1 /**
2  * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project.
3  *
4  * Copyright (c) 2001-2012 Anton Altaparmakov and Tuxera Inc.
5  * Copyright (c) 2002 Richard Russon
6  *
7  * This program/include file is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License as published
9  * by the Free Software Foundation; either version 2 of the License, or
10  * (at your option) any later version.
11  *
12  * This program/include file is distributed in the hope that it will be
13  * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
14  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program (in the main directory of the Linux-NTFS
19  * distribution in the file COPYING); if not, write to the Free Software
20  * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
21  */
22 
23 #include <linux/buffer_head.h>
24 #include <linux/slab.h>
25 #include <linux/swap.h>
26 
27 #include "attrib.h"
28 #include "aops.h"
29 #include "bitmap.h"
30 #include "debug.h"
31 #include "dir.h"
32 #include "lcnalloc.h"
33 #include "malloc.h"
34 #include "mft.h"
35 #include "ntfs.h"
36 
37 /**
38  * map_mft_record_page - map the page in which a specific mft record resides
39  * @ni:		ntfs inode whose mft record page to map
40  *
41  * This maps the page in which the mft record of the ntfs inode @ni is situated
42  * and returns a pointer to the mft record within the mapped page.
43  *
44  * Return value needs to be checked with IS_ERR() and if that is true PTR_ERR()
45  * contains the negative error code returned.
46  */
map_mft_record_page(ntfs_inode * ni)47 static inline MFT_RECORD *map_mft_record_page(ntfs_inode *ni)
48 {
49 	loff_t i_size;
50 	ntfs_volume *vol = ni->vol;
51 	struct inode *mft_vi = vol->mft_ino;
52 	struct page *page;
53 	unsigned long index, end_index;
54 	unsigned ofs;
55 
56 	BUG_ON(ni->page);
57 	/*
58 	 * The index into the page cache and the offset within the page cache
59 	 * page of the wanted mft record. FIXME: We need to check for
60 	 * overflowing the unsigned long, but I don't think we would ever get
61 	 * here if the volume was that big...
62 	 */
63 	index = (u64)ni->mft_no << vol->mft_record_size_bits >>
64 			PAGE_CACHE_SHIFT;
65 	ofs = (ni->mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
66 
67 	i_size = i_size_read(mft_vi);
68 	/* The maximum valid index into the page cache for $MFT's data. */
69 	end_index = i_size >> PAGE_CACHE_SHIFT;
70 
71 	/* If the wanted index is out of bounds the mft record doesn't exist. */
72 	if (unlikely(index >= end_index)) {
73 		if (index > end_index || (i_size & ~PAGE_CACHE_MASK) < ofs +
74 				vol->mft_record_size) {
75 			page = ERR_PTR(-ENOENT);
76 			ntfs_error(vol->sb, "Attempt to read mft record 0x%lx, "
77 					"which is beyond the end of the mft.  "
78 					"This is probably a bug in the ntfs "
79 					"driver.", ni->mft_no);
80 			goto err_out;
81 		}
82 	}
83 	/* Read, map, and pin the page. */
84 	page = ntfs_map_page(mft_vi->i_mapping, index);
85 	if (likely(!IS_ERR(page))) {
86 		/* Catch multi sector transfer fixup errors. */
87 		if (likely(ntfs_is_mft_recordp((le32*)(page_address(page) +
88 				ofs)))) {
89 			ni->page = page;
90 			ni->page_ofs = ofs;
91 			return page_address(page) + ofs;
92 		}
93 		ntfs_error(vol->sb, "Mft record 0x%lx is corrupt.  "
94 				"Run chkdsk.", ni->mft_no);
95 		ntfs_unmap_page(page);
96 		page = ERR_PTR(-EIO);
97 		NVolSetErrors(vol);
98 	}
99 err_out:
100 	ni->page = NULL;
101 	ni->page_ofs = 0;
102 	return (void*)page;
103 }
104 
105 /**
106  * map_mft_record - map, pin and lock an mft record
107  * @ni:		ntfs inode whose MFT record to map
108  *
109  * First, take the mrec_lock mutex.  We might now be sleeping, while waiting
110  * for the mutex if it was already locked by someone else.
111  *
112  * The page of the record is mapped using map_mft_record_page() before being
113  * returned to the caller.
114  *
115  * This in turn uses ntfs_map_page() to get the page containing the wanted mft
116  * record (it in turn calls read_cache_page() which reads it in from disk if
117  * necessary, increments the use count on the page so that it cannot disappear
118  * under us and returns a reference to the page cache page).
119  *
120  * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it
121  * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed
122  * and the post-read mst fixups on each mft record in the page have been
123  * performed, the page gets PG_uptodate set and PG_locked cleared (this is done
124  * in our asynchronous I/O completion handler end_buffer_read_mft_async()).
125  * ntfs_map_page() waits for PG_locked to become clear and checks if
126  * PG_uptodate is set and returns an error code if not. This provides
127  * sufficient protection against races when reading/using the page.
128  *
129  * However there is the write mapping to think about. Doing the above described
130  * checking here will be fine, because when initiating the write we will set
131  * PG_locked and clear PG_uptodate making sure nobody is touching the page
132  * contents. Doing the locking this way means that the commit to disk code in
133  * the page cache code paths is automatically sufficiently locked with us as
134  * we will not touch a page that has been locked or is not uptodate. The only
135  * locking problem then is them locking the page while we are accessing it.
136  *
137  * So that code will end up having to own the mrec_lock of all mft
138  * records/inodes present in the page before I/O can proceed. In that case we
139  * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be
140  * accessing anything without owning the mrec_lock mutex.  But we do need to
141  * use them because of the read_cache_page() invocation and the code becomes so
142  * much simpler this way that it is well worth it.
143  *
144  * The mft record is now ours and we return a pointer to it. You need to check
145  * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return
146  * the error code.
147  *
148  * NOTE: Caller is responsible for setting the mft record dirty before calling
149  * unmap_mft_record(). This is obviously only necessary if the caller really
150  * modified the mft record...
151  * Q: Do we want to recycle one of the VFS inode state bits instead?
152  * A: No, the inode ones mean we want to change the mft record, not we want to
153  * write it out.
154  */
map_mft_record(ntfs_inode * ni)155 MFT_RECORD *map_mft_record(ntfs_inode *ni)
156 {
157 	MFT_RECORD *m;
158 
159 	ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
160 
161 	/* Make sure the ntfs inode doesn't go away. */
162 	atomic_inc(&ni->count);
163 
164 	/* Serialize access to this mft record. */
165 	mutex_lock(&ni->mrec_lock);
166 
167 	m = map_mft_record_page(ni);
168 	if (likely(!IS_ERR(m)))
169 		return m;
170 
171 	mutex_unlock(&ni->mrec_lock);
172 	atomic_dec(&ni->count);
173 	ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m));
174 	return m;
175 }
176 
177 /**
178  * unmap_mft_record_page - unmap the page in which a specific mft record resides
179  * @ni:		ntfs inode whose mft record page to unmap
180  *
181  * This unmaps the page in which the mft record of the ntfs inode @ni is
182  * situated and returns. This is a NOOP if highmem is not configured.
183  *
184  * The unmap happens via ntfs_unmap_page() which in turn decrements the use
185  * count on the page thus releasing it from the pinned state.
186  *
187  * We do not actually unmap the page from memory of course, as that will be
188  * done by the page cache code itself when memory pressure increases or
189  * whatever.
190  */
unmap_mft_record_page(ntfs_inode * ni)191 static inline void unmap_mft_record_page(ntfs_inode *ni)
192 {
193 	BUG_ON(!ni->page);
194 
195 	// TODO: If dirty, blah...
196 	ntfs_unmap_page(ni->page);
197 	ni->page = NULL;
198 	ni->page_ofs = 0;
199 	return;
200 }
201 
202 /**
203  * unmap_mft_record - release a mapped mft record
204  * @ni:		ntfs inode whose MFT record to unmap
205  *
206  * We release the page mapping and the mrec_lock mutex which unmaps the mft
207  * record and releases it for others to get hold of. We also release the ntfs
208  * inode by decrementing the ntfs inode reference count.
209  *
210  * NOTE: If caller has modified the mft record, it is imperative to set the mft
211  * record dirty BEFORE calling unmap_mft_record().
212  */
unmap_mft_record(ntfs_inode * ni)213 void unmap_mft_record(ntfs_inode *ni)
214 {
215 	struct page *page = ni->page;
216 
217 	BUG_ON(!page);
218 
219 	ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no);
220 
221 	unmap_mft_record_page(ni);
222 	mutex_unlock(&ni->mrec_lock);
223 	atomic_dec(&ni->count);
224 	/*
225 	 * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to
226 	 * ntfs_clear_extent_inode() in the extent inode case, and to the
227 	 * caller in the non-extent, yet pure ntfs inode case, to do the actual
228 	 * tear down of all structures and freeing of all allocated memory.
229 	 */
230 	return;
231 }
232 
233 /**
234  * map_extent_mft_record - load an extent inode and attach it to its base
235  * @base_ni:	base ntfs inode
236  * @mref:	mft reference of the extent inode to load
237  * @ntfs_ino:	on successful return, pointer to the ntfs_inode structure
238  *
239  * Load the extent mft record @mref and attach it to its base inode @base_ni.
240  * Return the mapped extent mft record if IS_ERR(result) is false.  Otherwise
241  * PTR_ERR(result) gives the negative error code.
242  *
243  * On successful return, @ntfs_ino contains a pointer to the ntfs_inode
244  * structure of the mapped extent inode.
245  */
map_extent_mft_record(ntfs_inode * base_ni,MFT_REF mref,ntfs_inode ** ntfs_ino)246 MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref,
247 		ntfs_inode **ntfs_ino)
248 {
249 	MFT_RECORD *m;
250 	ntfs_inode *ni = NULL;
251 	ntfs_inode **extent_nis = NULL;
252 	int i;
253 	unsigned long mft_no = MREF(mref);
254 	u16 seq_no = MSEQNO(mref);
255 	bool destroy_ni = false;
256 
257 	ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).",
258 			mft_no, base_ni->mft_no);
259 	/* Make sure the base ntfs inode doesn't go away. */
260 	atomic_inc(&base_ni->count);
261 	/*
262 	 * Check if this extent inode has already been added to the base inode,
263 	 * in which case just return it. If not found, add it to the base
264 	 * inode before returning it.
265 	 */
266 	mutex_lock(&base_ni->extent_lock);
267 	if (base_ni->nr_extents > 0) {
268 		extent_nis = base_ni->ext.extent_ntfs_inos;
269 		for (i = 0; i < base_ni->nr_extents; i++) {
270 			if (mft_no != extent_nis[i]->mft_no)
271 				continue;
272 			ni = extent_nis[i];
273 			/* Make sure the ntfs inode doesn't go away. */
274 			atomic_inc(&ni->count);
275 			break;
276 		}
277 	}
278 	if (likely(ni != NULL)) {
279 		mutex_unlock(&base_ni->extent_lock);
280 		atomic_dec(&base_ni->count);
281 		/* We found the record; just have to map and return it. */
282 		m = map_mft_record(ni);
283 		/* map_mft_record() has incremented this on success. */
284 		atomic_dec(&ni->count);
285 		if (likely(!IS_ERR(m))) {
286 			/* Verify the sequence number. */
287 			if (likely(le16_to_cpu(m->sequence_number) == seq_no)) {
288 				ntfs_debug("Done 1.");
289 				*ntfs_ino = ni;
290 				return m;
291 			}
292 			unmap_mft_record(ni);
293 			ntfs_error(base_ni->vol->sb, "Found stale extent mft "
294 					"reference! Corrupt filesystem. "
295 					"Run chkdsk.");
296 			return ERR_PTR(-EIO);
297 		}
298 map_err_out:
299 		ntfs_error(base_ni->vol->sb, "Failed to map extent "
300 				"mft record, error code %ld.", -PTR_ERR(m));
301 		return m;
302 	}
303 	/* Record wasn't there. Get a new ntfs inode and initialize it. */
304 	ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no);
305 	if (unlikely(!ni)) {
306 		mutex_unlock(&base_ni->extent_lock);
307 		atomic_dec(&base_ni->count);
308 		return ERR_PTR(-ENOMEM);
309 	}
310 	ni->vol = base_ni->vol;
311 	ni->seq_no = seq_no;
312 	ni->nr_extents = -1;
313 	ni->ext.base_ntfs_ino = base_ni;
314 	/* Now map the record. */
315 	m = map_mft_record(ni);
316 	if (IS_ERR(m)) {
317 		mutex_unlock(&base_ni->extent_lock);
318 		atomic_dec(&base_ni->count);
319 		ntfs_clear_extent_inode(ni);
320 		goto map_err_out;
321 	}
322 	/* Verify the sequence number if it is present. */
323 	if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) {
324 		ntfs_error(base_ni->vol->sb, "Found stale extent mft "
325 				"reference! Corrupt filesystem. Run chkdsk.");
326 		destroy_ni = true;
327 		m = ERR_PTR(-EIO);
328 		goto unm_err_out;
329 	}
330 	/* Attach extent inode to base inode, reallocating memory if needed. */
331 	if (!(base_ni->nr_extents & 3)) {
332 		ntfs_inode **tmp;
333 		int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *);
334 
335 		tmp = kmalloc(new_size, GFP_NOFS);
336 		if (unlikely(!tmp)) {
337 			ntfs_error(base_ni->vol->sb, "Failed to allocate "
338 					"internal buffer.");
339 			destroy_ni = true;
340 			m = ERR_PTR(-ENOMEM);
341 			goto unm_err_out;
342 		}
343 		if (base_ni->nr_extents) {
344 			BUG_ON(!base_ni->ext.extent_ntfs_inos);
345 			memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size -
346 					4 * sizeof(ntfs_inode *));
347 			kfree(base_ni->ext.extent_ntfs_inos);
348 		}
349 		base_ni->ext.extent_ntfs_inos = tmp;
350 	}
351 	base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni;
352 	mutex_unlock(&base_ni->extent_lock);
353 	atomic_dec(&base_ni->count);
354 	ntfs_debug("Done 2.");
355 	*ntfs_ino = ni;
356 	return m;
357 unm_err_out:
358 	unmap_mft_record(ni);
359 	mutex_unlock(&base_ni->extent_lock);
360 	atomic_dec(&base_ni->count);
361 	/*
362 	 * If the extent inode was not attached to the base inode we need to
363 	 * release it or we will leak memory.
364 	 */
365 	if (destroy_ni)
366 		ntfs_clear_extent_inode(ni);
367 	return m;
368 }
369 
370 #ifdef NTFS_RW
371 
372 /**
373  * __mark_mft_record_dirty - set the mft record and the page containing it dirty
374  * @ni:		ntfs inode describing the mapped mft record
375  *
376  * Internal function.  Users should call mark_mft_record_dirty() instead.
377  *
378  * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni,
379  * as well as the page containing the mft record, dirty.  Also, mark the base
380  * vfs inode dirty.  This ensures that any changes to the mft record are
381  * written out to disk.
382  *
383  * NOTE:  We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES)
384  * on the base vfs inode, because even though file data may have been modified,
385  * it is dirty in the inode meta data rather than the data page cache of the
386  * inode, and thus there are no data pages that need writing out.  Therefore, a
387  * full mark_inode_dirty() is overkill.  A mark_inode_dirty_sync(), on the
388  * other hand, is not sufficient, because ->write_inode needs to be called even
389  * in case of fdatasync. This needs to happen or the file data would not
390  * necessarily hit the device synchronously, even though the vfs inode has the
391  * O_SYNC flag set.  Also, I_DIRTY_DATASYNC simply "feels" better than just
392  * I_DIRTY_SYNC, since the file data has not actually hit the block device yet,
393  * which is not what I_DIRTY_SYNC on its own would suggest.
