1 /**
2  * aops.c - NTFS kernel address space operations and page cache handling.
3  *
4  * Copyright (c) 2001-2014 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/errno.h>
24 #include <linux/fs.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/pagemap.h>
28 #include <linux/swap.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/bit_spinlock.h>
32 #include <linux/bio.h>
33 
34 #include "aops.h"
35 #include "attrib.h"
36 #include "debug.h"
37 #include "inode.h"
38 #include "mft.h"
39 #include "runlist.h"
40 #include "types.h"
41 #include "ntfs.h"
42 
43 /**
44  * ntfs_end_buffer_async_read - async io completion for reading attributes
45  * @bh:		buffer head on which io is completed
46  * @uptodate:	whether @bh is now uptodate or not
47  *
48  * Asynchronous I/O completion handler for reading pages belonging to the
49  * attribute address space of an inode.  The inodes can either be files or
50  * directories or they can be fake inodes describing some attribute.
51  *
52  * If NInoMstProtected(), perform the post read mst fixups when all IO on the
53  * page has been completed and mark the page uptodate or set the error bit on
54  * the page.  To determine the size of the records that need fixing up, we
55  * cheat a little bit by setting the index_block_size in ntfs_inode to the ntfs
56  * record size, and index_block_size_bits, to the log(base 2) of the ntfs
57  * record size.
58  */
ntfs_end_buffer_async_read(struct buffer_head * bh,int uptodate)59 static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
60 {
61 	unsigned long flags;
62 	struct buffer_head *first, *tmp;
63 	struct page *page;
64 	struct inode *vi;
65 	ntfs_inode *ni;
66 	int page_uptodate = 1;
67 
68 	page = bh->b_page;
69 	vi = page->mapping->host;
70 	ni = NTFS_I(vi);
71 
72 	if (likely(uptodate)) {
73 		loff_t i_size;
74 		s64 file_ofs, init_size;
75 
76 		set_buffer_uptodate(bh);
77 
78 		file_ofs = ((s64)page->index << PAGE_SHIFT) +
79 				bh_offset(bh);
80 		read_lock_irqsave(&ni->size_lock, flags);
81 		init_size = ni->initialized_size;
82 		i_size = i_size_read(vi);
83 		read_unlock_irqrestore(&ni->size_lock, flags);
84 		if (unlikely(init_size > i_size)) {
85 			/* Race with shrinking truncate. */
86 			init_size = i_size;
87 		}
88 		/* Check for the current buffer head overflowing. */
89 		if (unlikely(file_ofs + bh->b_size > init_size)) {
90 			int ofs;
91 			void *kaddr;
92 
93 			ofs = 0;
94 			if (file_ofs < init_size)
95 				ofs = init_size - file_ofs;
96 			kaddr = kmap_atomic(page);
97 			memset(kaddr + bh_offset(bh) + ofs, 0,
98 					bh->b_size - ofs);
99 			flush_dcache_page(page);
100 			kunmap_atomic(kaddr);
101 		}
102 	} else {
103 		clear_buffer_uptodate(bh);
104 		SetPageError(page);
105 		ntfs_error(ni->vol->sb, "Buffer I/O error, logical block "
106 				"0x%llx.", (unsigned long long)bh->b_blocknr);
107 	}
108 	first = page_buffers(page);
109 	local_irq_save(flags);
110 	bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
111 	clear_buffer_async_read(bh);
112 	unlock_buffer(bh);
113 	tmp = bh;
114 	do {
115 		if (!buffer_uptodate(tmp))
116 			page_uptodate = 0;
117 		if (buffer_async_read(tmp)) {
118 			if (likely(buffer_locked(tmp)))
119 				goto still_busy;
120 			/* Async buffers must be locked. */
121 			BUG();
122 		}
123 		tmp = tmp->b_this_page;
124 	} while (tmp != bh);
125 	bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
126 	local_irq_restore(flags);
127 	/*
128 	 * If none of the buffers had errors then we can set the page uptodate,
129 	 * but we first have to perform the post read mst fixups, if the
130 	 * attribute is mst protected, i.e. if NInoMstProteced(ni) is true.
131 	 * Note we ignore fixup errors as those are detected when
132 	 * map_mft_record() is called which gives us per record granularity
133 	 * rather than per page granularity.
134 	 */
135 	if (!NInoMstProtected(ni)) {
136 		if (likely(page_uptodate && !PageError(page)))
137 			SetPageUptodate(page);
138 	} else {
139 		u8 *kaddr;
140 		unsigned int i, recs;
141 		u32 rec_size;
142 
143 		rec_size = ni->itype.index.block_size;
144 		recs = PAGE_SIZE / rec_size;
145 		/* Should have been verified before we got here... */
146 		BUG_ON(!recs);
147 		kaddr = kmap_atomic(page);
148 		for (i = 0; i < recs; i++)
149 			post_read_mst_fixup((NTFS_RECORD*)(kaddr +
150 					i * rec_size), rec_size);
151 		kunmap_atomic(kaddr);
152 		flush_dcache_page(page);
153 		if (likely(page_uptodate && !PageError(page)))
154 			SetPageUptodate(page);
155 	}
156 	unlock_page(page);
157 	return;
158 still_busy:
159 	bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
160 	local_irq_restore(flags);
161 	return;
162 }
163 
164 /**
165  * ntfs_read_block - fill a @page of an address space with data
166  * @page:	page cache page to fill with data
167  *
168  * Fill the page @page of the address space belonging to the @page->host inode.
169  * We read each buffer asynchronously and when all buffers are read in, our io
170  * completion handler ntfs_end_buffer_read_async(), if required, automatically
171  * applies the mst fixups to the page before finally marking it uptodate and
172  * unlocking it.
173  *
174  * We only enforce allocated_size limit because i_size is checked for in
175  * generic_file_read().
176  *
177  * Return 0 on success and -errno on error.
178  *
179  * Contains an adapted version of fs/buffer.c::block_read_full_page().
180  */
ntfs_read_block(struct page * page)181 static int ntfs_read_block(struct page *page)
182 {
183 	loff_t i_size;
184 	VCN vcn;
185 	LCN lcn;
186 	s64 init_size;
187 	struct inode *vi;
188 	ntfs_inode *ni;
189 	ntfs_volume *vol;
190 	runlist_element *rl;
191 	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
192 	sector_t iblock, lblock, zblock;
193 	unsigned long flags;
194 	unsigned int blocksize, vcn_ofs;
195 	int i, nr;
196 	unsigned char blocksize_bits;
197 
198 	vi = page->mapping->host;
199 	ni = NTFS_I(vi);
200 	vol = ni->vol;
201 
202 	/* $MFT/$DATA must have its complete runlist in memory at all times. */
203 	BUG_ON(!ni->runlist.rl && !ni->mft_no && !NInoAttr(ni));
204 
205 	blocksize = vol->sb->s_blocksize;
206 	blocksize_bits = vol->sb->s_blocksize_bits;
207 
208 	if (!page_has_buffers(page)) {
209 		create_empty_buffers(page, blocksize, 0);
210 		if (unlikely(!page_has_buffers(page))) {
211 			unlock_page(page);
212 			return -ENOMEM;
213 		}
214 	}
215 	bh = head = page_buffers(page);
216 	BUG_ON(!bh);
217 
218 	/*
219 	 * We may be racing with truncate.  To avoid some of the problems we
220 	 * now take a snapshot of the various sizes and use those for the whole
221 	 * of the function.  In case of an extending truncate it just means we
222 	 * may leave some buffers unmapped which are now allocated.  This is
223 	 * not a problem since these buffers will just get mapped when a write
224 	 * occurs.  In case of a shrinking truncate, we will detect this later
225 	 * on due to the runlist being incomplete and if the page is being
226 	 * fully truncated, truncate will throw it away as soon as we unlock
227 	 * it so no need to worry what we do with it.
