE,
1785 				TRUE);
1786 		if (likely(!IS_ERR(rl2)))
1787 			break;
1788 		if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) {
1789 			ntfs_error(vol->sb, "Failed to allocate the minimal "
1790 					"number of clusters (%lli) for the "
1791 					"mft data attribute.", (long long)nr);
1792 			up_write(&mft_ni->runlist.lock);
1793 			return PTR_ERR(rl2);
1794 		}
1795 		/*
1796 		 * There is not enough space to do the allocation, but there
1797 		 * might be enough space to do a minimal allocation so try that
1798 		 * before failing.
1799 		 */
1800 		nr = min_nr;
1801 		ntfs_debug("Retrying mft data allocation with minimal cluster "
1802 				"count %lli.", (long long)nr);
1803 	} while (1);
1804 	rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2);
1805 	if (IS_ERR(rl)) {
1806 		up_write(&mft_ni->runlist.lock);
1807 		ntfs_error(vol->sb, "Failed to merge runlists for mft data "
1808 				"attribute.");
1809 		if (ntfs_cluster_free_from_rl(vol, rl2)) {
1810 			ntfs_error(vol->sb, "Failed to dealocate clusters "
1811 					"from the mft data attribute.%s", es);
1812 			NVolSetErrors(vol);
1813 		}
1814 		ntfs_free(rl2);
1815 		return PTR_ERR(rl);
1816 	}
1817 	mft_ni->runlist.rl = rl;
1818 	ntfs_debug("Allocated %lli clusters.", (long long)nr);
1819 	/* Find the last run in the new runlist. */
1820 	for (; rl[1].length; rl++)
1821 		;
1822 	/* Update the attribute record as well. */
1823 	mrec = map_mft_record(mft_ni);
1824 	if (IS_ERR(mrec)) {
1825 		ntfs_error(vol->sb, "Failed to map mft record.");
1826 		ret = PTR_ERR(mrec);
1827 		goto undo_alloc;
1828 	}
1829 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1830 	if (unlikely(!ctx)) {
1831 		ntfs_error(vol->sb, "Failed to get search context.");
1832 		ret = -ENOMEM;
1833 		goto undo_alloc;
1834 	}
1835 	ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1836 			CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx);
1837 	if (unlikely(ret)) {
1838 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1839 				"mft data attribute.");
1840 		if (ret == -ENOENT)
1841 			ret = -EIO;
1842 		goto undo_alloc;
1843 	}
1844 	a = ctx->attr;
1845 	ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1846 	/* Search back for the previous last allocated cluster of mft bitmap. */
1847 	for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) {
1848 		if (ll >= rl2->vcn)
1849 			break;
1850 	}
1851 	BUG_ON(ll < rl2->vcn);
1852 	BUG_ON(ll >= rl2->vcn + rl2->length);
1853 	/* Get the size for the new mapping pairs array for this extent. */
1854 	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1855 	if (unlikely(mp_size <= 0)) {
1856 		ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1857 				"mft data attribute extent.");
1858 		ret = mp_size;
1859 		if (!ret)
1860 			ret = -EIO;
1861 		goto undo_alloc;
1862 	}
1863 	/* Expand the attribute record if necessary. */
1864 	old_alen = le32_to_cpu(a->length);
1865 	ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1866 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1867 	if (unlikely(ret)) {
1868 		if (ret != -ENOSPC) {
1869 			ntfs_error(vol->sb, "Failed to resize attribute "
1870 					"record for mft data attribute.");
1871 			goto undo_alloc;
1872 		}
1873 		// TODO: Deal with this by moving this extent to a new mft
1874 		// record or by starting a new extent in a new mft record or by
1875 		// moving other attributes out of this mft record.
1876 		// Note: Use the special reserved mft records and ensure that
1877 		// this extent is not required to find the mft record in
1878 		// question.  If no free special records left we would need to
1879 		// move an existing record away, insert ours in its place, and
1880 		// then place the moved record into the newly allocated space
1881 		// and we would then need to update all references to this mft
1882 		// record appropriately.  This is rather complicated...
1883 		ntfs_error(vol->sb, "Not enough space in this mft record to "
1884 				"accomodate extended mft data attribute "
1885 				"extent.  Cannot handle this yet.");
1886 		ret = -EOPNOTSUPP;
1887 		goto undo_alloc;
1888 	}
1889 	mp_rebuilt = TRUE;
1890 	/* Generate the mapping pairs array directly into the attr record. */
1891 	ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1892 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1893 			mp_size, rl2, ll, -1, NULL);
1894 	if (unlikely(ret)) {
1895 		ntfs_error(vol->sb, "Failed to build mapping pairs array of "
1896 				"mft data attribute.");
1897 		goto undo_alloc;
1898 	}
1899 	/* Update the highest_vcn. */
1900 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1901 	/*
1902 	 * We now have extended the mft data allocated_size by nr clusters.
1903 	 * Reflect this in the ntfs_inode structure and the attribute record.
1904 	 * @rl is the last (non-terminator) runlist element of mft data
1905 	 * attribute.
1906 	 */
1907 	if (a->data.non_resident.lowest_vcn) {
1908 		/*
1909 		 * We are not in the first attribute extent, switch to it, but
1910 		 * first ensure the changes will make it to disk later.
1911 		 */
1912 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1913 		mark_mft_record_dirty(ctx->ntfs_ino);
1914 		ntfs_attr_reinit_search_ctx(ctx);
1915 		ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name,
1916 				mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0,
1917 				ctx);
1918 		if (unlikely(ret)) {
1919 			ntfs_error(vol->sb, "Failed to find first attribute "
1920 					"extent of mft data attribute.");
1921 			goto restore_undo_alloc;
1922 		}
1923 		a = ctx->attr;
1924 	}
1925 	write_lock_irqsave(&mft_ni->size_lock, flags);
1926 	mft_ni->allocated_size += nr << vol->cluster_size_bits;
1927 	a->data.non_resident.allocated_size =
1928 			cpu_to_sle64(mft_ni->allocated_size);
1929 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
1930 	/* Ensure the changes make it to disk. */
1931 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1932 	mark_mft_record_dirty(ctx->ntfs_ino);
1933 	ntfs_attr_put_search_ctx(ctx);
1934 	unmap_mft_record(mft_ni);
1935 	up_write(&mft_ni->runlist.lock);
1936 	ntfs_debug("Done.");
1937 	return 0;
1938 restore_undo_alloc:
1939 	ntfs_attr_reinit_search_ctx(ctx);
1940 	if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1941 			CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) {
1942 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1943 				"mft data attribute.%s", es);
1944 		write_lock_irqsave(&mft_ni->size_lock, flags);
1945 		mft_ni->allocated_size += nr << vol->cluster_size_bits;
1946 		write_unlock_irqrestore(&mft_ni->size_lock, flags);
1947 		ntfs_attr_put_search_ctx(ctx);
1948 		unmap_mft_record(mft_ni);
1949 		up_write(&mft_ni->runlist.lock);
1950 		/*
1951 		 * The only thing that is now wrong is ->allocated_size of the
1952 		 * base attribute extent which chkdsk should be able to fix.