394  */
__mark_mft_record_dirty(ntfs_inode * ni)395 void __mark_mft_record_dirty(ntfs_inode *ni)
396 {
397 	ntfs_inode *base_ni;
398 
399 	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
400 	BUG_ON(NInoAttr(ni));
401 	mark_ntfs_record_dirty(ni->page, ni->page_ofs);
402 	/* Determine the base vfs inode and mark it dirty, too. */
403 	mutex_lock(&ni->extent_lock);
404 	if (likely(ni->nr_extents >= 0))
405 		base_ni = ni;
406 	else
407 		base_ni = ni->ext.base_ntfs_ino;
408 	mutex_unlock(&ni->extent_lock);
409 	__mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC);
410 }
411 
412 static const char *ntfs_please_email = "Please email "
413 		"linux-ntfs-dev@lists.sourceforge.net and say that you saw "
414 		"this message.  Thank you.";
415 
416 /**
417  * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror
418  * @vol:	ntfs volume on which the mft record to synchronize resides
419  * @mft_no:	mft record number of mft record to synchronize
420  * @m:		mapped, mst protected (extent) mft record to synchronize
421  *
422  * Write the mapped, mst protected (extent) mft record @m with mft record
423  * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol,
424  * bypassing the page cache and the $MFTMirr inode itself.
425  *
426  * This function is only for use at umount time when the mft mirror inode has
427  * already been disposed off.  We BUG() if we are called while the mft mirror
428  * inode is still attached to the volume.
429  *
430  * On success return 0.  On error return -errno.
431  *
432  * NOTE:  This function is not implemented yet as I am not convinced it can
433  * actually be triggered considering the sequence of commits we do in super.c::
434  * ntfs_put_super().  But just in case we provide this place holder as the
435  * alternative would be either to BUG() or to get a NULL pointer dereference
436  * and Oops.
437  */
ntfs_sync_mft_mirror_umount(ntfs_volume * vol,const unsigned long mft_no,MFT_RECORD * m)438 static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol,
439 		const unsigned long mft_no, MFT_RECORD *m)
440 {
441 	BUG_ON(vol->mftmirr_ino);
442 	ntfs_error(vol->sb, "Umount time mft mirror syncing is not "
443 			"implemented yet.  %s", ntfs_please_email);
444 	return -EOPNOTSUPP;
445 }
446 
447 /**
448  * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror
449  * @vol:	ntfs volume on which the mft record to synchronize resides
450  * @mft_no:	mft record number of mft record to synchronize
451  * @m:		mapped, mst protected (extent) mft record to synchronize
452  * @sync:	if true, wait for i/o completion
453  *
454  * Write the mapped, mst protected (extent) mft record @m with mft record
455  * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol.
456  *
457  * On success return 0.  On error return -errno and set the volume errors flag
458  * in the ntfs volume @vol.
459  *
460  * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
461  *
462  * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
463  * schedule i/o via ->writepage or do it via kntfsd or whatever.
464  */
ntfs_sync_mft_mirror(ntfs_volume * vol,const unsigned long mft_no,MFT_RECORD * m,int sync)465 int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no,
466 		MFT_RECORD *m, int sync)
467 {
468 	struct page *page;
469 	unsigned int blocksize = vol->sb->s_blocksize;
470 	int max_bhs = vol->mft_record_size / blocksize;
471 	struct buffer_head *bhs[max_bhs];
472 	struct buffer_head *bh, *head;
473 	u8 *kmirr;
474 	runlist_element *rl;
475 	unsigned int block_start, block_end, m_start, m_end, page_ofs;
476 	int i_bhs, nr_bhs, err = 0;
477 	unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
478 
479 	ntfs_debug("Entering for inode 0x%lx.", mft_no);
480 	BUG_ON(!max_bhs);
481 	if (unlikely(!vol->mftmirr_ino)) {
482 		/* This could happen during umount... */
483 		err = ntfs_sync_mft_mirror_umount(vol, mft_no, m);
484 		if (likely(!err))
485 			return err;
486 		goto err_out;
487 	}
488 	/* Get the page containing the mirror copy of the mft record @m. */
489 	page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >>
490 			(PAGE_CACHE_SHIFT - vol->mft_record_size_bits));
491 	if (IS_ERR(page)) {
492 		ntfs_error(vol->sb, "Failed to map mft mirror page.");
493 		err = PTR_ERR(page);
494 		goto err_out;
495 	}
496 	lock_page(page);
497 	BUG_ON(!PageUptodate(page));
498 	ClearPageUptodate(page);
499 	/* Offset of the mft mirror record inside the page. */
500 	page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
501 	/* The address in the page of the mirror copy of the mft record @m. */
502 	kmirr = page_address(page) + page_ofs;
503 	/* Copy the mst protected mft record to the mirror. */
504 	memcpy(kmirr, m, vol->mft_record_size);
505 	/* Create uptodate buffers if not present. */
506 	if (unlikely(!page_has_buffers(page))) {
507 		struct buffer_head *tail;
508 
509 		bh = head = alloc_page_buffers(page, blocksize, 1);
510 		do {
511 			set_buffer_uptodate(bh);
512 			tail = bh;
513 			bh = bh->b_this_page;
514 		} while (bh);
515 		tail->b_this_page = head;
516 		attach_page_buffers(page, head);
517 	}
518 	bh = head = page_buffers(page);
519 	BUG_ON(!bh);
520 	rl = NULL;
521 	nr_bhs = 0;
522 	block_start = 0;
523 	m_start = kmirr - (u8*)page_address(page);
524 	m_end = m_start + vol->mft_record_size;
525 	do {
526 		block_end = block_start + blocksize;
527 		/* If the buffer is outside the mft record, skip it. */
528 		if (block_end <= m_start)
529 			continue;
530 		if (unlikely(block_start >= m_end))
531 			break;
532 		/* Need to map the buffer if it is not mapped already. */
533 		if (unlikely(!buffer_mapped(bh))) {
534 			VCN vcn;
535 			LCN lcn;
536 			unsigned int vcn_ofs;
537 
538 			bh->b_bdev = vol->sb->s_bdev;
539 			/* Obtain the vcn and offset of the current block. */
540 			vcn = ((VCN)mft_no << vol->mft_record_size_bits) +
541 					(block_start - m_start);
542 			vcn_ofs = vcn & vol->cluster_size_mask;
543 			vcn >>= vol->cluster_size_bits;
544 			if (!rl) {
545 				down_read(&NTFS_I(vol->mftmirr_ino)->
546 						runlist.lock);
547 				rl = NTFS_I(vol->mftmirr_ino)->runlist.rl;
548 				/*
549 				 * $MFTMirr always has the whole of its runlist
550 				 * in memory.
551 				 */
552 				BUG_ON(!rl);
553 			}
554 			/* Seek to element containing target vcn. */
555 			while (rl->length && rl[1].vcn <= vcn)
556 				rl++;
557 			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
558 			/* For $MFTMirr, only lcn >= 0 is a successful remap. */
559 			if (likely(lcn >= 0)) {
560 				/* Setup buffer head to correct block. */
561 				bh->b_blocknr = ((lcn <<
562 						vol->cluster_size_bits) +
563 						vcn_ofs) >> blocksize_bits;
564 				set_buffer_mapped(bh);
565 			} else {
566 				bh->b_blocknr = -1;
567 				ntfs_error(vol->sb, "Cannot write mft mirror "
568 						"record 0x%lx because its "
569 						"location on disk could not "
570 						"be determined (error code "
571 						"%lli).", mft_no,
572 						(long long)lcn);
573 				err = -EIO;
574 			}
575 		}
576 		BUG_ON(!buffer_uptodate(bh));
577 		BUG_ON(!nr_bhs && (m_start != block_start));
578 		BUG_ON(nr_bhs >= max_bhs);
579 		bhs[nr_bhs++] = bh;
580 		BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
581 	} while (block_start = block_end, (bh = bh->b_this_page) != head);
582 	if (unlikely(rl))
583 		up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock);
584 	if (likely(!err)) {
585 		/* Lock buffers and start synchronous write i/o on them. */
586 		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
587 			struct buffer_head *tbh = bhs[i_bhs];
588 
589 			if (!trylock_buffer(tbh))
590 				BUG();
591 			BUG_ON(!buffer_uptodate(tbh));
592 			clear_buffer_dirty(tbh);
593 			get_bh(tbh);
594 			tbh->b_end_io = end_buffer_write_sync;
595 			submit_bh(WRITE, tbh);
596 		}
597 		/* Wait on i/o completion of buffers. */
598 		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
599 			struct buffer_head *tbh = bhs[i_bhs];
600 
601 			wait_on_buffer(tbh);
602 			if (unlikely(!buffer_uptodate(tbh))) {
603 				err = -EIO;
604 				/*
605 				 * Set the buffer uptodate so the page and
606 				 * buffer states do not become out of sync.
607 				 */
608 				set_buffer_uptodate(tbh);
609 			}
610 		}
611 	} else /* if (unlikely(err)) */ {
612 		/* Clean the buffers. */
613 		for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
614 			clear_buffer_dirty(bhs[i_bhs]);
615 	}
616 	/* Current state: all buffers are clean, unlocked, and uptodate. */
617 	/* Remove the mst protection fixups again. */
618 	post_write_mst_fixup((NTFS_RECORD*)kmirr);
619 	flush_dcache_page(page);
620 	SetPageUptodate(page);
621 	unlock_page(page);
622 	ntfs_unmap_page(page);
623 	if (likely(!err)) {
624 		ntfs_debug("Done.");
625 	} else {
626 		ntfs_error(vol->sb, "I/O error while writing mft mirror "
627 				"record 0x%lx!", mft_no);
628 err_out:
629 		ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error "
630 				"code %i).  Volume will be left marked dirty "
631 				"on umount.  Run ntfsfix on the partition "
632 				"after umounting to correct this.", -err);
633 		NVolSetErrors(vol);
634 	}
635 	return err;
636 }
637 
638 /**
639  * write_mft_record_nolock - write out a mapped (extent) mft record
640  * @ni:		ntfs inode describing the mapped (extent) mft record
641  * @m:		mapped (extent) mft record to write
642  * @sync:	if true, wait for i/o completion
643  *
644  * Write the mapped (extent) mft record @m described by the (regular or extent)
645  * ntfs inode @ni to backing store.  If the mft record @m has a counterpart in
646  * the mft mirror, that is also updated.
647  *
648  * We only write the mft record if the ntfs inode @ni is dirty and the first
649  * buffer belonging to its mft record is dirty, too.  We ignore the dirty state
650  * of subsequent buffers because we could have raced with
651  * fs/ntfs/aops.c::mark_ntfs_record_dirty().
652  *
653  * On success, clean the mft record and return 0.  On error, leave the mft
654  * record dirty and return -errno.
655  *
656  * NOTE:  We always perform synchronous i/o and ignore the @sync parameter.
657  * However, if the mft record has a counterpart in the mft mirror and @sync is
658  * true, we write the mft record, wait for i/o completion, and only then write
659  * the mft mirror copy.  This ensures that if the system crashes either the mft
660  * or the mft mirror will contain a self-consistent mft record @m.  If @sync is
661  * false on the other hand, we start i/o on both and then wait for completion
662  * on them.  This provides a speedup but no longer guarantees that you will end
663  * up with a self-consistent mft record in the case of a crash but if you asked
664  * for asynchronous writing you probably do not care about that anyway.
665  *
666  * TODO:  If @sync is false, want to do truly asynchronous i/o, i.e. just
667  * schedule i/o via ->writepage or do it via kntfsd or whatever.
668  */
write_mft_record_nolock(ntfs_inode * ni,MFT_RECORD * m,int sync)669 int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync)
670 {
671 	ntfs_volume *vol = ni->vol;
672 	struct page *page = ni->page;
673 	unsigned int blocksize = vol->sb->s_blocksize;
674 	unsigned char blocksize_bits = vol->sb->s_blocksize_bits;
675 	int max_bhs = vol->mft_record_size / blocksize;
676 	struct buffer_head *bhs[max_bhs];
677 	struct buffer_head *bh, *head;
678 	runlist_element *rl;
679 	unsigned int block_start, block_end, m_start, m_end;
680 	int i_bhs, nr_bhs, err = 0;
681 
682 	ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
683 	BUG_ON(NInoAttr(ni));
684 	BUG_ON(!max_bhs);
685 	BUG_ON(!PageLocked(page));
686 	/*
687 	 * If the ntfs_inode is clean no need to do anything.  If it is dirty,
688 	 * mark it as clean now so that it can be redirtied later on if needed.
689 	 * There is no danger of races since the caller is holding the locks
690 	 * for the mft record @m and the page it is in.
691 	 */
692 	if (!NInoTestClearDirty(ni))
693 		goto done;
694 	bh = head = page_buffers(page);
695 	BUG_ON(!bh);
696 	rl = NULL;
697 	nr_bhs = 0;
698 	block_start = 0;
699 	m_start = ni->page_ofs;
700 	m_end = m_start + vol->mft_record_size;
701 	do {
702 		block_end = block_start + blocksize;
703 		/* If the buffer is outside the mft record, skip it. */
704 		if (block_end <= m_start)
705 			continue;
706 		if (unlikely(block_start >= m_end))
707 			break;
708 		/*
709 		 * If this block is not the first one in the record, we ignore
710 		 * the buffer's dirty state because we could have raced with a
711 		 * parallel mark_ntfs_record_dirty().