228 	 */
229 	iblock = (s64)page->index << (PAGE_SHIFT - blocksize_bits);
230 	read_lock_irqsave(&ni->size_lock, flags);
231 	lblock = (ni->allocated_size + blocksize - 1) >> blocksize_bits;
232 	init_size = ni->initialized_size;
233 	i_size = i_size_read(vi);
234 	read_unlock_irqrestore(&ni->size_lock, flags);
235 	if (unlikely(init_size > i_size)) {
236 		/* Race with shrinking truncate. */
237 		init_size = i_size;
238 	}
239 	zblock = (init_size + blocksize - 1) >> blocksize_bits;
240 
241 	/* Loop through all the buffers in the page. */
242 	rl = NULL;
243 	nr = i = 0;
244 	do {
245 		int err = 0;
246 
247 		if (unlikely(buffer_uptodate(bh)))
248 			continue;
249 		if (unlikely(buffer_mapped(bh))) {
250 			arr[nr++] = bh;
251 			continue;
252 		}
253 		bh->b_bdev = vol->sb->s_bdev;
254 		/* Is the block within the allowed limits? */
255 		if (iblock < lblock) {
256 			bool is_retry = false;
257 
258 			/* Convert iblock into corresponding vcn and offset. */
259 			vcn = (VCN)iblock << blocksize_bits >>
260 					vol->cluster_size_bits;
261 			vcn_ofs = ((VCN)iblock << blocksize_bits) &
262 					vol->cluster_size_mask;
263 			if (!rl) {
264 lock_retry_remap:
265 				down_read(&ni->runlist.lock);
266 				rl = ni->runlist.rl;
267 			}
268 			if (likely(rl != NULL)) {
269 				/* Seek to element containing target vcn. */
270 				while (rl->length && rl[1].vcn <= vcn)
271 					rl++;
272 				lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
273 			} else
274 				lcn = LCN_RL_NOT_MAPPED;
275 			/* Successful remap. */
276 			if (lcn >= 0) {
277 				/* Setup buffer head to correct block. */
278 				bh->b_blocknr = ((lcn << vol->cluster_size_bits)
279 						+ vcn_ofs) >> blocksize_bits;
280 				set_buffer_mapped(bh);
281 				/* Only read initialized data blocks. */
282 				if (iblock < zblock) {
283 					arr[nr++] = bh;
284 					continue;
285 				}
286 				/* Fully non-initialized data block, zero it. */
287 				goto handle_zblock;
288 			}
289 			/* It is a hole, need to zero it. */
290 			if (lcn == LCN_HOLE)
291 				goto handle_hole;
292 			/* If first try and runlist unmapped, map and retry. */
293 			if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
294 				is_retry = true;
295 				/*
296 				 * Attempt to map runlist, dropping lock for
297 				 * the duration.
298 				 */
299 				up_read(&ni->runlist.lock);
300 				err = ntfs_map_runlist(ni, vcn);
301 				if (likely(!err))
302 					goto lock_retry_remap;
303 				rl = NULL;
304 			} else if (!rl)
305 				up_read(&ni->runlist.lock);
306 			/*
307 			 * If buffer is outside the runlist, treat it as a
308 			 * hole.  This can happen due to concurrent truncate
309 			 * for example.
310 			 */
311 			if (err == -ENOENT || lcn == LCN_ENOENT) {
312 				err = 0;
313 				goto handle_hole;
314 			}
315 			/* Hard error, zero out region. */
316 			if (!err)
317 				err = -EIO;
318 			bh->b_blocknr = -1;
319 			SetPageError(page);
320 			ntfs_error(vol->sb, "Failed to read from inode 0x%lx, "
321 					"attribute type 0x%x, vcn 0x%llx, "
322 					"offset 0x%x because its location on "
323 					"disk could not be determined%s "
324 					"(error code %i).", ni->mft_no,
325 					ni->type, (unsigned long long)vcn,
326 					vcn_ofs, is_retry ? " even after "
327 					"retrying" : "", err);
328 		}
329 		/*
330 		 * Either iblock was outside lblock limits or
331 		 * ntfs_rl_vcn_to_lcn() returned error.  Just zero that portion
332 		 * of the page and set the buffer uptodate.
333 		 */
334 handle_hole:
335 		bh->b_blocknr = -1UL;
336 		clear_buffer_mapped(bh);
337 handle_zblock:
338 		zero_user(page, i * blocksize, blocksize);
339 		if (likely(!err))
340 			set_buffer_uptodate(bh);
341 	} while (i++, iblock++, (bh = bh->b_this_page) != head);
342 
343 	/* Release the lock if we took it. */
344 	if (rl)
345 		up_read(&ni->runlist.lock);
346 
347 	/* Check we have at least one buffer ready for i/o. */
348 	if (nr) {
349 		struct buffer_head *tbh;
350 
351 		/* Lock the buffers. */
352 		for (i = 0; i < nr; i++) {
353 			tbh = arr[i];
354 			lock_buffer(tbh);
355 			tbh->b_end_io = ntfs_end_buffer_async_read;
356 			set_buffer_async_read(tbh);
357 		}
358 		/* Finally, start i/o on the buffers. */
359 		for (i = 0; i < nr; i++) {
360 			tbh = arr[i];
361 			if (likely(!buffer_uptodate(tbh)))
362 				submit_bh(REQ_OP_READ, 0, tbh);
363 			else
364 				ntfs_end_buffer_async_read(tbh, 1);
365 		}
366 		return 0;
367 	}
368 	/* No i/o was scheduled on any of the buffers. */
369 	if (likely(!PageError(page)))
370 		SetPageUptodate(page);
371 	else /* Signal synchronous i/o error. */
372 		nr = -EIO;
373 	unlock_page(page);
374 	return nr;
375 }
376 
377 /**
378  * ntfs_readpage - fill a @page of a @file with data from the device
379  * @file:	open file to which the page @page belongs or NULL
380  * @page:	page cache page to fill with data
381  *
382  * For non-resident attributes, ntfs_readpage() fills the @page of the open
383  * file @file by calling the ntfs version of the generic block_read_full_page()
384  * function, ntfs_read_block(), which in turn creates and reads in the buffers
385  * associated with the page asynchronously.
386  *
387  * For resident attributes, OTOH, ntfs_readpage() fills @page by copying the
388  * data from the mft record (which at this stage is most likely in memory) and
389  * fills the remainder with zeroes. Thus, in this case, I/O is synchronous, as
390  * even if the mft record is not cached at this point in time, we need to wait
391  * for it to be read in before we can do the copy.
392  *
393  * Return 0 on success and -errno on error.
394  */
ntfs_readpage(struct file * file,struct page * page)395 static int ntfs_readpage(struct file *file, struct page *page)
396 {
397 	loff_t i_size;
398 	struct inode *vi;
399 	ntfs_inode *ni, *base_ni;
400 	u8 *addr;
401 	ntfs_attr_search_ctx *ctx;
402 	MFT_RECORD *mrec;
403 	unsigned long flags;
404 	u32 attr_len;
405 	int err = 0;
406 
407 retry_readpage:
408 	BUG_ON(!PageLocked(page));
409 	vi = page->mapping->host;
410 	i_size = i_size_read(vi);
411 	/* Is the page fully outside i_size? (truncate in progress) */
412 	if (unlikely(page->index >= (i_size + PAGE_SIZE - 1) >>
413 			PAGE_SHIFT)) {
414 		zero_user(page, 0, PAGE_SIZE);
415 		ntfs_debug("Read outside i_size - truncated?");
416 		goto done;
417 	}
418 	/*
419 	 * This can potentially happen because we clear PageUptodate() during
420 	 * ntfs_writepage() of MstProtected() attributes.