1953 		 */
1954 		NVolSetErrors(vol);
1955 		return ret;
1956 	}
1957 	ctx->attr->data.non_resident.highest_vcn =
1958 			cpu_to_sle64(old_last_vcn - 1);
1959 undo_alloc:
1960 	if (ntfs_cluster_free(mft_ni, old_last_vcn, -1, ctx) < 0) {
1961 		ntfs_error(vol->sb, "Failed to free clusters from mft data "
1962 				"attribute.%s", es);
1963 		NVolSetErrors(vol);
1964 	}
1965 	a = ctx->attr;
1966 	if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) {
1967 		ntfs_error(vol->sb, "Failed to truncate mft data attribute "
1968 				"runlist.%s", es);
1969 		NVolSetErrors(vol);
1970 	}
1971 	if (mp_rebuilt && !IS_ERR(ctx->mrec)) {
1972 		if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1973 				a->data.non_resident.mapping_pairs_offset),
1974 				old_alen - le16_to_cpu(
1975 				a->data.non_resident.mapping_pairs_offset),
1976 				rl2, ll, -1, NULL)) {
1977 			ntfs_error(vol->sb, "Failed to restore mapping pairs "
1978 					"array.%s", es);
1979 			NVolSetErrors(vol);
1980 		}
1981 		if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1982 			ntfs_error(vol->sb, "Failed to restore attribute "
1983 					"record.%s", es);
1984 			NVolSetErrors(vol);
1985 		}
1986 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1987 		mark_mft_record_dirty(ctx->ntfs_ino);
1988 	} else if (IS_ERR(ctx->mrec)) {
1989 		ntfs_error(vol->sb, "Failed to restore attribute search "
1990 				"context.%s", es);
1991 		NVolSetErrors(vol);
1992 	}
1993 	if (ctx)
1994 		ntfs_attr_put_search_ctx(ctx);
1995 	if (!IS_ERR(mrec))
1996 		unmap_mft_record(mft_ni);
1997 	up_write(&mft_ni->runlist.lock);
1998 	return ret;
1999 }
2000 
2001 /**
2002  * ntfs_mft_record_layout - layout an mft record into a memory buffer
2003  * @vol:	volume to which the mft record will belong
2004  * @mft_no:	mft reference specifying the mft record number
2005  * @m:		destination buffer of size >= @vol->mft_record_size bytes
2006  *
2007  * Layout an empty, unused mft record with the mft record number @mft_no into
2008  * the buffer @m.  The volume @vol is needed because the mft record structure
2009  * was modified in NTFS 3.1 so we need to know which volume version this mft
2010  * record will be used on.
2011  *
2012  * Return 0 on success and -errno on error.
2013  */
2014 static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no,
2015 		MFT_RECORD *m)
2016 {
2017 	ATTR_RECORD *a;
2018 
2019 	ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2020 	if (mft_no >= (1ll << 32)) {
2021 		ntfs_error(vol->sb, "Mft record number 0x%llx exceeds "
2022 				"maximum of 2^32.", (long long)mft_no);
2023 		return -ERANGE;
2024 	}
2025 	/* Start by clearing the whole mft record to gives us a clean slate. */
2026 	memset(m, 0, vol->mft_record_size);
2027 	/* Aligned to 2-byte boundary. */
2028 	if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver))
2029 		m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1);
2030 	else {
2031 		m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1);
2032 		/*
2033 		 * Set the NTFS 3.1+ specific fields while we know that the
2034 		 * volume version is 3.1+.
2035 		 */
2036 		m->reserved = 0;
2037 		m->mft_record_number = cpu_to_le32((u32)mft_no);
2038 	}
2039 	m->magic = magic_FILE;
2040 	if (vol->mft_record_size >= NTFS_BLOCK_SIZE)
2041 		m->usa_count = cpu_to_le16(vol->mft_record_size /
2042 				NTFS_BLOCK_SIZE + 1);
2043 	else {
2044 		m->usa_count = cpu_to_le16(1);
2045 		ntfs_warning(vol->sb, "Sector size is bigger than mft record "
2046 				"size.  Setting usa_count to 1.  If chkdsk "
2047 				"reports this as corruption, please email "
2048 				"linux-ntfs-dev@lists.sourceforge.net stating "
2049 				"that you saw this message and that the "
2050 				"modified filesystem created was corrupt.  "
2051 				"Thank you.");
2052 	}
2053 	/* Set the update sequence number to 1. */
2054 	*(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1);
2055 	m->lsn = 0;
2056 	m->sequence_number = cpu_to_le16(1);
2057 	m->link_count = 0;
2058 	/*
2059 	 * Place the attributes straight after the update sequence array,
2060 	 * aligned to 8-byte boundary.
2061 	 */
2062 	m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) +
2063 			(le16_to_cpu(m->usa_count) << 1) + 7) & ~7);
2064 	m->flags = 0;
2065 	/*
2066 	 * Using attrs_offset plus eight bytes (for the termination attribute).
2067 	 * attrs_offset is already aligned to 8-byte boundary, so no need to
2068 	 * align again.
2069 	 */
2070 	m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8);
2071 	m->bytes_allocated = cpu_to_le32(vol->mft_record_size);
2072 	m->base_mft_record = 0;
2073 	m->next_attr_instance = 0;
2074 	/* Add the termination attribute. */
2075 	a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset));
2076 	a->type = AT_END;
2077 	a->length = 0;
2078 	ntfs_debug("Done.");
2079 	return 0;
2080 }
2081 
2082 /**
2083  * ntfs_mft_record_format - format an mft record on an ntfs volume
2084  * @vol:	volume on which to format the mft record
2085  * @mft_no:	mft record number to format
2086  *
2087  * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused
2088  * mft record into the appropriate place of the mft data attribute.  This is
2089  * used when extending the mft data attribute.