712 		 */
713 		if (block_start == m_start) {
714 			/* This block is the first one in the record. */
715 			if (!buffer_dirty(bh)) {
716 				BUG_ON(nr_bhs);
717 				/* Clean records are not written out. */
718 				break;
719 			}
720 		}
721 		/* Need to map the buffer if it is not mapped already. */
722 		if (unlikely(!buffer_mapped(bh))) {
723 			VCN vcn;
724 			LCN lcn;
725 			unsigned int vcn_ofs;
726 
727 			bh->b_bdev = vol->sb->s_bdev;
728 			/* Obtain the vcn and offset of the current block. */
729 			vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) +
730 					(block_start - m_start);
731 			vcn_ofs = vcn & vol->cluster_size_mask;
732 			vcn >>= vol->cluster_size_bits;
733 			if (!rl) {
734 				down_read(&NTFS_I(vol->mft_ino)->runlist.lock);
735 				rl = NTFS_I(vol->mft_ino)->runlist.rl;
736 				BUG_ON(!rl);
737 			}
738 			/* Seek to element containing target vcn. */
739 			while (rl->length && rl[1].vcn <= vcn)
740 				rl++;
741 			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
742 			/* For $MFT, only lcn >= 0 is a successful remap. */
743 			if (likely(lcn >= 0)) {
744 				/* Setup buffer head to correct block. */
745 				bh->b_blocknr = ((lcn <<
746 						vol->cluster_size_bits) +
747 						vcn_ofs) >> blocksize_bits;
748 				set_buffer_mapped(bh);
749 			} else {
750 				bh->b_blocknr = -1;
751 				ntfs_error(vol->sb, "Cannot write mft record "
752 						"0x%lx because its location "
753 						"on disk could not be "
754 						"determined (error code %lli).",
755 						ni->mft_no, (long long)lcn);
756 				err = -EIO;
757 			}
758 		}
759 		BUG_ON(!buffer_uptodate(bh));
760 		BUG_ON(!nr_bhs && (m_start != block_start));
761 		BUG_ON(nr_bhs >= max_bhs);
762 		bhs[nr_bhs++] = bh;
763 		BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end));
764 	} while (block_start = block_end, (bh = bh->b_this_page) != head);
765 	if (unlikely(rl))
766 		up_read(&NTFS_I(vol->mft_ino)->runlist.lock);
767 	if (!nr_bhs)
768 		goto done;
769 	if (unlikely(err))
770 		goto cleanup_out;
771 	/* Apply the mst protection fixups. */
772 	err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size);
773 	if (err) {
774 		ntfs_error(vol->sb, "Failed to apply mst fixups!");
775 		goto cleanup_out;
776 	}
777 	flush_dcache_mft_record_page(ni);
778 	/* Lock buffers and start synchronous write i/o on them. */
779 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
780 		struct buffer_head *tbh = bhs[i_bhs];
781 
782 		if (!trylock_buffer(tbh))
783 			BUG();
784 		BUG_ON(!buffer_uptodate(tbh));
785 		clear_buffer_dirty(tbh);
786 		get_bh(tbh);
787 		tbh->b_end_io = end_buffer_write_sync;
788 		submit_bh(WRITE, tbh);
789 	}
790 	/* Synchronize the mft mirror now if not @sync. */
791 	if (!sync && ni->mft_no < vol->mftmirr_size)
792 		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
793 	/* Wait on i/o completion of buffers. */
794 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) {
795 		struct buffer_head *tbh = bhs[i_bhs];
796 
797 		wait_on_buffer(tbh);
798 		if (unlikely(!buffer_uptodate(tbh))) {
799 			err = -EIO;
800 			/*
801 			 * Set the buffer uptodate so the page and buffer
802 			 * states do not become out of sync.
803 			 */
804 			if (PageUptodate(page))
805 				set_buffer_uptodate(tbh);
806 		}
807 	}
808 	/* If @sync, now synchronize the mft mirror. */
809 	if (sync && ni->mft_no < vol->mftmirr_size)
810 		ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync);
811 	/* Remove the mst protection fixups again. */
812 	post_write_mst_fixup((NTFS_RECORD*)m);
813 	flush_dcache_mft_record_page(ni);
814 	if (unlikely(err)) {
815 		/* I/O error during writing.  This is really bad! */
816 		ntfs_error(vol->sb, "I/O error while writing mft record "
817 				"0x%lx!  Marking base inode as bad.  You "
818 				"should unmount the volume and run chkdsk.",
819 				ni->mft_no);
820 		goto err_out;
821 	}
822 done:
823 	ntfs_debug("Done.");
824 	return 0;
825 cleanup_out:
826 	/* Clean the buffers. */
827 	for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++)
828 		clear_buffer_dirty(bhs[i_bhs]);
829 err_out:
830 	/*
831 	 * Current state: all buffers are clean, unlocked, and uptodate.
832 	 * The caller should mark the base inode as bad so that no more i/o
833 	 * happens.  ->clear_inode() will still be invoked so all extent inodes
834 	 * and other allocated memory will be freed.
835 	 */
836 	if (err == -ENOMEM) {
837 		ntfs_error(vol->sb, "Not enough memory to write mft record.  "
838 				"Redirtying so the write is retried later.");
839 		mark_mft_record_dirty(ni);
840 		err = 0;
841 	} else
842 		NVolSetErrors(vol);
843 	return err;
844 }
845 
846 /**
847  * ntfs_may_write_mft_record - check if an mft record may be written out
848  * @vol:	[IN]  ntfs volume on which the mft record to check resides
849  * @mft_no:	[IN]  mft record number of the mft record to check
850  * @m:		[IN]  mapped mft record to check
851  * @locked_ni:	[OUT] caller has to unlock this ntfs inode if one is returned
852  *
853  * Check if the mapped (base or extent) mft record @m with mft record number
854  * @mft_no belonging to the ntfs volume @vol may be written out.  If necessary
855  * and possible the ntfs inode of the mft record is locked and the base vfs
856  * inode is pinned.  The locked ntfs inode is then returned in @locked_ni.  The
857  * caller is responsible for unlocking the ntfs inode and unpinning the base
858  * vfs inode.
859  *
860  * Return 'true' if the mft record may be written out and 'false' if not.
861  *
862  * The caller has locked the page and cleared the uptodate flag on it which
863  * means that we can safely write out any dirty mft records that do not have
864  * their inodes in icache as determined by ilookup5() as anyone
865  * opening/creating such an inode would block when attempting to map the mft
866  * record in read_cache_page() until we are finished with the write out.
867  *
868  * Here is a description of the tests we perform:
869  *
870  * If the inode is found in icache we know the mft record must be a base mft
871  * record.  If it is dirty, we do not write it and return 'false' as the vfs
872  * inode write paths will result in the access times being updated which would
873  * cause the base mft record to be redirtied and written out again.  (We know
874  * the access time update will modify the base mft record because Windows
875  * chkdsk complains if the standard information attribute is not in the base
876  * mft record.)
877  *
878  * If the inode is in icache and not dirty, we attempt to lock the mft record
879  * and if we find the lock was already taken, it is not safe to write the mft
880  * record and we return 'false'.
881  *
882  * If we manage to obtain the lock we have exclusive access to the mft record,
883  * which also allows us safe writeout of the mft record.  We then set
884  * @locked_ni to the locked ntfs inode and return 'true'.
885  *
886  * Note we cannot just lock the mft record and sleep while waiting for the lock
887  * because this would deadlock due to lock reversal (normally the mft record is
888  * locked before the page is locked but we already have the page locked here
889  * when we try to lock the mft record).
890  *
891  * If the inode is not in icache we need to perform further checks.
892  *
893  * If the mft record is not a FILE record or it is a base mft record, we can
894  * safely write it and return 'true'.
895  *
896  * We now know the mft record is an extent mft record.  We check if the inode
897  * corresponding to its base mft record is in icache and obtain a reference to
898  * it if it is.  If it is not, we can safely write it and return 'true'.
899  *
900  * We now have the base inode for the extent mft record.  We check if it has an
901  * ntfs inode for the extent mft record attached and if not it is safe to write
902  * the extent mft record and we return 'true'.
903  *
904  * The ntfs inode for the extent mft record is attached to the base inode so we
905  * attempt to lock the extent mft record and if we find the lock was already
906  * taken, it is not safe to write the extent mft record and we return 'false'.
907  *
908  * If we manage to obtain the lock we have exclusive access to the extent mft
909  * record, which also allows us safe writeout of the extent mft record.  We
910  * set the ntfs inode of the extent mft record clean and then set @locked_ni to
911  * the now locked ntfs inode and return 'true'.
912  *
913  * Note, the reason for actually writing dirty mft records here and not just
914  * relying on the vfs inode dirty code paths is that we can have mft records
915  * modified without them ever having actual inodes in memory.  Also we can have
916  * dirty mft records with clean ntfs inodes in memory.  None of the described
917  * cases would result in the dirty mft records being written out if we only
918  * relied on the vfs inode dirty code paths.  And these cases can really occur
919  * during allocation of new mft records and in particular when the
920  * initialized_size of the $MFT/$DATA attribute is extended and the new space
921  * is initialized using ntfs_mft_record_format().  The clean inode can then
922  * appear if the mft record is reused for a new inode before it got written
923  * out.
924  */
ntfs_may_write_mft_record(ntfs_volume * vol,const unsigned long mft_no,const MFT_RECORD * m,ntfs_inode ** locked_ni)925 bool ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
926 		const MFT_RECORD *m, ntfs_inode **locked_ni)
927 {
928 	struct super_block *sb = vol->sb;
929 	struct inode *mft_vi = vol->mft_ino;
930 	struct inode *vi;
931 	ntfs_inode *ni, *eni, **extent_nis;
932 	int i;
933 	ntfs_attr na;
934 
935 	ntfs_debug("Entering for inode 0x%lx.", mft_no);
936 	/*
937 	 * Normally we do not return a locked inode so set @locked_ni to NULL.
938 	 */
939 	BUG_ON(!locked_ni);
940 	*locked_ni = NULL;
941 	/*
942 	 * Check if the inode corresponding to this mft record is in the VFS
943 	 * inode cache and obtain a reference to it if it is.
944 	 */
945 	ntfs_debug("Looking for inode 0x%lx in icache.", mft_no);
946 	na.mft_no = mft_no;
947 	na.name = NULL;
948 	na.name_len = 0;
949 	na.type = AT_UNUSED;
950 	/*
951 	 * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
952 	 * we get here for it rather often.
953 	 */
954 	if (!mft_no) {
955 		/* Balance the below iput(). */
956 		vi = igrab(mft_vi);
957 		BUG_ON(vi != mft_vi);
958 	} else {
959 		/*
960 		 * Have to use ilookup5_nowait() since ilookup5() waits for the
961 		 * inode lock which causes ntfs to deadlock when a concurrent
962 		 * inode write via the inode dirty code paths and the page
963 		 * dirty code path of the inode dirty code path when writing
964 		 * $MFT occurs.
965 		 */
966 		vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
967 	}
968 	if (vi) {
969 		ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
970 		/* The inode is in icache. */
971 		ni = NTFS_I(vi);
972 		/* Take a reference to the ntfs inode. */
973 		atomic_inc(&ni->count);
974 		/* If the inode is dirty, do not write this record. */
975 		if (NInoDirty(ni)) {
976 			ntfs_debug("Inode 0x%lx is dirty, do not write it.",
977 					mft_no);
978 			atomic_dec(&ni->count);
979 			iput(vi);
980 			return false;
981 		}
982 		ntfs_debug("Inode 0x%lx is not dirty.", mft_no);
983 		/* The inode is not dirty, try to take the mft record lock. */
984 		if (unlikely(!mutex_trylock(&ni->mrec_lock))) {
985 			ntfs_debug("Mft record 0x%lx is already locked, do "
986 					"not write it.", mft_no);
987 			atomic_dec(&ni->count);
988 			iput(vi);
989 			return false;
990 		}
991 		ntfs_debug("Managed to lock mft record 0x%lx, write it.",
992 				mft_no);
993 		/*
994 		 * The write has to occur while we hold the mft record lock so
995 		 * return the locked ntfs inode.
996 		 */
997 		*locked_ni = ni;
998 		return true;
999 	}
1000 	ntfs_debug("Inode 0x%lx is not in icache.", mft_no);
1001 	/* The inode is not in icache. */
1002 	/* Write the record if it is not a mft record (type "FILE"). */
1003 	if (!ntfs_is_mft_record(m->magic)) {
1004 		ntfs_debug("Mft record 0x%lx is not a FILE record, write it.",
1005 				mft_no);
1006 		return true;
1007 	}
1008 	/* Write the mft record if it is a base inode. */
1009 	if (!m->base_mft_record) {
1010 		ntfs_debug("Mft record 0x%lx is a base record, write it.",
1011 				mft_no);
1012 		return true;
1013 	}
1014 	/*
1015 	 * This is an extent mft record.  Check if the inode corresponding to
1016 	 * its base mft record is in icache and obtain a reference to it if it
1017 	 * is.
1018 	 */
1019 	na.mft_no = MREF_LE(m->base_mft_record);
1020 	ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
1021 			"inode 0x%lx in icache.", mft_no, na.mft_no);
1022 	if (!na.mft_no) {
1023 		/* Balance the below iput(). */
1024 		vi = igrab(mft_vi);
1025 		BUG_ON(vi != mft_vi);
1026 	} else
1027 		vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
1028 				&na);
1029 	if (!vi) {
1030 		/*
1031 		 * The base inode is not in icache, write this extent mft
1032 		 * record.
1033 		 */
1034 		ntfs_debug("Base inode 0x%lx is not in icache, write the "
1035 				"extent record.", na.mft_no);
1036 		return true;
1037 	}
1038 	ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no);
1039 	/*
1040 	 * The base inode is in icache.  Check if it has the extent inode
1041 	 * corresponding to this extent mft record attached.
1042 	 */
1043 	ni = NTFS_I(vi);
1044 	mutex_lock(&ni->extent_lock);
1045 	if (ni->nr_extents <= 0) {
1046 		/*
1047 		 * The base inode has no attached extent inodes, write this
1048 		 * extent mft record.
1049 		 */
1050 		mutex_unlock(&ni->extent_lock);
1051 		iput(vi);
1052 		ntfs_debug("Base inode 0x%lx has no attached extent inodes, "
1053 				"write the extent record.", na.mft_no);
1054 		return true;
1055 	}
1056 	/* Iterate over the attached extent inodes. */
1057 	extent_nis = ni->ext.extent_ntfs_inos;
1058 	for (eni = NULL, i = 0; i < ni->nr_extents; ++i) {
1059 		if (mft_no == extent_nis[i]->mft_no) {
1060 			/*
1061 			 * Found the extent inode corresponding to this extent
1062 			 * mft record.
1063 			 */
1064 			eni = extent_nis[i];
1065 			break;
1066 		}
1067 	}
1068 	/*
1069 	 * If the extent inode was not attached to the base inode, write this
1070 	 * extent mft record.
1071 	 */
1072 	if (!eni) {
1073 		mutex_unlock(&ni->extent_lock);
1074 		iput(vi);
1075 		ntfs_debug("Extent inode 0x%lx is not attached to its base "
1076 				"inode 0x%lx, write the extent record.",
1077 				mft_no, na.mft_no);
1078 		return true;
1079 	}
1080 	ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.",
1081 			mft_no, na.mft_no);
1082 	/* Take a reference to the extent ntfs inode. */
1083 	atomic_inc(&eni->count);
1084 	mutex_unlock(&ni->extent_lock);
1085 	/*
1086 	 * Found the extent inode coresponding to this extent mft record.
1087 	 * Try to take the mft record lock.
1088 	 */
1089 	if (unlikely(!mutex_trylock(&eni->mrec_lock))) {
1090 		atomic_dec(&eni->count);
1091 		iput(vi);
1092 		ntfs_debug("Extent mft record 0x%lx is already locked, do "
1093 				"not write it.", mft_no);
1094 		return false;
1095 	}
1096 	ntfs_debug("Managed to lock extent mft record 0x%lx, write it.",
1097 			mft_no);
1098 	if (NInoTestClearDirty(eni))
1099 		ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.",
1100 				mft_no);
1101 	/*
1102 	 * The write has to occur while we hold the mft record lock so return
1103 	 * the locked extent ntfs inode.
1104 	 */
1105 	*locked_ni = eni;
1106 	return true;
1107 }
1108 
1109 static const char *es = "  Leaving inconsistent metadata.  Unmount and run "
1110 		"chkdsk.";
1111 
1112 /**
1113  * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name
1114  * @vol:	volume on which to search for a free mft record
1115  * @base_ni:	open base inode if allocating an extent mft record or NULL
1116  *
1117  * Search for a free mft record in the mft bitmap attribute on the ntfs volume
1118  * @vol.
1119  *
1120  * If @base_ni is NULL start the search at the default allocator position.
1121  *
1122  * If @base_ni is not NULL start the search at the mft record after the base
1123  * mft record @base_ni.
1124  *
1125  * Return the free mft record on success and -errno on error.  An error code of
1126  * -ENOSPC means that there are no free mft records in the currently
1127  * initialized mft bitmap.