421 	 */
422 	if (PageUptodate(page)) {
423 		unlock_page(page);
424 		return 0;
425 	}
426 	ni = NTFS_I(vi);
427 	/*
428 	 * Only $DATA attributes can be encrypted and only unnamed $DATA
429 	 * attributes can be compressed.  Index root can have the flags set but
430 	 * this means to create compressed/encrypted files, not that the
431 	 * attribute is compressed/encrypted.  Note we need to check for
432 	 * AT_INDEX_ALLOCATION since this is the type of both directory and
433 	 * index inodes.
434 	 */
435 	if (ni->type != AT_INDEX_ALLOCATION) {
436 		/* If attribute is encrypted, deny access, just like NT4. */
437 		if (NInoEncrypted(ni)) {
438 			BUG_ON(ni->type != AT_DATA);
439 			err = -EACCES;
440 			goto err_out;
441 		}
442 		/* Compressed data streams are handled in compress.c. */
443 		if (NInoNonResident(ni) && NInoCompressed(ni)) {
444 			BUG_ON(ni->type != AT_DATA);
445 			BUG_ON(ni->name_len);
446 			return ntfs_read_compressed_block(page);
447 		}
448 	}
449 	/* NInoNonResident() == NInoIndexAllocPresent() */
450 	if (NInoNonResident(ni)) {
451 		/* Normal, non-resident data stream. */
452 		return ntfs_read_block(page);
453 	}
454 	/*
455 	 * Attribute is resident, implying it is not compressed or encrypted.
456 	 * This also means the attribute is smaller than an mft record and
457 	 * hence smaller than a page, so can simply zero out any pages with
458 	 * index above 0.  Note the attribute can actually be marked compressed
459 	 * but if it is resident the actual data is not compressed so we are
460 	 * ok to ignore the compressed flag here.
461 	 */
462 	if (unlikely(page->index > 0)) {
463 		zero_user(page, 0, PAGE_SIZE);
464 		goto done;
465 	}
466 	if (!NInoAttr(ni))
467 		base_ni = ni;
468 	else
469 		base_ni = ni->ext.base_ntfs_ino;
470 	/* Map, pin, and lock the mft record. */
471 	mrec = map_mft_record(base_ni);
472 	if (IS_ERR(mrec)) {
473 		err = PTR_ERR(mrec);
474 		goto err_out;
475 	}
476 	/*
477 	 * If a parallel write made the attribute non-resident, drop the mft
478 	 * record and retry the readpage.
479 	 */
480 	if (unlikely(NInoNonResident(ni))) {
481 		unmap_mft_record(base_ni);
482 		goto retry_readpage;
483 	}
484 	ctx = ntfs_attr_get_search_ctx(base_ni, mrec);
485 	if (unlikely(!ctx)) {
486 		err = -ENOMEM;
487 		goto unm_err_out;
488 	}
489 	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
490 			CASE_SENSITIVE, 0, NULL, 0, ctx);
491 	if (unlikely(err))
492 		goto put_unm_err_out;
493 	attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
494 	read_lock_irqsave(&ni->size_lock, flags);
495 	if (unlikely(attr_len > ni->initialized_size))
496 		attr_len = ni->initialized_size;
497 	i_size = i_size_read(vi);
498 	read_unlock_irqrestore(&ni->size_lock, flags);
499 	if (unlikely(attr_len > i_size)) {
500 		/* Race with shrinking truncate. */
501 		attr_len = i_size;
502 	}
503 	addr = kmap_atomic(page);
504 	/* Copy the data to the page. */
505 	memcpy(addr, (u8*)ctx->attr +
506 			le16_to_cpu(ctx->attr->data.resident.value_offset),
507 			attr_len);
508 	/* Zero the remainder of the page. */
509 	memset(addr + attr_len, 0, PAGE_SIZE - attr_len);
510 	flush_dcache_page(page);
511 	kunmap_atomic(addr);
512 put_unm_err_out:
513 	ntfs_attr_put_search_ctx(ctx);
514 unm_err_out:
515 	unmap_mft_record(base_ni);
516 done:
517 	SetPageUptodate(page);
518 err_out:
519 	unlock_page(page);
520 	return err;
521 }
522 
523 #ifdef NTFS_RW
524 
525 /**
526  * ntfs_write_block - write a @page to the backing store
527  * @page:	page cache page to write out
528  * @wbc:	writeback control structure
529  *
530  * This function is for writing pages belonging to non-resident, non-mst
531  * protected attributes to their backing store.
532  *
533  * For a page with buffers, map and write the dirty buffers asynchronously
534  * under page writeback. For a page without buffers, create buffers for the
535  * page, then proceed as above.
536  *
537  * If a page doesn't have buffers the page dirty state is definitive. If a page
538  * does have buffers, the page dirty state is just a hint, and the buffer dirty
539  * state is definitive. (A hint which has rules: dirty buffers against a clean
540  * page is illegal. Other combinations are legal and need to be handled. In
541  * particular a dirty page containing clean buffers for example.)
542  *
543  * Return 0 on success and -errno on error.
544  *
545  * Based on ntfs_read_block() and __block_write_full_page().
546  */
ntfs_write_block(struct page * page,struct writeback_control * wbc)547 static int ntfs_write_block(struct page *page, struct writeback_control *wbc)
548 {
549 	VCN vcn;
550 	LCN lcn;
551 	s64 initialized_size;
552 	loff_t i_size;
553 	sector_t block, dblock, iblock;
554 	struct inode *vi;
555 	ntfs_inode *ni;
556 	ntfs_volume *vol;
557 	runlist_element *rl;
558 	struct buffer_head *bh, *head;
559 	unsigned long flags;
560 	unsigned int blocksize, vcn_ofs;
561 	int err;
562 	bool need_end_writeback;
563 	unsigned char blocksize_bits;
564 
565 	vi = page->mapping->host;
566 	ni = NTFS_I(vi);
567 	vol = ni->vol;
568 
569 	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
570 			"0x%lx.", ni->mft_no, ni->type, page->index);
571 
572 	BUG_ON(!NInoNonResident(ni));
573 	BUG_ON(NInoMstProtected(ni));
574 	blocksize = vol->sb->s_blocksize;
575 	blocksize_bits = vol->sb->s_blocksize_bits;
576 	if (!page_has_buffers(page)) {
577 		BUG_ON(!PageUptodate(page));
578 		create_empty_buffers(page, blocksize,
579 				(1 << BH_Uptodate) | (1 << BH_Dirty));
580 		if (unlikely(!page_has_buffers(page))) {
581 			ntfs_warning(vol->sb, "Error allocating page "
582 					"buffers.  Redirtying page so we try "
583 					"again later.");
584 			/*
585 			 * Put the page back on mapping->dirty_pages, but leave
586 			 * its buffers' dirty state as-is.