2090  *
2091  * Return 0 on success and -errno on error.
2092  */
2093 static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no)
2094 {
2095 	loff_t i_size;
2096 	struct inode *mft_vi = vol->mft_ino;
2097 	struct page *page;
2098 	MFT_RECORD *m;
2099 	pgoff_t index, end_index;
2100 	unsigned int ofs;
2101 	int err;
2102 
2103 	ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2104 	/*
2105 	 * The index into the page cache and the offset within the page cache
2106 	 * page of the wanted mft record.
2107 	 */
2108 	index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2109 	ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2110 	/* The maximum valid index into the page cache for $MFT's data. */
2111 	i_size = i_size_read(mft_vi);
2112 	end_index = i_size >> PAGE_CACHE_SHIFT;
2113 	if (unlikely(index >= end_index)) {
2114 		if (unlikely(index > end_index || ofs + vol->mft_record_size >=
2115 				(i_size & ~PAGE_CACHE_MASK))) {
2116 			ntfs_error(vol->sb, "Tried to format non-existing mft "
2117 					"record 0x%llx.", (long long)mft_no);
2118 			return -ENOENT;
2119 		}
2120 	}
2121 	/* Read, map, and pin the page containing the mft record. */
2122 	page = ntfs_map_page(mft_vi->i_mapping, index);
2123 	if (unlikely(IS_ERR(page))) {
2124 		ntfs_error(vol->sb, "Failed to map page containing mft record "
2125 				"to format 0x%llx.", (long long)mft_no);
2126 		return PTR_ERR(page);
2127 	}
2128 	lock_page(page);
2129 	BUG_ON(!PageUptodate(page));
2130 	ClearPageUptodate(page);
2131 	m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2132 	err = ntfs_mft_record_layout(vol, mft_no, m);
2133 	if (unlikely(err)) {
2134 		ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.",
2135 				(long long)mft_no);
2136 		SetPageUptodate(page);
2137 		unlock_page(page);
2138 		ntfs_unmap_page(page);
2139 		return err;
2140 	}
2141 	flush_dcache_page(page);
2142 	SetPageUptodate(page);
2143 	unlock_page(page);
2144 	/*
2145 	 * Make sure the mft record is written out to disk.  We could use
2146 	 * ilookup5() to check if an inode is in icache and so on but this is
2147 	 * unnecessary as ntfs_writepage() will write the dirty record anyway.
2148 	 */
2149 	mark_ntfs_record_dirty(page, ofs);
2150 	ntfs_unmap_page(page);
2151 	ntfs_debug("Done.");
2152 	return 0;
2153 }
2154 
2155 /**
2156  * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume
2157  * @vol:	[IN]  volume on which to allocate the mft record
2158  * @mode:	[IN]  mode if want a file or directory, i.e. base inode or 0
2159  * @base_ni:	[IN]  open base inode if allocating an extent mft record or NULL
2160  * @mrec:	[OUT] on successful return this is the mapped mft record
2161  *
2162  * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol.
2163  *
2164  * If @base_ni is NULL make the mft record a base mft record, i.e. a file or
2165  * direvctory inode, and allocate it at the default allocator position.  In
2166  * this case @mode is the file mode as given to us by the caller.  We in
2167  * particular use @mode to distinguish whether a file or a directory is being
2168  * created (S_IFDIR(mode) and S_IFREG(mode), respectively).
2169  *
2170  * If @base_ni is not NULL make the allocated mft record an extent record,
2171  * allocate it starting at the mft record after the base mft record and attach
2172  * the allocated and opened ntfs inode to the base inode @base_ni.  In this
2173  * case @mode must be 0 as it is meaningless for extent inodes.
2174  *
2175  * You need to check the return value with IS_ERR().  If false, the function
2176  * was successful and the return value is the now opened ntfs inode of the
2177  * allocated mft record.  *@mrec is then set to the allocated, mapped, pinned,
2178  * and locked mft record.  If IS_ERR() is true, the function failed and the
2179  * error code is obtained from PTR_ERR(return value).  *@mrec is undefined in
2180  * this case.
2181  *
2182  * Allocation strategy:
2183  *
2184  * To find a free mft record, we scan the mft bitmap for a zero bit.  To
2185  * optimize this we start scanning at the place specified by @base_ni or if
2186  * @base_ni is NULL we start where we last stopped and we perform wrap around
2187  * when we reach the end.  Note, we do not try to allocate mft records below
2188  * number 24 because numbers 0 to 15 are the defined system files anyway and 16
2189  * to 24 are special in that they are used for storing extension mft records
2190  * for the $DATA attribute of $MFT.  This is required to avoid the possibility
2191  * of creating a runlist with a circular dependency which once written to disk
2192  * can never be read in again.  Windows will only use records 16 to 24 for
2193  * normal files if the volume is completely out of space.  We never use them
2194  * which means that when the volume is really out of space we cannot create any
2195  * more files while Windows can still create up to 8 small files.  We can start
2196  * doing this at some later time, it does not matter much for now.
2197  *
2198  * When scanning the mft bitmap, we only search up to the last allocated mft
2199  * record.  If there are no free records left in the range 24 to number of
2200  * allocated mft records, then we extend the $MFT/$DATA attribute in order to
2201  * create free mft records.  We extend the allocated size of $MFT/$DATA by 16
2202  * records at a time or one cluster, if cluster size is above 16kiB.  If there
2203  * is not sufficient space to do this, we try to extend by a single mft record
2204  * or one cluster, if cluster size is above the mft record size.
2205  *
2206  * No matter how many mft records we allocate, we initialize only the first
2207  * allocated mft record, incrementing mft data size and initialized size
2208  * accordingly, open an ntfs_inode for it and return it to the caller, unless
2209  * there are less than 24 mft records, in which case we allocate and initialize
2210  * mft records until we reach record 24 which we consider as the first free mft
2211  * record for use by normal files.
2212  *
2213  * If during any stage we overflow the initialized data in the mft bitmap, we
2214  * extend the initialized size (and data size) by 8 bytes, allocating another
2215  * cluster if required.  The bitmap data size has to be at least equal to the
2216  * number of mft records in the mft, but it can be bigger, in which case the
2217  * superflous bits are padded with zeroes.