1128  *
1129  * Locking: Caller must hold vol->mftbmp_lock for writing.
1130  */
ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume * vol,ntfs_inode * base_ni)1131 static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol,
1132 		ntfs_inode *base_ni)
1133 {
1134 	s64 pass_end, ll, data_pos, pass_start, ofs, bit;
1135 	unsigned long flags;
1136 	struct address_space *mftbmp_mapping;
1137 	u8 *buf, *byte;
1138 	struct page *page;
1139 	unsigned int page_ofs, size;
1140 	u8 pass, b;
1141 
1142 	ntfs_debug("Searching for free mft record in the currently "
1143 			"initialized mft bitmap.");
1144 	mftbmp_mapping = vol->mftbmp_ino->i_mapping;
1145 	/*
1146 	 * Set the end of the pass making sure we do not overflow the mft
1147 	 * bitmap.
1148 	 */
1149 	read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags);
1150 	pass_end = NTFS_I(vol->mft_ino)->allocated_size >>
1151 			vol->mft_record_size_bits;
1152 	read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags);
1153 	read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1154 	ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3;
1155 	read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags);
1156 	if (pass_end > ll)
1157 		pass_end = ll;
1158 	pass = 1;
1159 	if (!base_ni)
1160 		data_pos = vol->mft_data_pos;
1161 	else
1162 		data_pos = base_ni->mft_no + 1;
1163 	if (data_pos < 24)
1164 		data_pos = 24;
1165 	if (data_pos >= pass_end) {
1166 		data_pos = 24;
1167 		pass = 2;
1168 		/* This happens on a freshly formatted volume. */
1169 		if (data_pos >= pass_end)
1170 			return -ENOSPC;
1171 	}
1172 	pass_start = data_pos;
1173 	ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, "
1174 			"pass_end 0x%llx, data_pos 0x%llx.", pass,
1175 			(long long)pass_start, (long long)pass_end,
1176 			(long long)data_pos);
1177 	/* Loop until a free mft record is found. */
1178 	for (; pass <= 2;) {
1179 		/* Cap size to pass_end. */
1180 		ofs = data_pos >> 3;
1181 		page_ofs = ofs & ~PAGE_CACHE_MASK;
1182 		size = PAGE_CACHE_SIZE - page_ofs;
1183 		ll = ((pass_end + 7) >> 3) - ofs;
1184 		if (size > ll)
1185 			size = ll;
1186 		size <<= 3;
1187 		/*
1188 		 * If we are still within the active pass, search the next page
1189 		 * for a zero bit.
1190 		 */
1191 		if (size) {
1192 			page = ntfs_map_page(mftbmp_mapping,
1193 					ofs >> PAGE_CACHE_SHIFT);
1194 			if (IS_ERR(page)) {
1195 				ntfs_error(vol->sb, "Failed to read mft "
1196 						"bitmap, aborting.");
1197 				return PTR_ERR(page);
1198 			}
1199 			buf = (u8*)page_address(page) + page_ofs;
1200 			bit = data_pos & 7;
1201 			data_pos &= ~7ull;
1202 			ntfs_debug("Before inner for loop: size 0x%x, "
1203 					"data_pos 0x%llx, bit 0x%llx", size,
1204 					(long long)data_pos, (long long)bit);
1205 			for (; bit < size && data_pos + bit < pass_end;
1206 					bit &= ~7ull, bit += 8) {
1207 				byte = buf + (bit >> 3);
1208 				if (*byte == 0xff)
1209 					continue;
1210 				b = ffz((unsigned long)*byte);
1211 				if (b < 8 && b >= (bit & 7)) {
1212 					ll = data_pos + (bit & ~7ull) + b;
1213 					if (unlikely(ll > (1ll << 32))) {
1214 						ntfs_unmap_page(page);
1215 						return -ENOSPC;
1216 					}
1217 					*byte |= 1 << b;
1218 					flush_dcache_page(page);
1219 					set_page_dirty(page);
1220 					ntfs_unmap_page(page);
1221 					ntfs_debug("Done.  (Found and "
1222 							"allocated mft record "
1223 							"0x%llx.)",
1224 							(long long)ll);
1225 					return ll;
1226 				}
1227 			}
1228 			ntfs_debug("After inner for loop: size 0x%x, "
1229 					"data_pos 0x%llx, bit 0x%llx", size,
1230 					(long long)data_pos, (long long)bit);
1231 			data_pos += size;
1232 			ntfs_unmap_page(page);
1233 			/*
1234 			 * If the end of the pass has not been reached yet,
1235 			 * continue searching the mft bitmap for a zero bit.
1236 			 */
1237 			if (data_pos < pass_end)
1238 				continue;
1239 		}
1240 		/* Do the next pass. */
1241 		if (++pass == 2) {
1242 			/*
1243 			 * Starting the second pass, in which we scan the first
1244 			 * part of the zone which we omitted earlier.
1245 			 */
1246 			pass_end = pass_start;
1247 			data_pos = pass_start = 24;
1248 			ntfs_debug("pass %i, pass_start 0x%llx, pass_end "
1249 					"0x%llx.", pass, (long long)pass_start,
1250 					(long long)pass_end);
1251 			if (data_pos >= pass_end)
1252 				break;
1253 		}
1254 	}
1255 	/* No free mft records in currently initialized mft bitmap. */
1256 	ntfs_debug("Done.  (No free mft records left in currently initialized "
1257 			"mft bitmap.)");
1258 	return -ENOSPC;
1259 }
1260 
1261 /**
1262  * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster
1263  * @vol:	volume on which to extend the mft bitmap attribute
1264  *
1265  * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster.
1266  *
1267  * Note: Only changes allocated_size, i.e. does not touch initialized_size or
1268  * data_size.
1269  *
1270  * Return 0 on success and -errno on error.
1271  *
1272  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1273  *	    - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for
1274  *	      writing and releases it before returning.
1275  *	    - This function takes vol->lcnbmp_lock for writing and releases it
1276  *	      before returning.
1277  */
ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume * vol)1278 static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol)
1279 {
1280 	LCN lcn;
1281 	s64 ll;
1282 	unsigned long flags;
1283 	struct page *page;
1284 	ntfs_inode *mft_ni, *mftbmp_ni;
1285 	runlist_element *rl, *rl2 = NULL;
1286 	ntfs_attr_search_ctx *ctx = NULL;
1287 	MFT_RECORD *mrec;
1288 	ATTR_RECORD *a = NULL;
1289 	int ret, mp_size;
1290 	u32 old_alen = 0;
1291 	u8 *b, tb;
1292 	struct {
1293 		u8 added_cluster:1;
1294 		u8 added_run:1;
1295 		u8 mp_rebuilt:1;
1296 	} status = { 0, 0, 0 };
1297 
1298 	ntfs_debug("Extending mft bitmap allocation.");
1299 	mft_ni = NTFS_I(vol->mft_ino);
1300 	mftbmp_ni = NTFS_I(vol->mftbmp_ino);
1301 	/*
1302 	 * Determine the last lcn of the mft bitmap.  The allocated size of the
1303 	 * mft bitmap cannot be zero so we are ok to do this.
1304 	 */
1305 	down_write(&mftbmp_ni->runlist.lock);
1306 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1307 	ll = mftbmp_ni->allocated_size;
1308 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1309 	rl = ntfs_attr_find_vcn_nolock(mftbmp_ni,
1310 			(ll - 1) >> vol->cluster_size_bits, NULL);
1311 	if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1312 		up_write(&mftbmp_ni->runlist.lock);
1313 		ntfs_error(vol->sb, "Failed to determine last allocated "
1314 				"cluster of mft bitmap attribute.");
1315 		if (!IS_ERR(rl))
1316 			ret = -EIO;
1317 		else
1318 			ret = PTR_ERR(rl);
1319 		return ret;
1320 	}
1321 	lcn = rl->lcn + rl->length;
1322 	ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.",
1323 			(long long)lcn);
1324 	/*
1325 	 * Attempt to get the cluster following the last allocated cluster by
1326 	 * hand as it may be in the MFT zone so the allocator would not give it
1327 	 * to us.
1328 	 */
1329 	ll = lcn >> 3;
1330 	page = ntfs_map_page(vol->lcnbmp_ino->i_mapping,
1331 			ll >> PAGE_CACHE_SHIFT);
1332 	if (IS_ERR(page)) {
1333 		up_write(&mftbmp_ni->runlist.lock);
1334 		ntfs_error(vol->sb, "Failed to read from lcn bitmap.");
1335 		return PTR_ERR(page);
1336 	}
1337 	b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK);
1338 	tb = 1 << (lcn & 7ull);
1339 	down_write(&vol->lcnbmp_lock);
1340 	if (*b != 0xff && !(*b & tb)) {
1341 		/* Next cluster is free, allocate it. */
1342 		*b |= tb;
1343 		flush_dcache_page(page);
1344 		set_page_dirty(page);
1345 		up_write(&vol->lcnbmp_lock);
1346 		ntfs_unmap_page(page);
1347 		/* Update the mft bitmap runlist. */
1348 		rl->length++;
1349 		rl[1].vcn++;
1350 		status.added_cluster = 1;
1351 		ntfs_debug("Appending one cluster to mft bitmap.");
1352 	} else {
1353 		up_write(&vol->lcnbmp_lock);
1354 		ntfs_unmap_page(page);
1355 		/* Allocate a cluster from the DATA_ZONE. */
1356 		rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE,
1357 				true);
1358 		if (IS_ERR(rl2)) {
1359 			up_write(&mftbmp_ni->runlist.lock);
1360 			ntfs_error(vol->sb, "Failed to allocate a cluster for "
1361 					"the mft bitmap.");
1362 			return PTR_ERR(rl2);
1363 		}
1364 		rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2);
1365 		if (IS_ERR(rl)) {
1366 			up_write(&mftbmp_ni->runlist.lock);
1367 			ntfs_error(vol->sb, "Failed to merge runlists for mft "
1368 					"bitmap.");
1369 			if (ntfs_cluster_free_from_rl(vol, rl2)) {
1370 				ntfs_error(vol->sb, "Failed to deallocate "
1371 						"allocated cluster.%s", es);
1372 				NVolSetErrors(vol);
1373 			}
1374 			ntfs_free(rl2);
1375 			return PTR_ERR(rl);
1376 		}
1377 		mftbmp_ni->runlist.rl = rl;
1378 		status.added_run = 1;
1379 		ntfs_debug("Adding one run to mft bitmap.");
1380 		/* Find the last run in the new runlist. */
1381 		for (; rl[1].length; rl++)
1382 			;
1383 	}
1384 	/*
1385 	 * Update the attribute record as well.  Note: @rl is the last
1386 	 * (non-terminator) runlist element of mft bitmap.
1387 	 */
1388 	mrec = map_mft_record(mft_ni);
1389 	if (IS_ERR(mrec)) {
1390 		ntfs_error(vol->sb, "Failed to map mft record.");
1391 		ret = PTR_ERR(mrec);
1392 		goto undo_alloc;
1393 	}
1394 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1395 	if (unlikely(!ctx)) {
1396 		ntfs_error(vol->sb, "Failed to get search context.");
1397 		ret = -ENOMEM;
1398 		goto undo_alloc;
1399 	}
1400 	ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1401 			mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1402 			0, ctx);
1403 	if (unlikely(ret)) {
1404 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1405 				"mft bitmap attribute.");
1406 		if (ret == -ENOENT)
1407 			ret = -EIO;
1408 		goto undo_alloc;
1409 	}
1410 	a = ctx->attr;
1411 	ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1412 	/* Search back for the previous last allocated cluster of mft bitmap. */
1413 	for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) {
1414 		if (ll >= rl2->vcn)
1415 			break;
1416 	}
1417 	BUG_ON(ll < rl2->vcn);
1418 	BUG_ON(ll >= rl2->vcn + rl2->length);
1419 	/* Get the size for the new mapping pairs array for this extent. */
1420 	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1421 	if (unlikely(mp_size <= 0)) {
1422 		ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1423 				"mft bitmap attribute extent.");
1424 		ret = mp_size;
1425 		if (!ret)
1426 			ret = -EIO;
1427 		goto undo_alloc;
1428 	}
1429 	/* Expand the attribute record if necessary. */
1430 	old_alen = le32_to_cpu(a->length);
1431 	ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1432 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1433 	if (unlikely(ret)) {
1434 		if (ret != -ENOSPC) {
1435 			ntfs_error(vol->sb, "Failed to resize attribute "
1436 					"record for mft bitmap attribute.");
1437 			goto undo_alloc;
1438 		}
1439 		// TODO: Deal with this by moving this extent to a new mft
1440 		// record or by starting a new extent in a new mft record or by
1441 		// moving other attributes out of this mft record.
1442 		// Note: It will need to be a special mft record and if none of
1443 		// those are available it gets rather complicated...
1444 		ntfs_error(vol->sb, "Not enough space in this mft record to "
1445 				"accommodate extended mft bitmap attribute "
1446 				"extent.  Cannot handle this yet.");
1447 		ret = -EOPNOTSUPP;
1448 		goto undo_alloc;
1449 	}
1450 	status.mp_rebuilt = 1;
1451 	/* Generate the mapping pairs array directly into the attr record. */
1452 	ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1453 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1454 			mp_size, rl2, ll, -1, NULL);
1455 	if (unlikely(ret)) {
1456 		ntfs_error(vol->sb, "Failed to build mapping pairs array for "
1457 				"mft bitmap attribute.");
1458 		goto undo_alloc;
1459 	}
1460 	/* Update the highest_vcn. */
1461 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1462 	/*
1463 	 * We now have extended the mft bitmap allocated_size by one cluster.
1464 	 * Reflect this in the ntfs_inode structure and the attribute record.
1465 	 */
1466 	if (a->data.non_resident.lowest_vcn) {
1467 		/*
1468 		 * We are not in the first attribute extent, switch to it, but
1469 		 * first ensure the changes will make it to disk later.
1470 		 */
1471 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1472 		mark_mft_record_dirty(ctx->ntfs_ino);
1473 		ntfs_attr_reinit_search_ctx(ctx);
1474 		ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1475 				mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL,
1476 				0, ctx);
1477 		if (unlikely(ret)) {
1478 			ntfs_error(vol->sb, "Failed to find first attribute "
1479 					"extent of mft bitmap attribute.");
1480 			goto restore_undo_alloc;
1481 		}
1482 		a = ctx->attr;
1483 	}
1484 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1485 	mftbmp_ni->allocated_size += vol->cluster_size;
1486 	a->data.non_resident.allocated_size =
1487 			cpu_to_sle64(mftbmp_ni->allocated_size);
1488 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1489 	/* Ensure the changes make it to disk. */
1490 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1491 	mark_mft_record_dirty(ctx->ntfs_ino);
1492 	ntfs_attr_put_search_ctx(ctx);
1493 	unmap_mft_record(mft_ni);
1494 	up_write(&mftbmp_ni->runlist.lock);
1495 	ntfs_debug("Done.");
1496 	return 0;
1497 restore_undo_alloc:
1498 	ntfs_attr_reinit_search_ctx(ctx);
1499 	if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1500 			mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL,
1501 			0, ctx)) {
1502 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1503 				"mft bitmap attribute.%s", es);
1504 		write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1505 		mftbmp_ni->allocated_size += vol->cluster_size;
1506 		write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1507 		ntfs_attr_put_search_ctx(ctx);
1508 		unmap_mft_record(mft_ni);
1509 		up_write(&mftbmp_ni->runlist.lock);
1510 		/*
1511 		 * The only thing that is now wrong is ->allocated_size of the
1512 		 * base attribute extent which chkdsk should be able to fix.