587 			 */
588 			redirty_page_for_writepage(wbc, page);
589 			unlock_page(page);
590 			return 0;
591 		}
592 	}
593 	bh = head = page_buffers(page);
594 	BUG_ON(!bh);
595 
596 	/* NOTE: Different naming scheme to ntfs_read_block()! */
597 
598 	/* The first block in the page. */
599 	block = (s64)page->index << (PAGE_SHIFT - blocksize_bits);
600 
601 	read_lock_irqsave(&ni->size_lock, flags);
602 	i_size = i_size_read(vi);
603 	initialized_size = ni->initialized_size;
604 	read_unlock_irqrestore(&ni->size_lock, flags);
605 
606 	/* The first out of bounds block for the data size. */
607 	dblock = (i_size + blocksize - 1) >> blocksize_bits;
608 
609 	/* The last (fully or partially) initialized block. */
610 	iblock = initialized_size >> blocksize_bits;
611 
612 	/*
613 	 * Be very careful.  We have no exclusion from __set_page_dirty_buffers
614 	 * here, and the (potentially unmapped) buffers may become dirty at
615 	 * any time.  If a buffer becomes dirty here after we've inspected it
616 	 * then we just miss that fact, and the page stays dirty.
617 	 *
618 	 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
619 	 * handle that here by just cleaning them.
620 	 */
621 
622 	/*
623 	 * Loop through all the buffers in the page, mapping all the dirty
624 	 * buffers to disk addresses and handling any aliases from the
625 	 * underlying block device's mapping.
626 	 */
627 	rl = NULL;
628 	err = 0;
629 	do {
630 		bool is_retry = false;
631 
632 		if (unlikely(block >= dblock)) {
633 			/*
634 			 * Mapped buffers outside i_size will occur, because
635 			 * this page can be outside i_size when there is a
636 			 * truncate in progress. The contents of such buffers
637 			 * were zeroed by ntfs_writepage().
638 			 *
639 			 * FIXME: What about the small race window where
640 			 * ntfs_writepage() has not done any clearing because
641 			 * the page was within i_size but before we get here,
642 			 * vmtruncate() modifies i_size?
643 			 */
644 			clear_buffer_dirty(bh);
645 			set_buffer_uptodate(bh);
646 			continue;
647 		}
648 
649 		/* Clean buffers are not written out, so no need to map them. */
650 		if (!buffer_dirty(bh))
651 			continue;
652 
653 		/* Make sure we have enough initialized size. */
654 		if (unlikely((block >= iblock) &&
655 				(initialized_size < i_size))) {
656 			/*
657 			 * If this page is fully outside initialized size, zero
658 			 * out all pages between the current initialized size
659 			 * and the current page. Just use ntfs_readpage() to do
660 			 * the zeroing transparently.
661 			 */
662 			if (block > iblock) {
663 				// TODO:
664 				// For each page do:
665 				// - read_cache_page()
666 				// Again for each page do:
667 				// - wait_on_page_locked()
668 				// - Check (PageUptodate(page) &&
669 				//			!PageError(page))
670 				// Update initialized size in the attribute and
671 				// in the inode.
672 				// Again, for each page do:
673 				//	__set_page_dirty_buffers();
674 				// put_page()
675 				// We don't need to wait on the writes.
676 				// Update iblock.
677 			}
678 			/*
679 			 * The current page straddles initialized size. Zero
680 			 * all non-uptodate buffers and set them uptodate (and
681 			 * dirty?). Note, there aren't any non-uptodate buffers
682 			 * if the page is uptodate.
683 			 * FIXME: For an uptodate page, the buffers may need to
684 			 * be written out because they were not initialized on
685 			 * disk before.
686 			 */
687 			if (!PageUptodate(page)) {
688 				// TODO:
689 				// Zero any non-uptodate buffers up to i_size.
690 				// Set them uptodate and dirty.
691 			}
692 			// TODO:
693 			// Update initialized size in the attribute and in the
694 			// inode (up to i_size).
695 			// Update iblock.
696 			// FIXME: This is inefficient. Try to batch the two
697 			// size changes to happen in one go.
698 			ntfs_error(vol->sb, "Writing beyond initialized size "
699 					"is not supported yet. Sorry.");
700 			err = -EOPNOTSUPP;
701 			break;
702 			// Do NOT set_buffer_new() BUT DO clear buffer range
703 			// outside write request range.
704 			// set_buffer_uptodate() on complete buffers as well as
705 			// set_buffer_dirty().
706 		}
707 
708 		/* No need to map buffers that are already mapped. */
709 		if (buffer_mapped(bh))
710 			continue;
711 
712 		/* Unmapped, dirty buffer. Need to map it. */
713 		bh->b_bdev = vol->sb->s_bdev;
714 
715 		/* Convert block into corresponding vcn and offset. */
716 		vcn = (VCN)block << blocksize_bits;
717 		vcn_ofs = vcn & vol->cluster_size_mask;
718 		vcn >>= vol->cluster_size_bits;
719 		if (!rl) {
720 lock_retry_remap:
721 			down_read(&ni->runlist.lock);
722 			rl = ni->runlist.rl;
723 		}
724 		if (likely(rl != NULL)) {
725 			/* Seek to element containing target vcn. */
726 			while (rl->length && rl[1].vcn <= vcn)
727 				rl++;
728 			lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
729 		} else
730 			lcn = LCN_RL_NOT_MAPPED;
731 		/* Successful remap. */
732 		if (lcn >= 0) {
733 			/* Setup buffer head to point to correct block. */
734 			bh->b_blocknr = ((lcn << vol->cluster_size_bits) +
735 					vcn_ofs) >> blocksize_bits;
736 			set_buffer_mapped(bh);
737 			continue;
738 		}
739 		/* It is a hole, need to instantiate it. */
740 		if (lcn == LCN_HOLE) {
741 			u8 *kaddr;
742 			unsigned long *bpos, *bend;
743 
744 			/* Check if the buffer is zero. */
745 			kaddr = kmap_atomic(page);
746 			bpos = (unsigned long *)(kaddr + bh_offset(bh));
747 			bend = (unsigned long *)((u8*)bpos + blocksize);
748 			do {
749 				if (unlikely(*bpos))
750 					break;
751 			} while (likely(++bpos < bend));
752 			kunmap_atomic(kaddr);
753 			if (bpos == bend) {
754 				/*
755 				 * Buffer is zero and sparse, no need to write
756 				 * it.
757 				 */
758 				bh->b_blocknr = -1;
759 				clear_buffer_dirty(bh);
760 				continue;
761 			}
762 			// TODO: Instantiate the hole.
763 			// clear_buffer_new(bh);
764 			// clean_bdev_bh_alias(bh);
765 			ntfs_error(vol->sb, "Writing into sparse regions is "
766 					"not supported yet. Sorry.");
767 			err = -EOPNOTSUPP;
768 			break;
769 		}
770 		/* If first try and runlist unmapped, map and retry. */
771 		if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
772 			is_retry = true;
773 			/*
774 			 * Attempt to map runlist, dropping lock for
775 			 * the duration.
776 			 */
777 			up_read(&ni->runlist.lock);
778 			err = ntfs_map_runlist(ni, vcn);
779 			if (likely(!err))
780 				goto lock_retry_remap;
781 			rl = NULL;
782 		} else if (!rl)
783 			up_read(&ni->runlist.lock);
784 		/*
785 		 * If buffer is outside the runlist, truncate has cut it out
786 		 * of the runlist.  Just clean and clear the buffer and set it
787 		 * uptodate so it can get discarded by the VM.