2218  *
2219  * Thus, when we return successfully (IS_ERR() is false), we will have:
2220  *	- initialized / extended the mft bitmap if necessary,
2221  *	- initialized / extended the mft data if necessary,
2222  *	- set the bit corresponding to the mft record being allocated in the
2223  *	  mft bitmap,
2224  *	- opened an ntfs_inode for the allocated mft record, and we will have
2225  *	- returned the ntfs_inode as well as the allocated mapped, pinned, and
2226  *	  locked mft record.
2227  *
2228  * On error, the volume will be left in a consistent state and no record will
2229  * be allocated.  If rolling back a partial operation fails, we may leave some
2230  * inconsistent metadata in which case we set NVolErrors() so the volume is
2231  * left dirty when unmounted.
2232  *
2233  * Note, this function cannot make use of most of the normal functions, like
2234  * for example for attribute resizing, etc, because when the run list overflows
2235  * the base mft record and an attribute list is used, it is very important that
2236  * the extension mft records used to store the $DATA attribute of $MFT can be
2237  * reached without having to read the information contained inside them, as
2238  * this would make it impossible to find them in the first place after the
2239  * volume is unmounted.  $MFT/$BITMAP probably does not need to follow this
2240  * rule because the bitmap is not essential for finding the mft records, but on
2241  * the other hand, handling the bitmap in this special way would make life
2242  * easier because otherwise there might be circular invocations of functions
2243  * when reading the bitmap.
2244  */
2245 ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode,
2246 		ntfs_inode *base_ni, MFT_RECORD **mrec)
2247 {
2248 	s64 ll, bit, old_data_initialized, old_data_size;
2249 	unsigned long flags;
2250 	struct inode *vi;
2251 	struct page *page;
2252 	ntfs_inode *mft_ni, *mftbmp_ni, *ni;
2253 	ntfs_attr_search_ctx *ctx;
2254 	MFT_RECORD *m;
2255 	ATTR_RECORD *a;
2256 	pgoff_t index;
2257 	unsigned int ofs;
2258 	int err;
2259 	le16 seq_no, usn;
2260 	BOOL record_formatted = FALSE;
2261 
2262 	if (base_ni) {
2263 		ntfs_debug("Entering (allocating an extent mft record for "
2264 				"base mft record 0x%llx).",
2265 				(long long)base_ni->mft_no);
2266 		/* @mode and @base_ni are mutually exclusive. */
2267 		BUG_ON(mode);
2268 	} else
2269 		ntfs_debug("Entering (allocating a base mft record).");
2270 	if (mode) {
2271 		/* @mode and @base_ni are mutually exclusive. */
2272 		BUG_ON(base_ni);
2273 		/* We only support creation of normal files and directories. */
2274 		if (!S_ISREG(mode) && !S_ISDIR(mode))
2275 			return ERR_PTR(-EOPNOTSUPP);
2276 	}
2277 	BUG_ON(!mrec);
2278 	mft_ni = NTFS_I(vol->mft_ino);
2279 	mftbmp_ni = NTFS_I(vol->mftbmp_ino);
2280 	down_write(&vol->mftbmp_lock);
2281 	bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni);
2282 	if (bit >= 0) {
2283 		ntfs_debug("Found and allocated free record (#1), bit 0x%llx.",
2284 				(long long)bit);
2285 		goto have_alloc_rec;
2286 	}
2287 	if (bit != -ENOSPC) {
2288 		up_write(&vol->mftbmp_lock);
2289 		return ERR_PTR(bit);
2290 	}
2291 	/*
2292 	 * No free mft records left.  If the mft bitmap already covers more
2293 	 * than the currently used mft records, the next records are all free,
2294 	 * so we can simply allocate the first unused mft record.
2295 	 * Note: We also have to make sure that the mft bitmap at least covers
2296 	 * the first 24 mft records as they are special and whilst they may not
2297 	 * be in use, we do not allocate from them.
2298 	 */
2299 	read_lock_irqsave(&mft_ni->size_lock, flags);
2300 	ll = mft_ni->initialized_size >> vol->mft_record_size_bits;
2301 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2302 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2303 	old_data_initialized = mftbmp_ni->initialized_size;
2304 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2305 	if (old_data_initialized << 3 > ll && old_data_initialized > 3) {
2306 		bit = ll;
2307 		if (bit < 24)
2308 			bit = 24;
2309 		if (unlikely(bit >= (1ll << 32)))
2310 			goto max_err_out;
2311 		ntfs_debug("Found free record (#2), bit 0x%llx.",
2312 				(long long)bit);
2313 		goto found_free_rec;
2314 	}
2315 	/*
2316 	 * The mft bitmap needs to be expanded until it covers the first unused
2317 	 * mft record that we can allocate.
2318 	 * Note: The smallest mft record we allocate is mft record 24.
2319 	 */
2320 	bit = old_data_initialized << 3;
2321 	if (unlikely(bit >= (1ll << 32)))
2322 		goto max_err_out;
2323 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2324 	old_data_size = mftbmp_ni->allocated_size;
2325 	ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, "
2326 			"data_size 0x%llx, initialized_size 0x%llx.",
2327 			(long long)old_data_size,
2328 			(long long)i_size_read(vol->mftbmp_ino),
2329 			(long long)old_data_initialized);
2330 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2331 	if (old_data_initialized + 8 > old_data_size) {
2332 		/* Need to extend bitmap by one more cluster. */
2333 		ntfs_debug("mftbmp: initialized_size + 8 > allocated_size.");
2334 		err = ntfs_mft_bitmap_extend_allocation_nolock(vol);
2335 		if (unlikely(err)) {
2336 			up_write(&vol->mftbmp_lock);
2337 			goto err_out;
2338 		}
2339 #ifdef DEBUG
2340 		read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2341 		ntfs_debug("Status of mftbmp after allocation extension: "
2342 				"allocated_size 0x%llx, data_size 0x%llx, "
2343 				"initialized_size 0x%llx.",
2344 				(long long)mftbmp_ni->allocated_size,
2345 				(long long)i_size_read(vol->mftbmp_ino),
2346 				(long long)mftbmp_ni->initialized_size);
2347 		read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2348 #endif /* DEBUG */
2349 	}
2350 	/*
2351 	 * We now have sufficient allocated space, extend the initialized_size
2352 	 * as well as the data_size if necessary and fill the new space with
2353 	 * zeroes.