1513 		 */
1514 		NVolSetErrors(vol);
1515 		return ret;
1516 	}
1517 	a = ctx->attr;
1518 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2);
1519 undo_alloc:
1520 	if (status.added_cluster) {
1521 		/* Truncate the last run in the runlist by one cluster. */
1522 		rl->length--;
1523 		rl[1].vcn--;
1524 	} else if (status.added_run) {
1525 		lcn = rl->lcn;
1526 		/* Remove the last run from the runlist. */
1527 		rl->lcn = rl[1].lcn;
1528 		rl->length = 0;
1529 	}
1530 	/* Deallocate the cluster. */
1531 	down_write(&vol->lcnbmp_lock);
1532 	if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1533 		ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es);
1534 		NVolSetErrors(vol);
1535 	}
1536 	up_write(&vol->lcnbmp_lock);
1537 	if (status.mp_rebuilt) {
1538 		if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1539 				a->data.non_resident.mapping_pairs_offset),
1540 				old_alen - le16_to_cpu(
1541 				a->data.non_resident.mapping_pairs_offset),
1542 				rl2, ll, -1, NULL)) {
1543 			ntfs_error(vol->sb, "Failed to restore mapping pairs "
1544 					"array.%s", es);
1545 			NVolSetErrors(vol);
1546 		}
1547 		if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1548 			ntfs_error(vol->sb, "Failed to restore attribute "
1549 					"record.%s", es);
1550 			NVolSetErrors(vol);
1551 		}
1552 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1553 		mark_mft_record_dirty(ctx->ntfs_ino);
1554 	}
1555 	if (ctx)
1556 		ntfs_attr_put_search_ctx(ctx);
1557 	if (!IS_ERR(mrec))
1558 		unmap_mft_record(mft_ni);
1559 	up_write(&mftbmp_ni->runlist.lock);
1560 	return ret;
1561 }
1562 
1563 /**
1564  * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data
1565  * @vol:	volume on which to extend the mft bitmap attribute
1566  *
1567  * Extend the initialized portion of the mft bitmap attribute on the ntfs
1568  * volume @vol by 8 bytes.
1569  *
1570  * Note:  Only changes initialized_size and data_size, i.e. requires that
1571  * allocated_size is big enough to fit the new initialized_size.
1572  *
1573  * Return 0 on success and -error on error.
1574  *
1575  * Locking: Caller must hold vol->mftbmp_lock for writing.
1576  */
ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume * vol)1577 static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol)
1578 {
1579 	s64 old_data_size, old_initialized_size;
1580 	unsigned long flags;
1581 	struct inode *mftbmp_vi;
1582 	ntfs_inode *mft_ni, *mftbmp_ni;
1583 	ntfs_attr_search_ctx *ctx;
1584 	MFT_RECORD *mrec;
1585 	ATTR_RECORD *a;
1586 	int ret;
1587 
1588 	ntfs_debug("Extending mft bitmap initiailized (and data) size.");
1589 	mft_ni = NTFS_I(vol->mft_ino);
1590 	mftbmp_vi = vol->mftbmp_ino;
1591 	mftbmp_ni = NTFS_I(mftbmp_vi);
1592 	/* Get the attribute record. */
1593 	mrec = map_mft_record(mft_ni);
1594 	if (IS_ERR(mrec)) {
1595 		ntfs_error(vol->sb, "Failed to map mft record.");
1596 		return PTR_ERR(mrec);
1597 	}
1598 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1599 	if (unlikely(!ctx)) {
1600 		ntfs_error(vol->sb, "Failed to get search context.");
1601 		ret = -ENOMEM;
1602 		goto unm_err_out;
1603 	}
1604 	ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1605 			mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx);
1606 	if (unlikely(ret)) {
1607 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
1608 				"mft bitmap attribute.");
1609 		if (ret == -ENOENT)
1610 			ret = -EIO;
1611 		goto put_err_out;
1612 	}
1613 	a = ctx->attr;
1614 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1615 	old_data_size = i_size_read(mftbmp_vi);
1616 	old_initialized_size = mftbmp_ni->initialized_size;
1617 	/*
1618 	 * We can simply update the initialized_size before filling the space
1619 	 * with zeroes because the caller is holding the mft bitmap lock for
1620 	 * writing which ensures that no one else is trying to access the data.
1621 	 */
1622 	mftbmp_ni->initialized_size += 8;
1623 	a->data.non_resident.initialized_size =
1624 			cpu_to_sle64(mftbmp_ni->initialized_size);
1625 	if (mftbmp_ni->initialized_size > old_data_size) {
1626 		i_size_write(mftbmp_vi, mftbmp_ni->initialized_size);
1627 		a->data.non_resident.data_size =
1628 				cpu_to_sle64(mftbmp_ni->initialized_size);
1629 	}
1630 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1631 	/* Ensure the changes make it to disk. */
1632 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1633 	mark_mft_record_dirty(ctx->ntfs_ino);
1634 	ntfs_attr_put_search_ctx(ctx);
1635 	unmap_mft_record(mft_ni);
1636 	/* Initialize the mft bitmap attribute value with zeroes. */
1637 	ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0);
1638 	if (likely(!ret)) {
1639 		ntfs_debug("Done.  (Wrote eight initialized bytes to mft "
1640 				"bitmap.");
1641 		return 0;
1642 	}
1643 	ntfs_error(vol->sb, "Failed to write to mft bitmap.");
1644 	/* Try to recover from the error. */
1645 	mrec = map_mft_record(mft_ni);
1646 	if (IS_ERR(mrec)) {
1647 		ntfs_error(vol->sb, "Failed to map mft record.%s", es);
1648 		NVolSetErrors(vol);
1649 		return ret;
1650 	}
1651 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1652 	if (unlikely(!ctx)) {
1653 		ntfs_error(vol->sb, "Failed to get search context.%s", es);
1654 		NVolSetErrors(vol);
1655 		goto unm_err_out;
1656 	}
1657 	if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name,
1658 			mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) {
1659 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
1660 				"mft bitmap attribute.%s", es);
1661 		NVolSetErrors(vol);
1662 put_err_out:
1663 		ntfs_attr_put_search_ctx(ctx);
1664 unm_err_out:
1665 		unmap_mft_record(mft_ni);
1666 		goto err_out;
1667 	}
1668 	a = ctx->attr;
1669 	write_lock_irqsave(&mftbmp_ni->size_lock, flags);
1670 	mftbmp_ni->initialized_size = old_initialized_size;
1671 	a->data.non_resident.initialized_size =
1672 			cpu_to_sle64(old_initialized_size);
1673 	if (i_size_read(mftbmp_vi) != old_data_size) {
1674 		i_size_write(mftbmp_vi, old_data_size);
1675 		a->data.non_resident.data_size = cpu_to_sle64(old_data_size);
1676 	}
1677 	write_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1678 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1679 	mark_mft_record_dirty(ctx->ntfs_ino);
1680 	ntfs_attr_put_search_ctx(ctx);
1681 	unmap_mft_record(mft_ni);
1682 #ifdef DEBUG
1683 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
1684 	ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, "
1685 			"data_size 0x%llx, initialized_size 0x%llx.",
1686 			(long long)mftbmp_ni->allocated_size,
1687 			(long long)i_size_read(mftbmp_vi),
1688 			(long long)mftbmp_ni->initialized_size);
1689 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
1690 #endif /* DEBUG */
1691 err_out:
1692 	return ret;
1693 }
1694 
1695 /**
1696  * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute
1697  * @vol:	volume on which to extend the mft data attribute
1698  *
1699  * Extend the mft data attribute on the ntfs volume @vol by 16 mft records
1700  * worth of clusters or if not enough space for this by one mft record worth
1701  * of clusters.
1702  *
1703  * Note:  Only changes allocated_size, i.e. does not touch initialized_size or
1704  * data_size.
1705  *
1706  * Return 0 on success and -errno on error.
1707  *
1708  * Locking: - Caller must hold vol->mftbmp_lock for writing.
1709  *	    - This function takes NTFS_I(vol->mft_ino)->runlist.lock for
1710  *	      writing and releases it before returning.
1711  *	    - This function calls functions which take vol->lcnbmp_lock for
1712  *	      writing and release it before returning.
1713  */
ntfs_mft_data_extend_allocation_nolock(ntfs_volume * vol)1714 static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol)
1715 {
1716 	LCN lcn;
1717 	VCN old_last_vcn;
1718 	s64 min_nr, nr, ll;
1719 	unsigned long flags;
1720 	ntfs_inode *mft_ni;
1721 	runlist_element *rl, *rl2;
1722 	ntfs_attr_search_ctx *ctx = NULL;
1723 	MFT_RECORD *mrec;
1724 	ATTR_RECORD *a = NULL;
1725 	int ret, mp_size;
1726 	u32 old_alen = 0;
1727 	bool mp_rebuilt = false;
1728 
1729 	ntfs_debug("Extending mft data allocation.");
1730 	mft_ni = NTFS_I(vol->mft_ino);
1731 	/*
1732 	 * Determine the preferred allocation location, i.e. the last lcn of
1733 	 * the mft data attribute.  The allocated size of the mft data
1734 	 * attribute cannot be zero so we are ok to do this.
1735 	 */
1736 	down_write(&mft_ni->runlist.lock);
1737 	read_lock_irqsave(&mft_ni->size_lock, flags);
1738 	ll = mft_ni->allocated_size;
1739 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
1740 	rl = ntfs_attr_find_vcn_nolock(mft_ni,
1741 			(ll - 1) >> vol->cluster_size_bits, NULL);
1742 	if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) {
1743 		up_write(&mft_ni->runlist.lock);
1744 		ntfs_error(vol->sb, "Failed to determine last allocated "
1745 				"cluster of mft data attribute.");
1746 		if (!IS_ERR(rl))
1747 			ret = -EIO;
1748 		else
1749 			ret = PTR_ERR(rl);
1750 		return ret;
1751 	}
1752 	lcn = rl->lcn + rl->length;
1753 	ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn);
1754 	/* Minimum allocation is one mft record worth of clusters. */
1755 	min_nr = vol->mft_record_size >> vol->cluster_size_bits;
1756 	if (!min_nr)
1757 		min_nr = 1;
1758 	/* Want to allocate 16 mft records worth of clusters. */
1759 	nr = vol->mft_record_size << 4 >> vol->cluster_size_bits;
1760 	if (!nr)
1761 		nr = min_nr;
1762 	/* Ensure we do not go above 2^32-1 mft records. */
1763 	read_lock_irqsave(&mft_ni->size_lock, flags);
1764 	ll = mft_ni->allocated_size;
1765 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
1766 	if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1767 			vol->mft_record_size_bits >= (1ll << 32))) {
1768 		nr = min_nr;
1769 		if (unlikely((ll + (nr << vol->cluster_size_bits)) >>
1770 				vol->mft_record_size_bits >= (1ll << 32))) {
1771 			ntfs_warning(vol->sb, "Cannot allocate mft record "
1772 					"because the maximum number of inodes "
1773 					"(2^32) has already been reached.");
1774 			up_write(&mft_ni->runlist.lock);
1775 			return -ENOSPC;
1776 		}
1777 	}
1778 	ntfs_debug("Trying mft data allocation with %s cluster count %lli.",
1779 			nr > min_nr ? "default" : "minimal", (long long)nr);
1780 	old_last_vcn = rl[1].vcn;
1781 	do {
1782 		rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE,
1783 				true);
1784 		if (likely(!IS_ERR(rl2)))
1785 			break;
1786 		if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) {
1787 			ntfs_error(vol->sb, "Failed to allocate the minimal "
1788 					"number of clusters (%lli) for the "
1789 					"mft data attribute.", (long long)nr);
1790 			up_write(&mft_ni->runlist.lock);
1791 			return PTR_ERR(rl2);
1792 		}
1793 		/*
1794 		 * There is not enough space to do the allocation, but there
1795 		 * might be enough space to do a minimal allocation so try that
1796 		 * before failing.
1797 		 */
1798 		nr = min_nr;
1799 		ntfs_debug("Retrying mft data allocation with minimal cluster "
1800 				"count %lli.", (long long)nr);
1801 	} while (1);
1802 	rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2);
1803 	if (IS_ERR(rl)) {
1804 		up_write(&mft_ni->runlist.lock);
1805 		ntfs_error(vol->sb, "Failed to merge runlists for mft data "
1806 				"attribute.");
1807 		if (ntfs_cluster_free_from_rl(vol, rl2)) {
1808 			ntfs_error(vol->sb, "Failed to deallocate clusters "
1809 					"from the mft data attribute.%s", es);
1810 			NVolSetErrors(vol);
1811 		}
1812 		ntfs_free(rl2);
1813 		return PTR_ERR(rl);
1814 	}
1815 	mft_ni->runlist.rl = rl;
1816 	ntfs_debug("Allocated %lli clusters.", (long long)nr);
1817 	/* Find the last run in the new runlist. */
1818 	for (; rl[1].length; rl++)
1819 		;
1820 	/* Update the attribute record as well. */
1821 	mrec = map_mft_record(mft_ni);
1822 	if (IS_ERR(mrec)) {
1823 		ntfs_error(vol->sb, "Failed to map mft record.");
1824 		ret = PTR_ERR(mrec);
1825 		goto undo_alloc;
1826 	}
1827 	ctx = ntfs_attr_get_search_ctx(mft_ni, mrec);
1828 	if (unlikely(!ctx)) {
1829 		ntfs_error(vol->sb, "Failed to get search context.");
1830 		ret = -ENOMEM;
1831 		goto undo_alloc;
1832 	}
1833 	ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1834 			CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx);
1835 	if (unlikely(ret)) {
1836 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1837 				"mft data attribute.");
1838 		if (ret == -ENOENT)
1839 			ret = -EIO;
1840 		goto undo_alloc;
1841 	}
1842 	a = ctx->attr;
1843 	ll = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1844 	/* Search back for the previous last allocated cluster of mft bitmap. */
1845 	for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) {
1846 		if (ll >= rl2->vcn)
1847 			break;
1848 	}
1849 	BUG_ON(ll < rl2->vcn);
1850 	BUG_ON(ll >= rl2->vcn + rl2->length);
1851 	/* Get the size for the new mapping pairs array for this extent. */
1852 	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1);
1853 	if (unlikely(mp_size <= 0)) {
1854 		ntfs_error(vol->sb, "Get size for mapping pairs failed for "
1855 				"mft data attribute extent.");
1856 		ret = mp_size;
1857 		if (!ret)
1858 			ret = -EIO;
1859 		goto undo_alloc;
1860 	}
1861 	/* Expand the attribute record if necessary. */
1862 	old_alen = le32_to_cpu(a->length);
1863 	ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size +
1864 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
1865 	if (unlikely(ret)) {
1866 		if (ret != -ENOSPC) {
1867 			ntfs_error(vol->sb, "Failed to resize attribute "
1868 					"record for mft data attribute.");
1869 			goto undo_alloc;
1870 		}
1871 		// TODO: Deal with this by moving this extent to a new mft
1872 		// record or by starting a new extent in a new mft record or by
1873 		// moving other attributes out of this mft record.