788 		 */
789 		if (err == -ENOENT || lcn == LCN_ENOENT) {
790 			bh->b_blocknr = -1;
791 			clear_buffer_dirty(bh);
792 			zero_user(page, bh_offset(bh), blocksize);
793 			set_buffer_uptodate(bh);
794 			err = 0;
795 			continue;
796 		}
797 		/* Failed to map the buffer, even after retrying. */
798 		if (!err)
799 			err = -EIO;
800 		bh->b_blocknr = -1;
801 		ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
802 				"attribute type 0x%x, vcn 0x%llx, offset 0x%x "
803 				"because its location on disk could not be "
804 				"determined%s (error code %i).", ni->mft_no,
805 				ni->type, (unsigned long long)vcn,
806 				vcn_ofs, is_retry ? " even after "
807 				"retrying" : "", err);
808 		break;
809 	} while (block++, (bh = bh->b_this_page) != head);
810 
811 	/* Release the lock if we took it. */
812 	if (rl)
813 		up_read(&ni->runlist.lock);
814 
815 	/* For the error case, need to reset bh to the beginning. */
816 	bh = head;
817 
818 	/* Just an optimization, so ->readpage() is not called later. */
819 	if (unlikely(!PageUptodate(page))) {
820 		int uptodate = 1;
821 		do {
822 			if (!buffer_uptodate(bh)) {
823 				uptodate = 0;
824 				bh = head;
825 				break;
826 			}
827 		} while ((bh = bh->b_this_page) != head);
828 		if (uptodate)
829 			SetPageUptodate(page);
830 	}
831 
832 	/* Setup all mapped, dirty buffers for async write i/o. */
833 	do {
834 		if (buffer_mapped(bh) && buffer_dirty(bh)) {
835 			lock_buffer(bh);
836 			if (test_clear_buffer_dirty(bh)) {
837 				BUG_ON(!buffer_uptodate(bh));
838 				mark_buffer_async_write(bh);
839 			} else
840 				unlock_buffer(bh);
841 		} else if (unlikely(err)) {
842 			/*
843 			 * For the error case. The buffer may have been set
844 			 * dirty during attachment to a dirty page.
845 			 */
846 			if (err != -ENOMEM)
847 				clear_buffer_dirty(bh);
848 		}
849 	} while ((bh = bh->b_this_page) != head);
850 
851 	if (unlikely(err)) {
852 		// TODO: Remove the -EOPNOTSUPP check later on...
853 		if (unlikely(err == -EOPNOTSUPP))
854 			err = 0;
855 		else if (err == -ENOMEM) {
856 			ntfs_warning(vol->sb, "Error allocating memory. "
857 					"Redirtying page so we try again "
858 					"later.");
859 			/*
860 			 * Put the page back on mapping->dirty_pages, but
861 			 * leave its buffer's dirty state as-is.
862 			 */
863 			redirty_page_for_writepage(wbc, page);
864 			err = 0;
865 		} else
866 			SetPageError(page);
867 	}
868 
869 	BUG_ON(PageWriteback(page));
870 	set_page_writeback(page);	/* Keeps try_to_free_buffers() away. */
871 
872 	/* Submit the prepared buffers for i/o. */
873 	need_end_writeback = true;
874 	do {
875 		struct buffer_head *next = bh->b_this_page;
876 		if (buffer_async_write(bh)) {
877 			submit_bh(REQ_OP_WRITE, 0, bh);
878 			need_end_writeback = false;
879 		}
880 		bh = next;
881 	} while (bh != head);
882 	unlock_page(page);
883 
884 	/* If no i/o was started, need to end_page_writeback(). */
885 	if (unlikely(need_end_writeback))
886 		end_page_writeback(page);
887 
888 	ntfs_debug("Done.");
889 	return err;
890 }
891 
892 /**
893  * ntfs_write_mst_block - write a @page to the backing store
894  * @page:	page cache page to write out
895  * @wbc:	writeback control structure
896  *
897  * This function is for writing pages belonging to non-resident, mst protected
898  * attributes to their backing store.  The only supported attributes are index
899  * allocation and $MFT/$DATA.  Both directory inodes and index inodes are
900  * supported for the index allocation case.
901  *
902  * The page must remain locked for the duration of the write because we apply
903  * the mst fixups, write, and then undo the fixups, so if we were to unlock the
904  * page before undoing the fixups, any other user of the page will see the
905  * page contents as corrupt.
906  *
907  * We clear the page uptodate flag for the duration of the function to ensure
908  * exclusion for the $MFT/$DATA case against someone mapping an mft record we
909  * are about to apply the mst fixups to.
910  *
911  * Return 0 on success and -errno on error.
912  *
913  * Based on ntfs_write_block(), ntfs_mft_writepage(), and
914  * write_mft_record_nolock().
915  */
ntfs_write_mst_block(struct page * page,struct writeback_control * wbc)916 static int ntfs_write_mst_block(struct page *page,
917 		struct writeback_control *wbc)
918 {
919 	sector_t block, dblock, rec_block;
920 	struct inode *vi = page->mapping->host;
921 	ntfs_inode *ni = NTFS_I(vi);
922 	ntfs_volume *vol = ni->vol;
923 	u8 *kaddr;
924 	unsigned int rec_size = ni->itype.index.block_size;
925 	ntfs_inode *locked_nis[PAGE_SIZE / NTFS_BLOCK_SIZE];
926 	struct buffer_head *bh, *head, *tbh, *rec_start_bh;
927 	struct buffer_head *bhs[MAX_BUF_PER_PAGE];
928 	runlist_element *rl;
929 	int i, nr_locked_nis, nr_recs, nr_bhs, max_bhs, bhs_per_rec, err, err2;
930 	unsigned bh_size, rec_size_bits;
931 	bool sync, is_mft, page_is_dirty, rec_is_dirty;
932 	unsigned char bh_size_bits;
933 
934 	if (WARN_ON(rec_size < NTFS_BLOCK_SIZE))
935 		return -EINVAL;
936 
937 	ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
938 			"0x%lx.", vi->i_ino, ni->type, page->index);
939 	BUG_ON(!NInoNonResident(ni));
940 	BUG_ON(!NInoMstProtected(ni));
941 	is_mft = (S_ISREG(vi->i_mode) && !vi->i_ino);
942 	/*
943 	 * NOTE: ntfs_write_mst_block() would be called for $MFTMirr if a page
944 	 * in its page cache were to be marked dirty.  However this should
945 	 * never happen with the current driver and considering we do not
946 	 * handle this case here we do want to BUG(), at least for now.
947 	 */
948 	BUG_ON(!(is_mft || S_ISDIR(vi->i_mode) ||
949 			(NInoAttr(ni) && ni->type == AT_INDEX_ALLOCATION)));
950 	bh_size = vol->sb->s_blocksize;
951 	bh_size_bits = vol->sb->s_blocksize_bits;
952 	max_bhs = PAGE_SIZE / bh_size;
953 	BUG_ON(!max_bhs);
954 	BUG_ON(max_bhs > MAX_BUF_PER_PAGE);
955 
956 	/* Were we called for sync purposes? */
957 	sync = (wbc->sync_mode == WB_SYNC_ALL);
958 
959 	/* Make sure we have mapped buffers. */
960 	bh = head = page_buffers(page);
961 	BUG_ON(!bh);
962 
963 	rec_size_bits = ni->itype.index.block_size_bits;
964 	BUG_ON(!(PAGE_SIZE >> rec_size_bits));
965 	bhs_per_rec = rec_size >> bh_size_bits;
966 	BUG_ON(!bhs_per_rec);
967 
968 	/* The first block in the page. */
969 	rec_block = block = (sector_t)page->index <<
970 			(PAGE_SHIFT - bh_size_bits);
971 
972 	/* The first out of bounds block for the data size. */
973 	dblock = (i_size_read(vi) + bh_size - 1) >> bh_size_bits;
974 
975 	rl = NULL;
976 	err = err2 = nr_bhs = nr_recs = nr_locked_nis = 0;
977 	page_is_dirty = rec_is_dirty = false;
978 	rec_start_bh = NULL;
979 	do {
980 		bool is_retry = false;
981 
982 		if (likely(block < rec_block)) {
983 			if (unlikely(block >= dblock)) {
984 				clear_buffer_dirty(bh);
985 				set_buffer_uptodate(bh);
986 				continue;
987 			}
988 			/*
989 			 * This block is not the first one in the record.  We
990 			 * ignore the buffer's dirty state because we could
991 			 * have raced with a parallel mark_ntfs_record_dirty().