2354 	 */
2355 	err = ntfs_mft_bitmap_extend_initialized_nolock(vol);
2356 	if (unlikely(err)) {
2357 		up_write(&vol->mftbmp_lock);
2358 		goto err_out;
2359 	}
2360 #ifdef DEBUG
2361 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2362 	ntfs_debug("Status of mftbmp after initialized extention: "
2363 			"allocated_size 0x%llx, data_size 0x%llx, "
2364 			"initialized_size 0x%llx.",
2365 			(long long)mftbmp_ni->allocated_size,
2366 			(long long)i_size_read(vol->mftbmp_ino),
2367 			(long long)mftbmp_ni->initialized_size);
2368 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2369 #endif /* DEBUG */
2370 	ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit);
2371 found_free_rec:
2372 	/* @bit is the found free mft record, allocate it in the mft bitmap. */
2373 	ntfs_debug("At found_free_rec.");
2374 	err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit);
2375 	if (unlikely(err)) {
2376 		ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap.");
2377 		up_write(&vol->mftbmp_lock);
2378 		goto err_out;
2379 	}
2380 	ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit);
2381 have_alloc_rec:
2382 	/*
2383 	 * The mft bitmap is now uptodate.  Deal with mft data attribute now.
2384 	 * Note, we keep hold of the mft bitmap lock for writing until all
2385 	 * modifications to the mft data attribute are complete, too, as they
2386 	 * will impact decisions for mft bitmap and mft record allocation done
2387 	 * by a parallel allocation and if the lock is not maintained a
2388 	 * parallel allocation could allocate the same mft record as this one.
2389 	 */
2390 	ll = (bit + 1) << vol->mft_record_size_bits;
2391 	read_lock_irqsave(&mft_ni->size_lock, flags);
2392 	old_data_initialized = mft_ni->initialized_size;
2393 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2394 	if (ll <= old_data_initialized) {
2395 		ntfs_debug("Allocated mft record already initialized.");
2396 		goto mft_rec_already_initialized;
2397 	}
2398 	ntfs_debug("Initializing allocated mft record.");
2399 	/*
2400 	 * The mft record is outside the initialized data.  Extend the mft data
2401 	 * attribute until it covers the allocated record.  The loop is only
2402 	 * actually traversed more than once when a freshly formatted volume is
2403 	 * first written to so it optimizes away nicely in the common case.
2404 	 */
2405 	read_lock_irqsave(&mft_ni->size_lock, flags);
2406 	ntfs_debug("Status of mft data before extension: "
2407 			"allocated_size 0x%llx, data_size 0x%llx, "
2408 			"initialized_size 0x%llx.",
2409 			(long long)mft_ni->allocated_size,
2410 			(long long)i_size_read(vol->mft_ino),
2411 			(long long)mft_ni->initialized_size);
2412 	while (ll > mft_ni->allocated_size) {
2413 		read_unlock_irqrestore(&mft_ni->size_lock, flags);
2414 		err = ntfs_mft_data_extend_allocation_nolock(vol);
2415 		if (unlikely(err)) {
2416 			ntfs_error(vol->sb, "Failed to extend mft data "
2417 					"allocation.");
2418 			goto undo_mftbmp_alloc_nolock;
2419 		}
2420 		read_lock_irqsave(&mft_ni->size_lock, flags);
2421 		ntfs_debug("Status of mft data after allocation extension: "
2422 				"allocated_size 0x%llx, data_size 0x%llx, "
2423 				"initialized_size 0x%llx.",
2424 				(long long)mft_ni->allocated_size,
2425 				(long long)i_size_read(vol->mft_ino),
2426 				(long long)mft_ni->initialized_size);
2427 	}
2428 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2429 	/*
2430 	 * Extend mft data initialized size (and data size of course) to reach
2431 	 * the allocated mft record, formatting the mft records allong the way.
2432 	 * Note: We only modify the ntfs_inode structure as that is all that is
2433 	 * needed by ntfs_mft_record_format().  We will update the attribute
2434 	 * record itself in one fell swoop later on.
2435 	 */
2436 	write_lock_irqsave(&mft_ni->size_lock, flags);
2437 	old_data_initialized = mft_ni->initialized_size;
2438 	old_data_size = vol->mft_ino->i_size;
2439 	while (ll > mft_ni->initialized_size) {
2440 		s64 new_initialized_size, mft_no;
2441 
2442 		new_initialized_size = mft_ni->initialized_size +
2443 				vol->mft_record_size;
2444 		mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits;
2445 		if (new_initialized_size > i_size_read(vol->mft_ino))
2446 			i_size_write(vol->mft_ino, new_initialized_size);
2447 		write_unlock_irqrestore(&mft_ni->size_lock, flags);
2448 		ntfs_debug("Initializing mft record 0x%llx.",
2449 				(long long)mft_no);
2450 		err = ntfs_mft_record_format(vol, mft_no);
2451 		if (unlikely(err)) {
2452 			ntfs_error(vol->sb, "Failed to format mft record.");
2453 			goto undo_data_init;
2454 		}
2455 		write_lock_irqsave(&mft_ni->size_lock, flags);
2456 		mft_ni->initialized_size = new_initialized_size;
2457 	}
2458 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
2459 	record_formatted = TRUE;
2460 	/* Update the mft data attribute record to reflect the new sizes. */
2461 	m = map_mft_record(mft_ni);
2462 	if (IS_ERR(m)) {
2463 		ntfs_error(vol->sb, "Failed to map mft record.");
2464 		err = PTR_ERR(m);
2465 		goto undo_data_init;
2466 	}
2467 	ctx = ntfs_attr_get_search_ctx(mft_ni, m);
2468 	if (unlikely(!ctx)) {
2469 		ntfs_error(vol->sb, "Failed to get search context.");
2470 		err = -ENOMEM;
2471 		unmap_mft_record(mft_ni);
2472 		goto undo_data_init;
2473 	}
2474 	err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
2475 			CASE_SENSITIVE, 0, NULL, 0, ctx);
2476 	if (unlikely(err)) {
2477 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
2478 				"mft data attribute.");
2479 		ntfs_attr_put_search_ctx(ctx);
2480 		unmap_mft_record(mft_ni);
2481 		goto undo_data_init;
2482 	}
2483 	a = ctx->attr;
2484 	read_lock_irqsave(&mft_ni->size_lock, flags);
2485 	a->data.non_resident.initialized_size =
2486 			cpu_to_sle64(mft_ni->initialized_size);
2487 	a->data.non_resident.data_size =
2488 			cpu_to_sle64(i_size_read(vol->mft_ino));
2489 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2490 	/* Ensure the changes make it to disk. */
2491 	flush_dcache_mft_record_page(ctx->ntfs_ino);
2492 	mark_mft_record_dirty(ctx->ntfs_ino);
2493 	ntfs_attr_put_search_ctx(ctx);
2494 	unmap_mft_record(mft_ni);
2495 	read_lock_irqsave(&mft_ni->size_lock, flags);
2496 	ntfs_debug("Status of mft data after mft record initialization: "
2497 			"allocated_size 0x%llx, data_size 0x%llx, "
2498 			"initialized_size 0x%llx.",
2499 			(long long)mft_ni->allocated_size,
2500 			(long long)i_size_read(vol->mft_ino),
2501 			(long long)mft_ni->initialized_size);
2502 	BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size);
2503 	BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino));
2504 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2505 mft_rec_already_initialized:
2506 	/*
2507 	 * We can finally drop the mft bitmap lock as the mft data attribute
2508 	 * has been fully updated.  The only disparity left is that the
2509 	 * allocated mft record still needs to be marked as in use to match the
2510 	 * set bit in the mft bitmap but this is actually not a problem since
2511 	 * this mft record is not referenced from anywhere yet and the fact
2512 	 * that it is allocated in the mft bitmap means that no-one will try to
2513 	 * allocate it either.