1874 		// Note: Use the special reserved mft records and ensure that
1875 		// this extent is not required to find the mft record in
1876 		// question.  If no free special records left we would need to
1877 		// move an existing record away, insert ours in its place, and
1878 		// then place the moved record into the newly allocated space
1879 		// and we would then need to update all references to this mft
1880 		// record appropriately.  This is rather complicated...
1881 		ntfs_error(vol->sb, "Not enough space in this mft record to "
1882 				"accommodate extended mft data attribute "
1883 				"extent.  Cannot handle this yet.");
1884 		ret = -EOPNOTSUPP;
1885 		goto undo_alloc;
1886 	}
1887 	mp_rebuilt = true;
1888 	/* Generate the mapping pairs array directly into the attr record. */
1889 	ret = ntfs_mapping_pairs_build(vol, (u8*)a +
1890 			le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
1891 			mp_size, rl2, ll, -1, NULL);
1892 	if (unlikely(ret)) {
1893 		ntfs_error(vol->sb, "Failed to build mapping pairs array of "
1894 				"mft data attribute.");
1895 		goto undo_alloc;
1896 	}
1897 	/* Update the highest_vcn. */
1898 	a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1);
1899 	/*
1900 	 * We now have extended the mft data allocated_size by nr clusters.
1901 	 * Reflect this in the ntfs_inode structure and the attribute record.
1902 	 * @rl is the last (non-terminator) runlist element of mft data
1903 	 * attribute.
1904 	 */
1905 	if (a->data.non_resident.lowest_vcn) {
1906 		/*
1907 		 * We are not in the first attribute extent, switch to it, but
1908 		 * first ensure the changes will make it to disk later.
1909 		 */
1910 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1911 		mark_mft_record_dirty(ctx->ntfs_ino);
1912 		ntfs_attr_reinit_search_ctx(ctx);
1913 		ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name,
1914 				mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0,
1915 				ctx);
1916 		if (unlikely(ret)) {
1917 			ntfs_error(vol->sb, "Failed to find first attribute "
1918 					"extent of mft data attribute.");
1919 			goto restore_undo_alloc;
1920 		}
1921 		a = ctx->attr;
1922 	}
1923 	write_lock_irqsave(&mft_ni->size_lock, flags);
1924 	mft_ni->allocated_size += nr << vol->cluster_size_bits;
1925 	a->data.non_resident.allocated_size =
1926 			cpu_to_sle64(mft_ni->allocated_size);
1927 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
1928 	/* Ensure the changes make it to disk. */
1929 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1930 	mark_mft_record_dirty(ctx->ntfs_ino);
1931 	ntfs_attr_put_search_ctx(ctx);
1932 	unmap_mft_record(mft_ni);
1933 	up_write(&mft_ni->runlist.lock);
1934 	ntfs_debug("Done.");
1935 	return 0;
1936 restore_undo_alloc:
1937 	ntfs_attr_reinit_search_ctx(ctx);
1938 	if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
1939 			CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) {
1940 		ntfs_error(vol->sb, "Failed to find last attribute extent of "
1941 				"mft data attribute.%s", es);
1942 		write_lock_irqsave(&mft_ni->size_lock, flags);
1943 		mft_ni->allocated_size += nr << vol->cluster_size_bits;
1944 		write_unlock_irqrestore(&mft_ni->size_lock, flags);
1945 		ntfs_attr_put_search_ctx(ctx);
1946 		unmap_mft_record(mft_ni);
1947 		up_write(&mft_ni->runlist.lock);
1948 		/*
1949 		 * The only thing that is now wrong is ->allocated_size of the
1950 		 * base attribute extent which chkdsk should be able to fix.
1951 		 */
1952 		NVolSetErrors(vol);
1953 		return ret;
1954 	}
1955 	ctx->attr->data.non_resident.highest_vcn =
1956 			cpu_to_sle64(old_last_vcn - 1);
1957 undo_alloc:
1958 	if (ntfs_cluster_free(mft_ni, old_last_vcn, -1, ctx) < 0) {
1959 		ntfs_error(vol->sb, "Failed to free clusters from mft data "
1960 				"attribute.%s", es);
1961 		NVolSetErrors(vol);
1962 	}
1963 	a = ctx->attr;
1964 	if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) {
1965 		ntfs_error(vol->sb, "Failed to truncate mft data attribute "
1966 				"runlist.%s", es);
1967 		NVolSetErrors(vol);
1968 	}
1969 	if (mp_rebuilt && !IS_ERR(ctx->mrec)) {
1970 		if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1971 				a->data.non_resident.mapping_pairs_offset),
1972 				old_alen - le16_to_cpu(
1973 				a->data.non_resident.mapping_pairs_offset),
1974 				rl2, ll, -1, NULL)) {
1975 			ntfs_error(vol->sb, "Failed to restore mapping pairs "
1976 					"array.%s", es);
1977 			NVolSetErrors(vol);
1978 		}
1979 		if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) {
1980 			ntfs_error(vol->sb, "Failed to restore attribute "
1981 					"record.%s", es);
1982 			NVolSetErrors(vol);
1983 		}
1984 		flush_dcache_mft_record_page(ctx->ntfs_ino);
1985 		mark_mft_record_dirty(ctx->ntfs_ino);
1986 	} else if (IS_ERR(ctx->mrec)) {
1987 		ntfs_error(vol->sb, "Failed to restore attribute search "
1988 				"context.%s", es);
1989 		NVolSetErrors(vol);
1990 	}
1991 	if (ctx)
1992 		ntfs_attr_put_search_ctx(ctx);
1993 	if (!IS_ERR(mrec))
1994 		unmap_mft_record(mft_ni);
1995 	up_write(&mft_ni->runlist.lock);
1996 	return ret;
1997 }
1998 
1999 /**
2000  * ntfs_mft_record_layout - layout an mft record into a memory buffer
2001  * @vol:	volume to which the mft record will belong
2002  * @mft_no:	mft reference specifying the mft record number
2003  * @m:		destination buffer of size >= @vol->mft_record_size bytes
2004  *
2005  * Layout an empty, unused mft record with the mft record number @mft_no into
2006  * the buffer @m.  The volume @vol is needed because the mft record structure
2007  * was modified in NTFS 3.1 so we need to know which volume version this mft
2008  * record will be used on.
2009  *
2010  * Return 0 on success and -errno on error.
2011  */
ntfs_mft_record_layout(const ntfs_volume * vol,const s64 mft_no,MFT_RECORD * m)2012 static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no,
2013 		MFT_RECORD *m)
2014 {
2015 	ATTR_RECORD *a;
2016 
2017 	ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2018 	if (mft_no >= (1ll << 32)) {
2019 		ntfs_error(vol->sb, "Mft record number 0x%llx exceeds "
2020 				"maximum of 2^32.", (long long)mft_no);
2021 		return -ERANGE;
2022 	}
2023 	/* Start by clearing the whole mft record to gives us a clean slate. */
2024 	memset(m, 0, vol->mft_record_size);
2025 	/* Aligned to 2-byte boundary. */
2026 	if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver))
2027 		m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1);
2028 	else {
2029 		m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1);
2030 		/*
2031 		 * Set the NTFS 3.1+ specific fields while we know that the
2032 		 * volume version is 3.1+.
2033 		 */
2034 		m->reserved = 0;
2035 		m->mft_record_number = cpu_to_le32((u32)mft_no);
2036 	}
2037 	m->magic = magic_FILE;
2038 	if (vol->mft_record_size >= NTFS_BLOCK_SIZE)
2039 		m->usa_count = cpu_to_le16(vol->mft_record_size /
2040 				NTFS_BLOCK_SIZE + 1);
2041 	else {
2042 		m->usa_count = cpu_to_le16(1);
2043 		ntfs_warning(vol->sb, "Sector size is bigger than mft record "
2044 				"size.  Setting usa_count to 1.  If chkdsk "
2045 				"reports this as corruption, please email "
2046 				"linux-ntfs-dev@lists.sourceforge.net stating "
2047 				"that you saw this message and that the "
2048 				"modified filesystem created was corrupt.  "
2049 				"Thank you.");
2050 	}
2051 	/* Set the update sequence number to 1. */
2052 	*(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1);
2053 	m->lsn = 0;
2054 	m->sequence_number = cpu_to_le16(1);
2055 	m->link_count = 0;
2056 	/*
2057 	 * Place the attributes straight after the update sequence array,
2058 	 * aligned to 8-byte boundary.
2059 	 */
2060 	m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) +
2061 			(le16_to_cpu(m->usa_count) << 1) + 7) & ~7);
2062 	m->flags = 0;
2063 	/*
2064 	 * Using attrs_offset plus eight bytes (for the termination attribute).
2065 	 * attrs_offset is already aligned to 8-byte boundary, so no need to
2066 	 * align again.
2067 	 */
2068 	m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8);
2069 	m->bytes_allocated = cpu_to_le32(vol->mft_record_size);
2070 	m->base_mft_record = 0;
2071 	m->next_attr_instance = 0;
2072 	/* Add the termination attribute. */
2073 	a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset));
2074 	a->type = AT_END;
2075 	a->length = 0;
2076 	ntfs_debug("Done.");
2077 	return 0;
2078 }
2079 
2080 /**
2081  * ntfs_mft_record_format - format an mft record on an ntfs volume
2082  * @vol:	volume on which to format the mft record
2083  * @mft_no:	mft record number to format
2084  *
2085  * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused
2086  * mft record into the appropriate place of the mft data attribute.  This is
2087  * used when extending the mft data attribute.
2088  *
2089  * Return 0 on success and -errno on error.
2090  */
ntfs_mft_record_format(const ntfs_volume * vol,const s64 mft_no)2091 static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no)
2092 {
2093 	loff_t i_size;
2094 	struct inode *mft_vi = vol->mft_ino;
2095 	struct page *page;
2096 	MFT_RECORD *m;
2097 	pgoff_t index, end_index;
2098 	unsigned int ofs;
2099 	int err;
2100 
2101 	ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no);
2102 	/*
2103 	 * The index into the page cache and the offset within the page cache
2104 	 * page of the wanted mft record.
2105 	 */
2106 	index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2107 	ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2108 	/* The maximum valid index into the page cache for $MFT's data. */
2109 	i_size = i_size_read(mft_vi);
2110 	end_index = i_size >> PAGE_CACHE_SHIFT;
2111 	if (unlikely(index >= end_index)) {
2112 		if (unlikely(index > end_index || ofs + vol->mft_record_size >=
2113 				(i_size & ~PAGE_CACHE_MASK))) {
2114 			ntfs_error(vol->sb, "Tried to format non-existing mft "
2115 					"record 0x%llx.", (long long)mft_no);
2116 			return -ENOENT;
2117 		}
2118 	}
2119 	/* Read, map, and pin the page containing the mft record. */
2120 	page = ntfs_map_page(mft_vi->i_mapping, index);
2121 	if (IS_ERR(page)) {
2122 		ntfs_error(vol->sb, "Failed to map page containing mft record "
2123 				"to format 0x%llx.", (long long)mft_no);
2124 		return PTR_ERR(page);
2125 	}
2126 	lock_page(page);
2127 	BUG_ON(!PageUptodate(page));
2128 	ClearPageUptodate(page);
2129 	m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2130 	err = ntfs_mft_record_layout(vol, mft_no, m);
2131 	if (unlikely(err)) {
2132 		ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.",
2133 				(long long)mft_no);
2134 		SetPageUptodate(page);
2135 		unlock_page(page);
2136 		ntfs_unmap_page(page);
2137 		return err;
2138 	}
2139 	flush_dcache_page(page);
2140 	SetPageUptodate(page);
2141 	unlock_page(page);
2142 	/*
2143 	 * Make sure the mft record is written out to disk.  We could use
2144 	 * ilookup5() to check if an inode is in icache and so on but this is
2145 	 * unnecessary as ntfs_writepage() will write the dirty record anyway.
2146 	 */
2147 	mark_ntfs_record_dirty(page, ofs);
2148 	ntfs_unmap_page(page);
2149 	ntfs_debug("Done.");
2150 	return 0;
2151 }
2152 
2153 /**
2154  * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume
2155  * @vol:	[IN]  volume on which to allocate the mft record
2156  * @mode:	[IN]  mode if want a file or directory, i.e. base inode or 0
2157  * @base_ni:	[IN]  open base inode if allocating an extent mft record or NULL
2158  * @mrec:	[OUT] on successful return this is the mapped mft record
2159  *
2160  * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol.
2161  *
2162  * If @base_ni is NULL make the mft record a base mft record, i.e. a file or
2163  * direvctory inode, and allocate it at the default allocator position.  In
2164  * this case @mode is the file mode as given to us by the caller.  We in
2165  * particular use @mode to distinguish whether a file or a directory is being
2166  * created (S_IFDIR(mode) and S_IFREG(mode), respectively).
2167  *
2168  * If @base_ni is not NULL make the allocated mft record an extent record,
2169  * allocate it starting at the mft record after the base mft record and attach
2170  * the allocated and opened ntfs inode to the base inode @base_ni.  In this
2171  * case @mode must be 0 as it is meaningless for extent inodes.
2172  *
2173  * You need to check the return value with IS_ERR().  If false, the function
2174  * was successful and the return value is the now opened ntfs inode of the
2175  * allocated mft record.  *@mrec is then set to the allocated, mapped, pinned,
2176  * and locked mft record.  If IS_ERR() is true, the function failed and the
2177  * error code is obtained from PTR_ERR(return value).  *@mrec is undefined in
2178  * this case.
2179  *
2180  * Allocation strategy:
2181  *
2182  * To find a free mft record, we scan the mft bitmap for a zero bit.  To
2183  * optimize this we start scanning at the place specified by @base_ni or if
2184  * @base_ni is NULL we start where we last stopped and we perform wrap around
2185  * when we reach the end.  Note, we do not try to allocate mft records below
2186  * number 24 because numbers 0 to 15 are the defined system files anyway and 16
2187  * to 24 are special in that they are used for storing extension mft records
2188  * for the $DATA attribute of $MFT.  This is required to avoid the possibility
2189  * of creating a runlist with a circular dependency which once written to disk
2190  * can never be read in again.  Windows will only use records 16 to 24 for
2191  * normal files if the volume is completely out of space.  We never use them
2192  * which means that when the volume is really out of space we cannot create any
2193  * more files while Windows can still create up to 8 small files.  We can start
2194  * doing this at some later time, it does not matter much for now.
2195  *
2196  * When scanning the mft bitmap, we only search up to the last allocated mft
2197  * record.  If there are no free records left in the range 24 to number of
2198  * allocated mft records, then we extend the $MFT/$DATA attribute in order to
2199  * create free mft records.  We extend the allocated size of $MFT/$DATA by 16
2200  * records at a time or one cluster, if cluster size is above 16kiB.  If there
2201  * is not sufficient space to do this, we try to extend by a single mft record
2202  * or one cluster, if cluster size is above the mft record size.
2203  *
2204  * No matter how many mft records we allocate, we initialize only the first
2205  * allocated mft record, incrementing mft data size and initialized size
2206  * accordingly, open an ntfs_inode for it and return it to the caller, unless
2207  * there are less than 24 mft records, in which case we allocate and initialize
2208  * mft records until we reach record 24 which we consider as the first free mft
2209  * record for use by normal files.
2210  *
2211  * If during any stage we overflow the initialized data in the mft bitmap, we
2212  * extend the initialized size (and data size) by 8 bytes, allocating another
2213  * cluster if required.  The bitmap data size has to be at least equal to the
2214  * number of mft records in the mft, but it can be bigger, in which case the
2215  * superflous bits are padded with zeroes.