992 			 */
993 			if (!rec_is_dirty)
994 				continue;
995 			if (unlikely(err2)) {
996 				if (err2 != -ENOMEM)
997 					clear_buffer_dirty(bh);
998 				continue;
999 			}
1000 		} else /* if (block == rec_block) */ {
1001 			BUG_ON(block > rec_block);
1002 			/* This block is the first one in the record. */
1003 			rec_block += bhs_per_rec;
1004 			err2 = 0;
1005 			if (unlikely(block >= dblock)) {
1006 				clear_buffer_dirty(bh);
1007 				continue;
1008 			}
1009 			if (!buffer_dirty(bh)) {
1010 				/* Clean records are not written out. */
1011 				rec_is_dirty = false;
1012 				continue;
1013 			}
1014 			rec_is_dirty = true;
1015 			rec_start_bh = bh;
1016 		}
1017 		/* Need to map the buffer if it is not mapped already. */
1018 		if (unlikely(!buffer_mapped(bh))) {
1019 			VCN vcn;
1020 			LCN lcn;
1021 			unsigned int vcn_ofs;
1022 
1023 			bh->b_bdev = vol->sb->s_bdev;
1024 			/* Obtain the vcn and offset of the current block. */
1025 			vcn = (VCN)block << bh_size_bits;
1026 			vcn_ofs = vcn & vol->cluster_size_mask;
1027 			vcn >>= vol->cluster_size_bits;
1028 			if (!rl) {
1029 lock_retry_remap:
1030 				down_read(&ni->runlist.lock);
1031 				rl = ni->runlist.rl;
1032 			}
1033 			if (likely(rl != NULL)) {
1034 				/* Seek to element containing target vcn. */
1035 				while (rl->length && rl[1].vcn <= vcn)
1036 					rl++;
1037 				lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
1038 			} else
1039 				lcn = LCN_RL_NOT_MAPPED;
1040 			/* Successful remap. */
1041 			if (likely(lcn >= 0)) {
1042 				/* Setup buffer head to correct block. */
1043 				bh->b_blocknr = ((lcn <<
1044 						vol->cluster_size_bits) +
1045 						vcn_ofs) >> bh_size_bits;
1046 				set_buffer_mapped(bh);
1047 			} else {
1048 				/*
1049 				 * Remap failed.  Retry to map the runlist once
1050 				 * unless we are working on $MFT which always
1051 				 * has the whole of its runlist in memory.
1052 				 */
1053 				if (!is_mft && !is_retry &&
1054 						lcn == LCN_RL_NOT_MAPPED) {
1055 					is_retry = true;
1056 					/*
1057 					 * Attempt to map runlist, dropping
1058 					 * lock for the duration.
1059 					 */
1060 					up_read(&ni->runlist.lock);
1061 					err2 = ntfs_map_runlist(ni, vcn);
1062 					if (likely(!err2))
1063 						goto lock_retry_remap;
1064 					if (err2 == -ENOMEM)
1065 						page_is_dirty = true;
1066 					lcn = err2;
1067 				} else {
1068 					err2 = -EIO;
1069 					if (!rl)
1070 						up_read(&ni->runlist.lock);
1071 				}
1072 				/* Hard error.  Abort writing this record. */
1073 				if (!err || err == -ENOMEM)
1074 					err = err2;
1075 				bh->b_blocknr = -1;
1076 				ntfs_error(vol->sb, "Cannot write ntfs record "
1077 						"0x%llx (inode 0x%lx, "
1078 						"attribute type 0x%x) because "
1079 						"its location on disk could "
1080 						"not be determined (error "
1081 						"code %lli).",
1082 						(long long)block <<
1083 						bh_size_bits >>
1084 						vol->mft_record_size_bits,
1085 						ni->mft_no, ni->type,
1086 						(long long)lcn);
1087 				/*
1088 				 * If this is not the first buffer, remove the
1089 				 * buffers in this record from the list of
1090 				 * buffers to write and clear their dirty bit
1091 				 * if not error -ENOMEM.
1092 				 */
1093 				if (rec_start_bh != bh) {
1094 					while (bhs[--nr_bhs] != rec_start_bh)
1095 						;
1096 					if (err2 != -ENOMEM) {
1097 						do {
1098 							clear_buffer_dirty(
1099 								rec_start_bh);
1100 						} while ((rec_start_bh =
1101 								rec_start_bh->
1102 								b_this_page) !=
1103 								bh);
1104 					}
1105 				}
1106 				continue;
1107 			}
1108 		}
1109 		BUG_ON(!buffer_uptodate(bh));
1110 		BUG_ON(nr_bhs >= max_bhs);
1111 		bhs[nr_bhs++] = bh;
1112 	} while (block++, (bh = bh->b_this_page) != head);
1113 	if (unlikely(rl))
1114 		up_read(&ni->runlist.lock);
1115 	/* If there were no dirty buffers, we are done. */
1116 	if (!nr_bhs)
1117 		goto done;
1118 	/* Map the page so we can access its contents. */
1119 	kaddr = kmap(page);
1120 	/* Clear the page uptodate flag whilst the mst fixups are applied. */
1121 	BUG_ON(!PageUptodate(page));
1122 	ClearPageUptodate(page);
1123 	for (i = 0; i < nr_bhs; i++) {
1124 		unsigned int ofs;
1125 
1126 		/* Skip buffers which are not at the beginning of records. */
1127 		if (i % bhs_per_rec)
1128 			continue;
1129 		tbh = bhs[i];
1130 		ofs = bh_offset(tbh);
1131 		if (is_mft) {
1132 			ntfs_inode *tni;
1133 			unsigned long mft_no;
1134 
1135 			/* Get the mft record number. */
1136 			mft_no = (((s64)page->index << PAGE_SHIFT) + ofs)
1137 					>> rec_size_bits;
1138 			/* Check whether to write this mft record. */
1139 			tni = NULL;
1140 			if (!ntfs_may_write_mft_record(vol, mft_no,
1141 					(MFT_RECORD*)(kaddr + ofs), &tni)) {
1142 				/*
1143 				 * The record should not be written.  This
1144 				 * means we need to redirty the page before
1145 				 * returning.
1146 				 */
1147 				page_is_dirty = true;
1148 				/*
1149 				 * Remove the buffers in this mft record from
1150 				 * the list of buffers to write.
1151 				 */
1152 				do {
1153 					bhs[i] = NULL;
1154 				} while (++i % bhs_per_rec);
1155 				continue;
1156 			}
1157 			/*
1158 			 * The record should be written.  If a locked ntfs
1159 			 * inode was returned, add it to the array of locked
1160 			 * ntfs inodes.
1161 			 */
1162 			if (tni)
1163 				locked_nis[nr_locked_nis++] = tni;
1164 		}
1165 		/* Apply the mst protection fixups. */
1166 		err2 = pre_write_mst_fixup((NTFS_RECORD*)(kaddr + ofs),
1167 				rec_size);
1168 		if (unlikely(err2)) {
1169 			if (!err || err == -ENOMEM)
1170 				err = -EIO;
1171 			ntfs_error(vol->sb, "Failed to apply mst fixups "
1172 					"(inode 0x%lx, attribute type 0x%x, "
1173 					"page index 0x%lx, page offset 0x%x)!"