2514 	 */
2515 	up_write(&vol->mftbmp_lock);
2516 	/*
2517 	 * We now have allocated and initialized the mft record.  Calculate the
2518 	 * index of and the offset within the page cache page the record is in.
2519 	 */
2520 	index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2521 	ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2522 	/* Read, map, and pin the page containing the mft record. */
2523 	page = ntfs_map_page(vol->mft_ino->i_mapping, index);
2524 	if (unlikely(IS_ERR(page))) {
2525 		ntfs_error(vol->sb, "Failed to map page containing allocated "
2526 				"mft record 0x%llx.", (long long)bit);
2527 		err = PTR_ERR(page);
2528 		goto undo_mftbmp_alloc;
2529 	}
2530 	lock_page(page);
2531 	BUG_ON(!PageUptodate(page));
2532 	ClearPageUptodate(page);
2533 	m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2534 	/* If we just formatted the mft record no need to do it again. */
2535 	if (!record_formatted) {
2536 		/* Sanity check that the mft record is really not in use. */
2537 		if (ntfs_is_file_record(m->magic) &&
2538 				(m->flags & MFT_RECORD_IN_USE)) {
2539 			ntfs_error(vol->sb, "Mft record 0x%llx was marked "
2540 					"free in mft bitmap but is marked "
2541 					"used itself.  Corrupt filesystem.  "
2542 					"Unmount and run chkdsk.",
2543 					(long long)bit);
2544 			err = -EIO;
2545 			SetPageUptodate(page);
2546 			unlock_page(page);
2547 			ntfs_unmap_page(page);
2548 			NVolSetErrors(vol);
2549 			goto undo_mftbmp_alloc;
2550 		}
2551 		/*
2552 		 * We need to (re-)format the mft record, preserving the
2553 		 * sequence number if it is not zero as well as the update
2554 		 * sequence number if it is not zero or -1 (0xffff).  This
2555 		 * means we do not need to care whether or not something went
2556 		 * wrong with the previous mft record.
2557 		 */
2558 		seq_no = m->sequence_number;
2559 		usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs));
2560 		err = ntfs_mft_record_layout(vol, bit, m);
2561 		if (unlikely(err)) {
2562 			ntfs_error(vol->sb, "Failed to layout allocated mft "
2563 					"record 0x%llx.", (long long)bit);
2564 			SetPageUptodate(page);
2565 			unlock_page(page);
2566 			ntfs_unmap_page(page);
2567 			goto undo_mftbmp_alloc;
2568 		}
2569 		if (seq_no)
2570 			m->sequence_number = seq_no;
2571 		if (usn && le16_to_cpu(usn) != 0xffff)
2572 			*(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn;
2573 	}
2574 	/* Set the mft record itself in use. */
2575 	m->flags |= MFT_RECORD_IN_USE;
2576 	if (S_ISDIR(mode))
2577 		m->flags |= MFT_RECORD_IS_DIRECTORY;
2578 	flush_dcache_page(page);
2579 	SetPageUptodate(page);
2580 	if (base_ni) {
2581 		/*
2582 		 * Setup the base mft record in the extent mft record.  This
2583 		 * completes initialization of the allocated extent mft record
2584 		 * and we can simply use it with map_extent_mft_record().
2585 		 */
2586 		m->base_mft_record = MK_LE_MREF(base_ni->mft_no,
2587 				base_ni->seq_no);
2588 		/*
2589 		 * Allocate an extent inode structure for the new mft record,
2590 		 * attach it to the base inode @base_ni and map, pin, and lock
2591 		 * its, i.e. the allocated, mft record.
2592 		 */
2593 		m = map_extent_mft_record(base_ni, bit, &ni);
2594 		if (IS_ERR(m)) {
2595 			ntfs_error(vol->sb, "Failed to map allocated extent "
2596 					"mft record 0x%llx.", (long long)bit);
2597 			err = PTR_ERR(m);
2598 			/* Set the mft record itself not in use. */
2599 			m->flags &= cpu_to_le16(
2600 					~le16_to_cpu(MFT_RECORD_IN_USE));
2601 			flush_dcache_page(page);
2602 			/* Make sure the mft record is written out to disk. */
2603 			mark_ntfs_record_dirty(page, ofs);
2604 			unlock_page(page);
2605 			ntfs_unmap_page(page);
2606 			goto undo_mftbmp_alloc;
2607 		}
2608 		/*
2609 		 * Make sure the allocated mft record is written out to disk.
2610 		 * No need to set the inode dirty because the caller is going
2611 		 * to do that anyway after finishing with the new extent mft
2612 		 * record (e.g. at a minimum a new attribute will be added to
2613 		 * the mft record.
2614 		 */
2615 		mark_ntfs_record_dirty(page, ofs);
2616 		unlock_page(page);
2617 		/*
2618 		 * Need to unmap the page since map_extent_mft_record() mapped
2619 		 * it as well so we have it mapped twice at the moment.