2216  *
2217  * Thus, when we return successfully (IS_ERR() is false), we will have:
2218  *	- initialized / extended the mft bitmap if necessary,
2219  *	- initialized / extended the mft data if necessary,
2220  *	- set the bit corresponding to the mft record being allocated in the
2221  *	  mft bitmap,
2222  *	- opened an ntfs_inode for the allocated mft record, and we will have
2223  *	- returned the ntfs_inode as well as the allocated mapped, pinned, and
2224  *	  locked mft record.
2225  *
2226  * On error, the volume will be left in a consistent state and no record will
2227  * be allocated.  If rolling back a partial operation fails, we may leave some
2228  * inconsistent metadata in which case we set NVolErrors() so the volume is
2229  * left dirty when unmounted.
2230  *
2231  * Note, this function cannot make use of most of the normal functions, like
2232  * for example for attribute resizing, etc, because when the run list overflows
2233  * the base mft record and an attribute list is used, it is very important that
2234  * the extension mft records used to store the $DATA attribute of $MFT can be
2235  * reached without having to read the information contained inside them, as
2236  * this would make it impossible to find them in the first place after the
2237  * volume is unmounted.  $MFT/$BITMAP probably does not need to follow this
2238  * rule because the bitmap is not essential for finding the mft records, but on
2239  * the other hand, handling the bitmap in this special way would make life
2240  * easier because otherwise there might be circular invocations of functions
2241  * when reading the bitmap.
2242  */
ntfs_mft_record_alloc(ntfs_volume * vol,const int mode,ntfs_inode * base_ni,MFT_RECORD ** mrec)2243 ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode,
2244 		ntfs_inode *base_ni, MFT_RECORD **mrec)
2245 {
2246 	s64 ll, bit, old_data_initialized, old_data_size;
2247 	unsigned long flags;
2248 	struct inode *vi;
2249 	struct page *page;
2250 	ntfs_inode *mft_ni, *mftbmp_ni, *ni;
2251 	ntfs_attr_search_ctx *ctx;
2252 	MFT_RECORD *m;
2253 	ATTR_RECORD *a;
2254 	pgoff_t index;
2255 	unsigned int ofs;
2256 	int err;
2257 	le16 seq_no, usn;
2258 	bool record_formatted = false;
2259 
2260 	if (base_ni) {
2261 		ntfs_debug("Entering (allocating an extent mft record for "
2262 				"base mft record 0x%llx).",
2263 				(long long)base_ni->mft_no);
2264 		/* @mode and @base_ni are mutually exclusive. */
2265 		BUG_ON(mode);
2266 	} else
2267 		ntfs_debug("Entering (allocating a base mft record).");
2268 	if (mode) {
2269 		/* @mode and @base_ni are mutually exclusive. */
2270 		BUG_ON(base_ni);
2271 		/* We only support creation of normal files and directories. */
2272 		if (!S_ISREG(mode) && !S_ISDIR(mode))
2273 			return ERR_PTR(-EOPNOTSUPP);
2274 	}
2275 	BUG_ON(!mrec);
2276 	mft_ni = NTFS_I(vol->mft_ino);
2277 	mftbmp_ni = NTFS_I(vol->mftbmp_ino);
2278 	down_write(&vol->mftbmp_lock);
2279 	bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni);
2280 	if (bit >= 0) {
2281 		ntfs_debug("Found and allocated free record (#1), bit 0x%llx.",
2282 				(long long)bit);
2283 		goto have_alloc_rec;
2284 	}
2285 	if (bit != -ENOSPC) {
2286 		up_write(&vol->mftbmp_lock);
2287 		return ERR_PTR(bit);
2288 	}
2289 	/*
2290 	 * No free mft records left.  If the mft bitmap already covers more
2291 	 * than the currently used mft records, the next records are all free,
2292 	 * so we can simply allocate the first unused mft record.
2293 	 * Note: We also have to make sure that the mft bitmap at least covers
2294 	 * the first 24 mft records as they are special and whilst they may not
2295 	 * be in use, we do not allocate from them.
2296 	 */
2297 	read_lock_irqsave(&mft_ni->size_lock, flags);
2298 	ll = mft_ni->initialized_size >> vol->mft_record_size_bits;
2299 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2300 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2301 	old_data_initialized = mftbmp_ni->initialized_size;
2302 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2303 	if (old_data_initialized << 3 > ll && old_data_initialized > 3) {
2304 		bit = ll;
2305 		if (bit < 24)
2306 			bit = 24;
2307 		if (unlikely(bit >= (1ll << 32)))
2308 			goto max_err_out;
2309 		ntfs_debug("Found free record (#2), bit 0x%llx.",
2310 				(long long)bit);
2311 		goto found_free_rec;
2312 	}
2313 	/*
2314 	 * The mft bitmap needs to be expanded until it covers the first unused
2315 	 * mft record that we can allocate.
2316 	 * Note: The smallest mft record we allocate is mft record 24.
2317 	 */
2318 	bit = old_data_initialized << 3;
2319 	if (unlikely(bit >= (1ll << 32)))
2320 		goto max_err_out;
2321 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2322 	old_data_size = mftbmp_ni->allocated_size;
2323 	ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, "
2324 			"data_size 0x%llx, initialized_size 0x%llx.",
2325 			(long long)old_data_size,
2326 			(long long)i_size_read(vol->mftbmp_ino),
2327 			(long long)old_data_initialized);
2328 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2329 	if (old_data_initialized + 8 > old_data_size) {
2330 		/* Need to extend bitmap by one more cluster. */
2331 		ntfs_debug("mftbmp: initialized_size + 8 > allocated_size.");
2332 		err = ntfs_mft_bitmap_extend_allocation_nolock(vol);
2333 		if (unlikely(err)) {
2334 			up_write(&vol->mftbmp_lock);
2335 			goto err_out;
2336 		}
2337 #ifdef DEBUG
2338 		read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2339 		ntfs_debug("Status of mftbmp after allocation extension: "
2340 				"allocated_size 0x%llx, data_size 0x%llx, "
2341 				"initialized_size 0x%llx.",
2342 				(long long)mftbmp_ni->allocated_size,
2343 				(long long)i_size_read(vol->mftbmp_ino),
2344 				(long long)mftbmp_ni->initialized_size);
2345 		read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2346 #endif /* DEBUG */
2347 	}
2348 	/*
2349 	 * We now have sufficient allocated space, extend the initialized_size
2350 	 * as well as the data_size if necessary and fill the new space with
2351 	 * zeroes.
2352 	 */
2353 	err = ntfs_mft_bitmap_extend_initialized_nolock(vol);
2354 	if (unlikely(err)) {
2355 		up_write(&vol->mftbmp_lock);
2356 		goto err_out;
2357 	}
2358 #ifdef DEBUG
2359 	read_lock_irqsave(&mftbmp_ni->size_lock, flags);
2360 	ntfs_debug("Status of mftbmp after initialized extension: "
2361 			"allocated_size 0x%llx, data_size 0x%llx, "
2362 			"initialized_size 0x%llx.",
2363 			(long long)mftbmp_ni->allocated_size,
2364 			(long long)i_size_read(vol->mftbmp_ino),
2365 			(long long)mftbmp_ni->initialized_size);
2366 	read_unlock_irqrestore(&mftbmp_ni->size_lock, flags);
2367 #endif /* DEBUG */
2368 	ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit);
2369 found_free_rec:
2370 	/* @bit is the found free mft record, allocate it in the mft bitmap. */
2371 	ntfs_debug("At found_free_rec.");
2372 	err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit);
2373 	if (unlikely(err)) {
2374 		ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap.");
2375 		up_write(&vol->mftbmp_lock);
2376 		goto err_out;
2377 	}
2378 	ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit);
2379 have_alloc_rec:
2380 	/*
2381 	 * The mft bitmap is now uptodate.  Deal with mft data attribute now.
2382 	 * Note, we keep hold of the mft bitmap lock for writing until all
2383 	 * modifications to the mft data attribute are complete, too, as they
2384 	 * will impact decisions for mft bitmap and mft record allocation done
2385 	 * by a parallel allocation and if the lock is not maintained a
2386 	 * parallel allocation could allocate the same mft record as this one.
2387 	 */
2388 	ll = (bit + 1) << vol->mft_record_size_bits;
2389 	read_lock_irqsave(&mft_ni->size_lock, flags);
2390 	old_data_initialized = mft_ni->initialized_size;
2391 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2392 	if (ll <= old_data_initialized) {
2393 		ntfs_debug("Allocated mft record already initialized.");
2394 		goto mft_rec_already_initialized;
2395 	}
2396 	ntfs_debug("Initializing allocated mft record.");
2397 	/*
2398 	 * The mft record is outside the initialized data.  Extend the mft data
2399 	 * attribute until it covers the allocated record.  The loop is only
2400 	 * actually traversed more than once when a freshly formatted volume is
2401 	 * first written to so it optimizes away nicely in the common case.
2402 	 */
2403 	read_lock_irqsave(&mft_ni->size_lock, flags);
2404 	ntfs_debug("Status of mft data before extension: "
2405 			"allocated_size 0x%llx, data_size 0x%llx, "
2406 			"initialized_size 0x%llx.",
2407 			(long long)mft_ni->allocated_size,
2408 			(long long)i_size_read(vol->mft_ino),
2409 			(long long)mft_ni->initialized_size);
2410 	while (ll > mft_ni->allocated_size) {
2411 		read_unlock_irqrestore(&mft_ni->size_lock, flags);
2412 		err = ntfs_mft_data_extend_allocation_nolock(vol);
2413 		if (unlikely(err)) {
2414 			ntfs_error(vol->sb, "Failed to extend mft data "
2415 					"allocation.");
2416 			goto undo_mftbmp_alloc_nolock;
2417 		}
2418 		read_lock_irqsave(&mft_ni->size_lock, flags);
2419 		ntfs_debug("Status of mft data after allocation extension: "
2420 				"allocated_size 0x%llx, data_size 0x%llx, "
2421 				"initialized_size 0x%llx.",
2422 				(long long)mft_ni->allocated_size,
2423 				(long long)i_size_read(vol->mft_ino),
2424 				(long long)mft_ni->initialized_size);
2425 	}
2426 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2427 	/*
2428 	 * Extend mft data initialized size (and data size of course) to reach
2429 	 * the allocated mft record, formatting the mft records allong the way.
2430 	 * Note: We only modify the ntfs_inode structure as that is all that is
2431 	 * needed by ntfs_mft_record_format().  We will update the attribute
2432 	 * record itself in one fell swoop later on.
2433 	 */
2434 	write_lock_irqsave(&mft_ni->size_lock, flags);
2435 	old_data_initialized = mft_ni->initialized_size;
2436 	old_data_size = vol->mft_ino->i_size;
2437 	while (ll > mft_ni->initialized_size) {
2438 		s64 new_initialized_size, mft_no;
2439 
2440 		new_initialized_size = mft_ni->initialized_size +
2441 				vol->mft_record_size;
2442 		mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits;
2443 		if (new_initialized_size > i_size_read(vol->mft_ino))
2444 			i_size_write(vol->mft_ino, new_initialized_size);
2445 		write_unlock_irqrestore(&mft_ni->size_lock, flags);
2446 		ntfs_debug("Initializing mft record 0x%llx.",
2447 				(long long)mft_no);
2448 		err = ntfs_mft_record_format(vol, mft_no);
2449 		if (unlikely(err)) {
2450 			ntfs_error(vol->sb, "Failed to format mft record.");
2451 			goto undo_data_init;
2452 		}
2453 		write_lock_irqsave(&mft_ni->size_lock, flags);
2454 		mft_ni->initialized_size = new_initialized_size;
2455 	}
2456 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
2457 	record_formatted = true;
2458 	/* Update the mft data attribute record to reflect the new sizes. */
2459 	m = map_mft_record(mft_ni);
2460 	if (IS_ERR(m)) {
2461 		ntfs_error(vol->sb, "Failed to map mft record.");
2462 		err = PTR_ERR(m);
2463 		goto undo_data_init;
2464 	}
2465 	ctx = ntfs_attr_get_search_ctx(mft_ni, m);
2466 	if (unlikely(!ctx)) {
2467 		ntfs_error(vol->sb, "Failed to get search context.");
2468 		err = -ENOMEM;
2469 		unmap_mft_record(mft_ni);
2470 		goto undo_data_init;
2471 	}
2472 	err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len,
2473 			CASE_SENSITIVE, 0, NULL, 0, ctx);
2474 	if (unlikely(err)) {
2475 		ntfs_error(vol->sb, "Failed to find first attribute extent of "
2476 				"mft data attribute.");
2477 		ntfs_attr_put_search_ctx(ctx);
2478 		unmap_mft_record(mft_ni);
2479 		goto undo_data_init;
2480 	}
2481 	a = ctx->attr;
2482 	read_lock_irqsave(&mft_ni->size_lock, flags);
2483 	a->data.non_resident.initialized_size =
2484 			cpu_to_sle64(mft_ni->initialized_size);
2485 	a->data.non_resident.data_size =
2486 			cpu_to_sle64(i_size_read(vol->mft_ino));
2487 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2488 	/* Ensure the changes make it to disk. */
2489 	flush_dcache_mft_record_page(ctx->ntfs_ino);
2490 	mark_mft_record_dirty(ctx->ntfs_ino);
2491 	ntfs_attr_put_search_ctx(ctx);
2492 	unmap_mft_record(mft_ni);
2493 	read_lock_irqsave(&mft_ni->size_lock, flags);
2494 	ntfs_debug("Status of mft data after mft record initialization: "
2495 			"allocated_size 0x%llx, data_size 0x%llx, "
2496 			"initialized_size 0x%llx.",
2497 			(long long)mft_ni->allocated_size,
2498 			(long long)i_size_read(vol->mft_ino),
2499 			(long long)mft_ni->initialized_size);
2500 	BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size);
2501 	BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino));
2502 	read_unlock_irqrestore(&mft_ni->size_lock, flags);
2503 mft_rec_already_initialized:
2504 	/*
2505 	 * We can finally drop the mft bitmap lock as the mft data attribute
2506 	 * has been fully updated.  The only disparity left is that the
2507 	 * allocated mft record still needs to be marked as in use to match the
2508 	 * set bit in the mft bitmap but this is actually not a problem since
2509 	 * this mft record is not referenced from anywhere yet and the fact
2510 	 * that it is allocated in the mft bitmap means that no-one will try to
2511 	 * allocate it either.
2512 	 */
2513 	up_write(&vol->mftbmp_lock);
2514 	/*
2515 	 * We now have allocated and initialized the mft record.  Calculate the
2516 	 * index of and the offset within the page cache page the record is in.