1174 					"  Unmount and run chkdsk.", vi->i_ino,
1175 					ni->type, page->index, ofs);
1176 			/*
1177 			 * Mark all the buffers in this record clean as we do
1178 			 * not want to write corrupt data to disk.
1179 			 */
1180 			do {
1181 				clear_buffer_dirty(bhs[i]);
1182 				bhs[i] = NULL;
1183 			} while (++i % bhs_per_rec);
1184 			continue;
1185 		}
1186 		nr_recs++;
1187 	}
1188 	/* If no records are to be written out, we are done. */
1189 	if (!nr_recs)
1190 		goto unm_done;
1191 	flush_dcache_page(page);
1192 	/* Lock buffers and start synchronous write i/o on them. */
1193 	for (i = 0; i < nr_bhs; i++) {
1194 		tbh = bhs[i];
1195 		if (!tbh)
1196 			continue;
1197 		if (!trylock_buffer(tbh))
1198 			BUG();
1199 		/* The buffer dirty state is now irrelevant, just clean it. */
1200 		clear_buffer_dirty(tbh);
1201 		BUG_ON(!buffer_uptodate(tbh));
1202 		BUG_ON(!buffer_mapped(tbh));
1203 		get_bh(tbh);
1204 		tbh->b_end_io = end_buffer_write_sync;
1205 		submit_bh(REQ_OP_WRITE, 0, tbh);
1206 	}
1207 	/* Synchronize the mft mirror now if not @sync. */
1208 	if (is_mft && !sync)
1209 		goto do_mirror;
1210 do_wait:
1211 	/* Wait on i/o completion of buffers. */
1212 	for (i = 0; i < nr_bhs; i++) {
1213 		tbh = bhs[i];
1214 		if (!tbh)
1215 			continue;
1216 		wait_on_buffer(tbh);
1217 		if (unlikely(!buffer_uptodate(tbh))) {
1218 			ntfs_error(vol->sb, "I/O error while writing ntfs "
1219 					"record buffer (inode 0x%lx, "
1220 					"attribute type 0x%x, page index "
1221 					"0x%lx, page offset 0x%lx)!  Unmount "
1222 					"and run chkdsk.", vi->i_ino, ni->type,
1223 					page->index, bh_offset(tbh));
1224 			if (!err || err == -ENOMEM)
1225 				err = -EIO;
1226 			/*
1227 			 * Set the buffer uptodate so the page and buffer
1228 			 * states do not become out of sync.
1229 			 */
1230 			set_buffer_uptodate(tbh);
1231 		}
1232 	}
1233 	/* If @sync, now synchronize the mft mirror. */
1234 	if (is_mft && sync) {
1235 do_mirror:
1236 		for (i = 0; i < nr_bhs; i++) {
1237 			unsigned long mft_no;
1238 			unsigned int ofs;
1239 
1240 			/*
1241 			 * Skip buffers which are not at the beginning of
1242 			 * records.
1243 			 */
1244 			if (i % bhs_per_rec)
1245 				continue;
1246 			tbh = bhs[i];
1247 			/* Skip removed buffers (and hence records). */
1248 			if (!tbh)
1249 				continue;
1250 			ofs = bh_offset(tbh);
1251 			/* Get the mft record number. */
1252 			mft_no = (((s64)page->index << PAGE_SHIFT) + ofs)
1253 					>> rec_size_bits;
1254 			if (mft_no < vol->mftmirr_size)
1255 				ntfs_sync_mft_mirror(vol, mft_no,
1256 						(MFT_RECORD*)(kaddr + ofs),
1257 						sync);
1258 		}
1259 		if (!sync)
1260 			goto do_wait;
1261 	}
1262 	/* Remove the mst protection fixups again. */
1263 	for (i = 0; i < nr_bhs; i++) {
1264 		if (!(i % bhs_per_rec)) {
1265 			tbh = bhs[i];
1266 			if (!tbh)
1267 				continue;
1268 			post_write_mst_fixup((NTFS_RECORD*)(kaddr +
1269 					bh_offset(tbh)));
1270 		}
1271 	}
1272 	flush_dcache_page(page);
1273 unm_done:
1274 	/* Unlock any locked inodes. */
1275 	while (nr_locked_nis-- > 0) {
1276 		ntfs_inode *tni, *base_tni;
1277 
1278 		tni = locked_nis[nr_locked_nis];
1279 		/* Get the base inode. */
1280 		mutex_lock(&tni->extent_lock);
1281 		if (tni->nr_extents >= 0)
1282 			base_tni = tni;
1283 		else {
1284 			base_tni = tni->ext.base_ntfs_ino;
1285 			BUG_ON(!base_tni);
1286 		}
1287 		mutex_unlock(&tni->extent_lock);
1288 		ntfs_debug("Unlocking %s inode 0x%lx.",
1289 				tni == base_tni ? "base" : "extent",
1290 				tni->mft_no);
1291 		mutex_unlock(&tni->mrec_lock);
1292 		atomic_dec(&tni->count);
1293 		iput(VFS_I(base_tni));
1294 	}
1295 	SetPageUptodate(page);
1296 	kunmap(page);
1297 done:
1298 	if (unlikely(err && err != -ENOMEM)) {
1299 		/*
1300 		 * Set page error if there is only one ntfs record in the page.
1301 		 * Otherwise we would loose per-record granularity.
1302 		 */
1303 		if (ni->itype.index.block_size == PAGE_SIZE)
1304 			SetPageError(page);
1305 		NVolSetErrors(vol);
1306 	}
1307 	if (page_is_dirty) {
1308 		ntfs_debug("Page still contains one or more dirty ntfs "
1309 				"records.  Redirtying the page starting at "
1310 				"record 0x%lx.", page->index <<
1311 				(PAGE_SHIFT - rec_size_bits));
1312 		redirty_page_for_writepage(wbc, page);
1313 		unlock_page(page);
1314 	} else {
1315 		/*
1316 		 * Keep the VM happy.  This must be done otherwise the
1317 		 * radix-tree tag PAGECACHE_TAG_DIRTY remains set even though
1318 		 * the page is clean.
1319 		 */
1320 		BUG_ON(PageWriteback(page));
1321 		set_page_writeback(page);
1322 		unlock_page(page);
1323 		end_page_writeback(page);
1324 	}
1325 	if (likely(!err))
1326 		ntfs_debug("Done.");
1327 	return err;
1328 }
1329 
1330 /**
1331  * ntfs_writepage - write a @page to the backing store
1332  * @page:	page cache page to write out
1333  * @wbc:	writeback control structure
1334  *
1335  * This is called from the VM when it wants to have a dirty ntfs page cache
1336  * page cleaned.  The VM has already locked the page and marked it clean.
1337  *
1338  * For non-resident attributes, ntfs_writepage() writes the @page by calling
1339  * the ntfs version of the generic block_write_full_page() function,
1340  * ntfs_write_block(), which in turn if necessary creates and writes the
1341  * buffers associated with the page asynchronously.
1342  *
1343  * For resident attributes, OTOH, ntfs_writepage() writes the @page by copying
1344  * the data to the mft record (which at this stage is most likely in memory).
1345  * The mft record is then marked dirty and written out asynchronously via the
1346  * vfs inode dirty code path for the inode the mft record belongs to or via the
1347  * vm page dirty code path for the page the mft record is in.
1348  *
1349  * Based on ntfs_readpage() and fs/buffer.c::block_write_full_page().
1350  *
1351  * Return 0 on success and -errno on error.