2620 		 */
2621 		ntfs_unmap_page(page);
2622 	} else {
2623 		/*
2624 		 * Allocate a new VFS inode and set it up.  NOTE: @vi->i_nlink
2625 		 * is set to 1 but the mft record->link_count is 0.  The caller
2626 		 * needs to bear this in mind.
2627 		 */
2628 		vi = new_inode(vol->sb);
2629 		if (unlikely(!vi)) {
2630 			err = -ENOMEM;
2631 			/* Set the mft record itself not in use. */
2632 			m->flags &= cpu_to_le16(
2633 					~le16_to_cpu(MFT_RECORD_IN_USE));
2634 			flush_dcache_page(page);
2635 			/* Make sure the mft record is written out to disk. */
2636 			mark_ntfs_record_dirty(page, ofs);
2637 			unlock_page(page);
2638 			ntfs_unmap_page(page);
2639 			goto undo_mftbmp_alloc;
2640 		}
2641 		vi->i_ino = bit;
2642 		/*
2643 		 * This is the optimal IO size (for stat), not the fs block
2644 		 * size.
2645 		 */
2646 		vi->i_blksize = PAGE_CACHE_SIZE;
2647 		/*
2648 		 * This is for checking whether an inode has changed w.r.t. a
2649 		 * file so that the file can be updated if necessary (compare
2650 		 * with f_version).
2651 		 */
2652 		vi->i_version = 1;
2653 
2654 		/* The owner and group come from the ntfs volume. */
2655 		vi->i_uid = vol->uid;
2656 		vi->i_gid = vol->gid;
2657 
2658 		/* Initialize the ntfs specific part of @vi. */
2659 		ntfs_init_big_inode(vi);
2660 		ni = NTFS_I(vi);
2661 		/*
2662 		 * Set the appropriate mode, attribute type, and name.  For
2663 		 * directories, also setup the index values to the defaults.
2664 		 */
2665 		if (S_ISDIR(mode)) {
2666 			vi->i_mode = S_IFDIR | S_IRWXUGO;
2667 			vi->i_mode &= ~vol->dmask;
2668 
2669 			NInoSetMstProtected(ni);
2670 			ni->type = AT_INDEX_ALLOCATION;
2671 			ni->name = I30;
2672 			ni->name_len = 4;
2673 
2674 			ni->itype.index.block_size = 4096;
2675 			ni->itype.index.block_size_bits = generic_ffs(4096) - 1;
2676 			ni->itype.index.collation_rule = COLLATION_FILE_NAME;
2677 			if (vol->cluster_size <= ni->itype.index.block_size) {
2678 				ni->itype.index.vcn_size = vol->cluster_size;
2679 				ni->itype.index.vcn_size_bits =
2680 						vol->cluster_size_bits;
2681 			} else {
2682 				ni->itype.index.vcn_size = vol->sector_size;
2683 				ni->itype.index.vcn_size_bits =
2684 						vol->sector_size_bits;
2685 			}
2686 		} else {
2687 			vi->i_mode = S_IFREG | S_IRWXUGO;
2688 			vi->i_mode &= ~vol->fmask;
2689 
2690 			ni->type = AT_DATA;
2691 			ni->name = NULL;
2692 			ni->name_len = 0;
2693 		}
2694 		if (IS_RDONLY(vi))
2695 			vi->i_mode &= ~S_IWUGO;
2696 
2697 		/* Set the inode times to the current time. */
2698 		vi->i_atime = vi->i_mtime = vi->i_ctime =
2699 			current_fs_time(vi->i_sb);
2700 		/*
2701 		 * Set the file size to 0, the ntfs inode sizes are set to 0 by
2702 		 * the call to ntfs_init_big_inode() below.
2703 		 */
2704 		vi->i_size = 0;
2705 		vi->i_blocks = 0;
2706 
2707 		/* Set the sequence number. */
2708 		vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
2709 		/*
2710 		 * Manually map, pin, and lock the mft record as we already
2711 		 * have its page mapped and it is very easy to do.
2712 		 */
2713 		atomic_inc(&ni->count);
2714 		down(&ni->mrec_lock);
2715 		ni->page = page;
2716 		ni->page_ofs = ofs;
2717 		/*
2718 		 * Make sure the allocated mft record is written out to disk.
2719 		 * NOTE: We do not set the ntfs inode dirty because this would
2720 		 * fail in ntfs_write_inode() because the inode does not have a
2721 		 * standard information attribute yet.  Also, there is no need
2722 		 * to set the inode dirty because the caller is going to do
2723 		 * that anyway after finishing with the new mft record (e.g. at
2724 		 * a minimum some new attributes will be added to the mft
2725 		 * record.
2726 		 */
2727 		mark_ntfs_record_dirty(page, ofs);
2728 		unlock_page(page);
2729 
2730 		/* Add the inode to the inode hash for the superblock. */
2731 		insert_inode_hash(vi);
2732 
2733 		/* Update the default mft allocation position. */
2734 		vol->mft_data_pos = bit + 1;
2735 	}
2736 	/*
2737 	 * Return the opened, allocated inode of the allocated mft record as
2738 	 * well as the mapped, pinned, and locked mft record.
2739 	 */
2740 	ntfs_debug("Returning opened, allocated %sinode 0x%llx.",
2741 			base_ni ? "extent " : "", (long long)bit);
2742 	*mrec = m;
2743 	return ni;
2744 undo_data_init:
2745 	write_lock_irqsave(&mft_ni->size_lock, flags);
2746 	mft_ni->initialized_size = old_data_initialized;
2747 	i_size_write(vol->mft_ino, old_data_size);
2748 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
2749 	goto undo_mftbmp_alloc_nolock;
2750 undo_mftbmp_alloc:
2751 	down_write(&vol->mftbmp_lock);
2752 undo_mftbmp_alloc_nolock:
2753 	if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) {
2754 		ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2755 		NVolSetErrors(vol);
2756 	}
2757 	up_write(&vol->mftbmp_lock);
2758 err_out:
2759 	return ERR_PTR(err);
2760 max_err_out:
2761 	ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum "
2762 			"number of inodes (2^32) has already been reached.");
2763 	up_write(&vol->mftbmp_lock);
2764 	return ERR_PTR(-ENOSPC);
2765 }
2766 
2767 /**
2768  * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume
2769  * @ni:		ntfs inode of the mapped extent mft record to free
2770  * @m:		mapped extent mft record of the ntfs inode @ni
2771  *
2772  * Free the mapped extent mft record @m of the extent ntfs inode @ni.