2517 	 */
2518 	index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT;
2519 	ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK;
2520 	/* Read, map, and pin the page containing the mft record. */
2521 	page = ntfs_map_page(vol->mft_ino->i_mapping, index);
2522 	if (IS_ERR(page)) {
2523 		ntfs_error(vol->sb, "Failed to map page containing allocated "
2524 				"mft record 0x%llx.", (long long)bit);
2525 		err = PTR_ERR(page);
2526 		goto undo_mftbmp_alloc;
2527 	}
2528 	lock_page(page);
2529 	BUG_ON(!PageUptodate(page));
2530 	ClearPageUptodate(page);
2531 	m = (MFT_RECORD*)((u8*)page_address(page) + ofs);
2532 	/* If we just formatted the mft record no need to do it again. */
2533 	if (!record_formatted) {
2534 		/* Sanity check that the mft record is really not in use. */
2535 		if (ntfs_is_file_record(m->magic) &&
2536 				(m->flags & MFT_RECORD_IN_USE)) {
2537 			ntfs_error(vol->sb, "Mft record 0x%llx was marked "
2538 					"free in mft bitmap but is marked "
2539 					"used itself.  Corrupt filesystem.  "
2540 					"Unmount and run chkdsk.",
2541 					(long long)bit);
2542 			err = -EIO;
2543 			SetPageUptodate(page);
2544 			unlock_page(page);
2545 			ntfs_unmap_page(page);
2546 			NVolSetErrors(vol);
2547 			goto undo_mftbmp_alloc;
2548 		}
2549 		/*
2550 		 * We need to (re-)format the mft record, preserving the
2551 		 * sequence number if it is not zero as well as the update
2552 		 * sequence number if it is not zero or -1 (0xffff).  This
2553 		 * means we do not need to care whether or not something went
2554 		 * wrong with the previous mft record.
2555 		 */
2556 		seq_no = m->sequence_number;
2557 		usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs));
2558 		err = ntfs_mft_record_layout(vol, bit, m);
2559 		if (unlikely(err)) {
2560 			ntfs_error(vol->sb, "Failed to layout allocated mft "
2561 					"record 0x%llx.", (long long)bit);
2562 			SetPageUptodate(page);
2563 			unlock_page(page);
2564 			ntfs_unmap_page(page);
2565 			goto undo_mftbmp_alloc;
2566 		}
2567 		if (seq_no)
2568 			m->sequence_number = seq_no;
2569 		if (usn && le16_to_cpu(usn) != 0xffff)
2570 			*(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn;
2571 	}
2572 	/* Set the mft record itself in use. */
2573 	m->flags |= MFT_RECORD_IN_USE;
2574 	if (S_ISDIR(mode))
2575 		m->flags |= MFT_RECORD_IS_DIRECTORY;
2576 	flush_dcache_page(page);
2577 	SetPageUptodate(page);
2578 	if (base_ni) {
2579 		MFT_RECORD *m_tmp;
2580 
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_tmp = map_extent_mft_record(base_ni, bit, &ni);
2594 		if (IS_ERR(m_tmp)) {
2595 			ntfs_error(vol->sb, "Failed to map allocated extent "
2596 					"mft record 0x%llx.", (long long)bit);
2597 			err = PTR_ERR(m_tmp);
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 		BUG_ON(m != m_tmp);
2609 		/*
2610 		 * Make sure the allocated mft record is written out to disk.
2611 		 * No need to set the inode dirty because the caller is going
2612 		 * to do that anyway after finishing with the new extent mft
2613 		 * record (e.g. at a minimum a new attribute will be added to
2614 		 * the mft record.
2615 		 */
2616 		mark_ntfs_record_dirty(page, ofs);
2617 		unlock_page(page);
2618 		/*
2619 		 * Need to unmap the page since map_extent_mft_record() mapped
2620 		 * it as well so we have it mapped twice at the moment.
2621 		 */
2622 		ntfs_unmap_page(page);
2623 	} else {
2624 		/*
2625 		 * Allocate a new VFS inode and set it up.  NOTE: @vi->i_nlink
2626 		 * is set to 1 but the mft record->link_count is 0.  The caller
2627 		 * needs to bear this in mind.
2628 		 */
2629 		vi = new_inode(vol->sb);
2630 		if (unlikely(!vi)) {
2631 			err = -ENOMEM;
2632 			/* Set the mft record itself not in use. */
2633 			m->flags &= cpu_to_le16(
2634 					~le16_to_cpu(MFT_RECORD_IN_USE));
2635 			flush_dcache_page(page);
2636 			/* Make sure the mft record is written out to disk. */
2637 			mark_ntfs_record_dirty(page, ofs);
2638 			unlock_page(page);
2639 			ntfs_unmap_page(page);
2640 			goto undo_mftbmp_alloc;
2641 		}
2642 		vi->i_ino = bit;
2643 		/*
2644 		 * This is for checking whether an inode has changed w.r.t. a
2645 		 * file so that the file can be updated if necessary (compare
2646 		 * with f_version).
2647 		 */
2648 		vi->i_version = 1;
2649 
2650 		/* The owner and group come from the ntfs volume. */
2651 		vi->i_uid = vol->uid;
2652 		vi->i_gid = vol->gid;
2653 
2654 		/* Initialize the ntfs specific part of @vi. */
2655 		ntfs_init_big_inode(vi);
2656 		ni = NTFS_I(vi);
2657 		/*
2658 		 * Set the appropriate mode, attribute type, and name.  For
2659 		 * directories, also setup the index values to the defaults.
2660 		 */
2661 		if (S_ISDIR(mode)) {
2662 			vi->i_mode = S_IFDIR | S_IRWXUGO;
2663 			vi->i_mode &= ~vol->dmask;
2664 
2665 			NInoSetMstProtected(ni);
2666 			ni->type = AT_INDEX_ALLOCATION;
2667 			ni->name = I30;
2668 			ni->name_len = 4;
2669 
2670 			ni->itype.index.block_size = 4096;
2671 			ni->itype.index.block_size_bits = ntfs_ffs(4096) - 1;
2672 			ni->itype.index.collation_rule = COLLATION_FILE_NAME;
2673 			if (vol->cluster_size <= ni->itype.index.block_size) {
2674 				ni->itype.index.vcn_size = vol->cluster_size;
2675 				ni->itype.index.vcn_size_bits =
2676 						vol->cluster_size_bits;
2677 			} else {
2678 				ni->itype.index.vcn_size = vol->sector_size;
2679 				ni->itype.index.vcn_size_bits =
2680 						vol->sector_size_bits;
2681 			}
2682 		} else {
2683 			vi->i_mode = S_IFREG | S_IRWXUGO;
2684 			vi->i_mode &= ~vol->fmask;
2685 
2686 			ni->type = AT_DATA;
2687 			ni->name = NULL;
2688 			ni->name_len = 0;
2689 		}
2690 		if (IS_RDONLY(vi))
2691 			vi->i_mode &= ~S_IWUGO;
2692 
2693 		/* Set the inode times to the current time. */
2694 		vi->i_atime = vi->i_mtime = vi->i_ctime =
2695 			current_fs_time(vi->i_sb);
2696 		/*
2697 		 * Set the file size to 0, the ntfs inode sizes are set to 0 by
2698 		 * the call to ntfs_init_big_inode() below.
2699 		 */
2700 		vi->i_size = 0;
2701 		vi->i_blocks = 0;
2702 
2703 		/* Set the sequence number. */
2704 		vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
2705 		/*
2706 		 * Manually map, pin, and lock the mft record as we already
2707 		 * have its page mapped and it is very easy to do.
2708 		 */
2709 		atomic_inc(&ni->count);
2710 		mutex_lock(&ni->mrec_lock);
2711 		ni->page = page;
2712 		ni->page_ofs = ofs;
2713 		/*
2714 		 * Make sure the allocated mft record is written out to disk.
2715 		 * NOTE: We do not set the ntfs inode dirty because this would
2716 		 * fail in ntfs_write_inode() because the inode does not have a
2717 		 * standard information attribute yet.  Also, there is no need
2718 		 * to set the inode dirty because the caller is going to do
2719 		 * that anyway after finishing with the new mft record (e.g. at
2720 		 * a minimum some new attributes will be added to the mft
2721 		 * record.
2722 		 */
2723 		mark_ntfs_record_dirty(page, ofs);
2724 		unlock_page(page);
2725 
2726 		/* Add the inode to the inode hash for the superblock. */
2727 		insert_inode_hash(vi);
2728 
2729 		/* Update the default mft allocation position. */
2730 		vol->mft_data_pos = bit + 1;
2731 	}
2732 	/*
2733 	 * Return the opened, allocated inode of the allocated mft record as
2734 	 * well as the mapped, pinned, and locked mft record.
2735 	 */
2736 	ntfs_debug("Returning opened, allocated %sinode 0x%llx.",
2737 			base_ni ? "extent " : "", (long long)bit);
2738 	*mrec = m;
2739 	return ni;
2740 undo_data_init:
2741 	write_lock_irqsave(&mft_ni->size_lock, flags);
2742 	mft_ni->initialized_size = old_data_initialized;
2743 	i_size_write(vol->mft_ino, old_data_size);
2744 	write_unlock_irqrestore(&mft_ni->size_lock, flags);
2745 	goto undo_mftbmp_alloc_nolock;
2746 undo_mftbmp_alloc:
2747 	down_write(&vol->mftbmp_lock);
2748 undo_mftbmp_alloc_nolock:
2749 	if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) {
2750 		ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2751 		NVolSetErrors(vol);
2752 	}
2753 	up_write(&vol->mftbmp_lock);
2754 err_out:
2755 	return ERR_PTR(err);
2756 max_err_out:
2757 	ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum "
2758 			"number of inodes (2^32) has already been reached.");
2759 	up_write(&vol->mftbmp_lock);
2760 	return ERR_PTR(-ENOSPC);
2761 }
2762 
2763 /**
2764  * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume
2765  * @ni:		ntfs inode of the mapped extent mft record to free
2766  * @m:		mapped extent mft record of the ntfs inode @ni
2767  *
2768  * Free the mapped extent mft record @m of the extent ntfs inode @ni.
2769  *
2770  * Note that this function unmaps the mft record and closes and destroys @ni
2771  * internally and hence you cannot use either @ni nor @m any more after this
2772  * function returns success.
2773  *
2774  * On success return 0 and on error return -errno.  @ni and @m are still valid
2775  * in this case and have not been freed.
2776  *
2777  * For some errors an error message is displayed and the success code 0 is
2778  * returned and the volume is then left dirty on umount.  This makes sense in
2779  * case we could not rollback the changes that were already done since the
2780  * caller no longer wants to reference this mft record so it does not matter to
2781  * the caller if something is wrong with it as long as it is properly detached
2782  * from the base inode.
2783  */
ntfs_extent_mft_record_free(ntfs_inode * ni,MFT_RECORD * m)2784 int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m)
2785 {
2786 	unsigned long mft_no = ni->mft_no;
2787 	ntfs_volume *vol = ni->vol;
2788 	ntfs_inode *base_ni;
2789 	ntfs_inode **extent_nis;
2790 	int i, err;
2791 	le16 old_seq_no;
2792 	u16 seq_no;
2793 
2794 	BUG_ON(NInoAttr(ni));
2795 	BUG_ON(ni->nr_extents != -1);
2796 
2797 	mutex_lock(&ni->extent_lock);
2798 	base_ni = ni->ext.base_ntfs_ino;
2799 	mutex_unlock(&ni->extent_lock);
2800 
2801 	BUG_ON(base_ni->nr_extents <= 0);
2802 
2803 	ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n",
2804 			mft_no, base_ni->mft_no);
2805 
2806 	mutex_lock(&base_ni->extent_lock);
2807 
2808 	/* Make sure we are holding the only reference to the extent inode. */
2809 	if (atomic_read(&ni->count) > 2) {
2810 		ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, "
2811 				"not freeing.", base_ni->mft_no);
2812 		mutex_unlock(&base_ni->extent_lock);
2813 		return -EBUSY;
2814 	}
2815 
2816 	/* Dissociate the ntfs inode from the base inode. */
2817 	extent_nis = base_ni->ext.extent_ntfs_inos;
2818 	err = -ENOENT;
2819 	for (i = 0; i < base_ni->nr_extents; i++) {
2820 		if (ni != extent_nis[i])
2821 			continue;
2822 		extent_nis += i;
2823 		base_ni->nr_extents--;
2824 		memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) *
2825 				sizeof(ntfs_inode*));
2826 		err = 0;
2827 		break;
2828 	}
2829 
2830 	mutex_unlock(&base_ni->extent_lock);
2831 
2832 	if (unlikely(err)) {
2833 		ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to "
2834 				"its base inode 0x%lx.", mft_no,
2835 				base_ni->mft_no);
2836 		BUG();
2837 	}
2838 
2839 	/*
2840 	 * The extent inode is no longer attached to the base inode so no one
2841 	 * can get a reference to it any more.
2842 	 */
2843 
2844 	/* Mark the mft record as not in use. */
2845 	m->flags &= ~MFT_RECORD_IN_USE;
2846 
2847 	/* Increment the sequence number, skipping zero, if it is not zero. */
2848 	old_seq_no = m->sequence_number;
2849 	seq_no = le16_to_cpu(old_seq_no);
2850 	if (seq_no == 0xffff)
2851 		seq_no = 1;
2852 	else if (seq_no)
2853 		seq_no++;
2854 	m->sequence_number = cpu_to_le16(seq_no);
2855 
2856 	/*
2857 	 * Set the ntfs inode dirty and write it out.  We do not need to worry
2858 	 * about the base inode here since whatever caused the extent mft
2859 	 * record to be freed is guaranteed to do it already.
2860 	 */
2861 	NInoSetDirty(ni);
2862 	err = write_mft_record(ni, m, 0);
2863 	if (unlikely(err)) {
2864 		ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not "
2865 				"freeing.", mft_no);
2866 		goto rollback;
2867 	}
2868 rollback_error:
2869 	/* Unmap and throw away the now freed extent inode. */
2870 	unmap_extent_mft_record(ni);
2871 	ntfs_clear_extent_inode(ni);
2872 
2873 	/* Clear the bit in the $MFT/$BITMAP corresponding to this record. */
2874 	down_write(&vol->mftbmp_lock);
2875 	err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no);
2876 	up_write(&vol->mftbmp_lock);
2877 	if (unlikely(err)) {
2878 		/*
2879 		 * The extent inode is gone but we failed to deallocate it in
2880 		 * the mft bitmap.  Just emit a warning and leave the volume
2881 		 * dirty on umount.
2882 		 */
2883 		ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es);
2884 		NVolSetErrors(vol);
2885 	}
2886 	return 0;
2887 rollback:
2888 	/* Rollback what we did... */
2889 	mutex_lock(&base_ni->extent_lock);
2890 	extent_nis = base_ni->ext.extent_ntfs_inos;
2891 	if (!(base_ni->nr_extents & 3)) {
2892 		int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*);
2893 
2894 		extent_nis = kmalloc(new_size, GFP_NOFS);
2895 		if (unlikely(!extent_nis)) {
2896 			ntfs_error(vol->sb, "Failed to allocate internal "
2897 					"buffer during rollback.%s", es);
2898 			mutex_unlock(&base_ni->extent_lock);
2899 			NVolSetErrors(vol);
2900 			goto rollback_error;
2901 		}
2902 		if (base_ni->nr_extents) {
2903 			BUG_ON(!base_ni->ext.extent_ntfs_inos);
2904 			memcpy(extent_nis, base_ni->ext.extent_ntfs_inos,
2905 					new_size - 4 * sizeof(ntfs_inode*));
2906 			kfree(base_ni->ext.extent_ntfs_inos);
2907 		}
2908 		base_ni->ext.extent_ntfs_inos = extent_nis;
2909 	}
2910 	m->flags |= MFT_RECORD_IN_USE;
2911 	m->sequence_number = old_seq_no;
2912 	extent_nis[base_ni->nr_extents++] = ni;
2913 	mutex_unlock(&base_ni->extent_lock);
2914 	mark_mft_record_dirty(ni);
2915 	return err;
2916 }
2917 #endif /* NTFS_RW */
2918