2773  *
2774  * Note that this function unmaps the mft record and closes and destroys @ni
2775  * internally and hence you cannot use either @ni nor @m any more after this
2776  * function returns success.
2777  *
2778  * On success return 0 and on error return -errno.  @ni and @m are still valid
2779  * in this case and have not been freed.
2780  *
2781  * For some errors an error message is displayed and the success code 0 is
2782  * returned and the volume is then left dirty on umount.  This makes sense in
2783  * case we could not rollback the changes that were already done since the
2784  * caller no longer wants to reference this mft record so it does not matter to
2785  * the caller if something is wrong with it as long as it is properly detached
2786  * from the base inode.
2787  */
2788 int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m)
2789 {
2790 	unsigned long mft_no = ni->mft_no;
2791 	ntfs_volume *vol = ni->vol;
2792 	ntfs_inode *base_ni;
2793 	ntfs_inode **extent_nis;
2794 	int i, err;
2795 	le16 old_seq_no;
2796 	u16 seq_no;
2797 
2798 	BUG_ON(NInoAttr(ni));
2799 	BUG_ON(ni->nr_extents != -1);
2800 
2801 	down(&ni->extent_lock);
2802 	base_ni = ni->ext.base_ntfs_ino;
2803 	up(&ni->extent_lock);
2804 
2805 	BUG_ON(base_ni->nr_extents <= 0);
2806 
2807 	ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n",
2808 			mft_no, base_ni->mft_no);
2809 
2810 	down(&base_ni->extent_lock);
2811 
2812 	/* Make sure we are holding the only reference to the extent inode. */
2813 	if (atomic_read(&ni->count) > 2) {
2814 		ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, "
2815 				"not freeing.", base_ni->mft_no);
2816 		up(&base_ni->extent_lock);
2817 		return -EBUSY;
2818 	}
2819 
2820 	/* Dissociate the ntfs inode from the base inode. */
2821 	extent_nis = base_ni->ext.extent_ntfs_inos;
2822 	err = -ENOENT;
2823 	for (i = 0; i < base_ni->nr_extents; i++) {
2824 		if (ni != extent_nis[i])
2825 			continue;
2826 		extent_nis += i;
2827 		base_ni->nr_extents--;
2828 		memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) *
2829 				sizeof(ntfs_inode*));
2830 		err = 0;
2831 		break;
2832 	}
2833 
2834 	up(&base_ni->extent_lock);
2835 
2836 	if (unlikely(err)) {
2837 		ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to "
2838 				"its base inode 0x%lx.", mft_no,
2839 				base_ni->mft_no);
2840 		BUG();
2841 	}
2842 
2843 	/*
2844 	 * The extent inode is no longer attached to the base inode so no one
2845 	 * can get a reference to it any more.
2846 	 */
2847 
2848 	/* Mark the mft record as not in use. */
2849 	m->flags &= const_cpu_to_le16(~const_le16_to_cpu(MFT_RECORD_IN_USE));
2850 
2851 	/* Increment the sequence number, skipping zero, if it is not zero. */
2852 	old_seq_no = m->sequence_number;
2853 	seq_no = le16_to_cpu(old_seq_no);
2854 	if (seq_no == 0xffff)
2855 		seq_no = 1;
2856 	else if (seq_no)
2857 		seq_no++;
2858 	m->sequence_number = cpu_to_le16(seq_no);
2859 
2860 	/*
2861 	 * Set the ntfs inode dirty and write it out.  We do not need to worry
2862 	 * about the base inode here since whatever caused the extent mft
2863 	 * record to be freed is guaranteed to do it already.
2864 	 */
2865 	NInoSetDirty(ni);
2866 	err = write_mft_record(ni, m, 0);
2867 	if (unlikely(err)) {
2868 		ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not "
2869 				"freeing.", mft_no);
2870 		goto rollback;
2871 	}
2872 rollback_error:
2873 	/* Unmap and throw away the now freed extent inode. */
2874 	unmap_extent_mft_record(ni);
2875 	ntfs_clear_extent_inode(ni);
2876 
2877 	/* Clear the bit in the $MFT/$BITMAP corresponding to this record. */
2878 	down_write(&vol->mftbmp_lock);
2879 	err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no);
2880 	up_write(&vol->mftbmp_lock);
2881 	if (unlikely(err)) {
2882 		/*
2883 		 * The extent inode is gone but we failed to deallocate it in
2884 		 * the mft bitmap.  Just emit a warning and leave the volume
2885 		 * dirty on umount.
2886 		 */
2887 		ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2888 		NVolSetErrors(vol);
2889 	}
2890 	return 0;
2891 rollback:
2892 	/* Rollback what we did... */
2893 	down(&base_ni->extent_lock);
2894 	extent_nis = base_ni->ext.extent_ntfs_inos;
2895 	if (!(base_ni->nr_extents & 3)) {
2896 		int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*);
2897 
2898 		extent_nis = (ntfs_inode**)kmalloc(new_size, GFP_NOFS);
2899 		if (unlikely(!extent_nis)) {
2900 			ntfs_error(vol->sb, "Failed to allocate internal "
2901 					"buffer during rollback.%s", es);
2902 			up(&base_ni->extent_lock);
2903 			NVolSetErrors(vol);
2904 			goto rollback_error;
2905 		}
2906 		if (base_ni->nr_extents) {
2907 			BUG_ON(!base_ni->ext.extent_ntfs_inos);
2908 			memcpy(extent_nis, base_ni->ext.extent_ntfs_inos,
2909 					new_size - 4 * sizeof(ntfs_inode*));
2910 			kfree(base_ni->ext.extent_ntfs_inos);
2911 		}
2912 		base_ni->ext.extent_ntfs_inos = extent_nis;
2913 	}
2914 	m->flags |= MFT_RECORD_IN_USE;
2915 	m->sequence_number = old_seq_no;
2916 	extent_nis[base_ni->nr_extents++] = ni;
2917 	up(&base_ni->extent_lock);
2918 	mark_mft_record_dirty(ni);
2919 	return err;
2920 }
2921 #endif /* NTFS_RW */
2922