1352  */
ntfs_writepage(struct page * page,struct writeback_control * wbc)1353 static int ntfs_writepage(struct page *page, struct writeback_control *wbc)
1354 {
1355 	loff_t i_size;
1356 	struct inode *vi = page->mapping->host;
1357 	ntfs_inode *base_ni = NULL, *ni = NTFS_I(vi);
1358 	char *addr;
1359 	ntfs_attr_search_ctx *ctx = NULL;
1360 	MFT_RECORD *m = NULL;
1361 	u32 attr_len;
1362 	int err;
1363 
1364 retry_writepage:
1365 	BUG_ON(!PageLocked(page));
1366 	i_size = i_size_read(vi);
1367 	/* Is the page fully outside i_size? (truncate in progress) */
1368 	if (unlikely(page->index >= (i_size + PAGE_SIZE - 1) >>
1369 			PAGE_SHIFT)) {
1370 		/*
1371 		 * The page may have dirty, unmapped buffers.  Make them
1372 		 * freeable here, so the page does not leak.
1373 		 */
1374 		block_invalidatepage(page, 0, PAGE_SIZE);
1375 		unlock_page(page);
1376 		ntfs_debug("Write outside i_size - truncated?");
1377 		return 0;
1378 	}
1379 	/*
1380 	 * Only $DATA attributes can be encrypted and only unnamed $DATA
1381 	 * attributes can be compressed.  Index root can have the flags set but
1382 	 * this means to create compressed/encrypted files, not that the
1383 	 * attribute is compressed/encrypted.  Note we need to check for
1384 	 * AT_INDEX_ALLOCATION since this is the type of both directory and
1385 	 * index inodes.
1386 	 */
1387 	if (ni->type != AT_INDEX_ALLOCATION) {
1388 		/* If file is encrypted, deny access, just like NT4. */
1389 		if (NInoEncrypted(ni)) {
1390 			unlock_page(page);
1391 			BUG_ON(ni->type != AT_DATA);
1392 			ntfs_debug("Denying write access to encrypted file.");
1393 			return -EACCES;
1394 		}
1395 		/* Compressed data streams are handled in compress.c. */
1396 		if (NInoNonResident(ni) && NInoCompressed(ni)) {
1397 			BUG_ON(ni->type != AT_DATA);
1398 			BUG_ON(ni->name_len);
1399 			// TODO: Implement and replace this with
1400 			// return ntfs_write_compressed_block(page);
1401 			unlock_page(page);
1402 			ntfs_error(vi->i_sb, "Writing to compressed files is "
1403 					"not supported yet.  Sorry.");
1404 			return -EOPNOTSUPP;
1405 		}
1406 		// TODO: Implement and remove this check.
1407 		if (NInoNonResident(ni) && NInoSparse(ni)) {
1408 			unlock_page(page);
1409 			ntfs_error(vi->i_sb, "Writing to sparse files is not "
1410 					"supported yet.  Sorry.");
1411 			return -EOPNOTSUPP;
1412 		}
1413 	}
1414 	/* NInoNonResident() == NInoIndexAllocPresent() */
1415 	if (NInoNonResident(ni)) {
1416 		/* We have to zero every time due to mmap-at-end-of-file. */
1417 		if (page->index >= (i_size >> PAGE_SHIFT)) {
1418 			/* The page straddles i_size. */
1419 			unsigned int ofs = i_size & ~PAGE_MASK;
1420 			zero_user_segment(page, ofs, PAGE_SIZE);
1421 		}
1422 		/* Handle mst protected attributes. */
1423 		if (NInoMstProtected(ni))
1424 			return ntfs_write_mst_block(page, wbc);
1425 		/* Normal, non-resident data stream. */
1426 		return ntfs_write_block(page, wbc);
1427 	}
1428 	/*
1429 	 * Attribute is resident, implying it is not compressed, encrypted, or
1430 	 * mst protected.  This also means the attribute is smaller than an mft
1431 	 * record and hence smaller than a page, so can simply return error on
1432 	 * any pages with index above 0.  Note the attribute can actually be
1433 	 * marked compressed but if it is resident the actual data is not
1434 	 * compressed so we are ok to ignore the compressed flag here.
1435 	 */
1436 	BUG_ON(page_has_buffers(page));
1437 	BUG_ON(!PageUptodate(page));
1438 	if (unlikely(page->index > 0)) {
1439 		ntfs_error(vi->i_sb, "BUG()! page->index (0x%lx) > 0.  "
1440 				"Aborting write.", page->index);
1441 		BUG_ON(PageWriteback(page));
1442 		set_page_writeback(page);
1443 		unlock_page(page);
1444 		end_page_writeback(page);
1445 		return -EIO;
1446 	}
1447 	if (!NInoAttr(ni))
1448 		base_ni = ni;
1449 	else
1450 		base_ni = ni->ext.base_ntfs_ino;
1451 	/* Map, pin, and lock the mft record. */
1452 	m = map_mft_record(base_ni);
1453 	if (IS_ERR(m)) {
1454 		err = PTR_ERR(m);
1455 		m = NULL;
1456 		ctx = NULL;
1457 		goto err_out;
1458 	}
1459 	/*
1460 	 * If a parallel write made the attribute non-resident, drop the mft
1461 	 * record and retry the writepage.
1462 	 */
1463 	if (unlikely(NInoNonResident(ni))) {
1464 		unmap_mft_record(base_ni);
1465 		goto retry_writepage;
1466 	}
1467 	ctx = ntfs_attr_get_search_ctx(base_ni, m);
1468 	if (unlikely(!ctx)) {
1469 		err = -ENOMEM;
1470 		goto err_out;
1471 	}
1472 	err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1473 			CASE_SENSITIVE, 0, NULL, 0, ctx);
1474 	if (unlikely(err))
1475 		goto err_out;
1476 	/*
1477 	 * Keep the VM happy.  This must be done otherwise the radix-tree tag
1478 	 * PAGECACHE_TAG_DIRTY remains set even though the page is clean.
1479 	 */
1480 	BUG_ON(PageWriteback(page));
1481 	set_page_writeback(page);
1482 	unlock_page(page);
1483 	attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
1484 	i_size = i_size_read(vi);
1485 	if (unlikely(attr_len > i_size)) {
1486 		/* Race with shrinking truncate or a failed truncate. */
1487 		attr_len = i_size;
1488 		/*
1489 		 * If the truncate failed, fix it up now.  If a concurrent
1490 		 * truncate, we do its job, so it does not have to do anything.
1491 		 */
1492 		err = ntfs_resident_attr_value_resize(ctx->mrec, ctx->attr,
1493 				attr_len);
1494 		/* Shrinking cannot fail. */
1495 		BUG_ON(err);
1496 	}
1497 	addr = kmap_atomic(page);
1498 	/* Copy the data from the page to the mft record. */
1499 	memcpy((u8*)ctx->attr +
1500 			le16_to_cpu(ctx->attr->data.resident.value_offset),
1501 			addr, attr_len);
1502 	/* Zero out of bounds area in the page cache page. */
1503 	memset(addr + attr_len, 0, PAGE_SIZE - attr_len);
1504 	kunmap_atomic(addr);
1505 	flush_dcache_page(page);
1506 	flush_dcache_mft_record_page(ctx->ntfs_ino);
1507 	/* We are done with the page. */
1508 	end_page_writeback(page);
1509 	/* Finally, mark the mft record dirty, so it gets written back. */
1510 	mark_mft_record_dirty(ctx->ntfs_ino);
1511 	ntfs_attr_put_search_ctx(ctx);
1512 	unmap_mft_record(base_ni);
1513 	return 0;
1514 err_out:
1515 	if (err == -ENOMEM) {
1516 		ntfs_warning(vi->i_sb, "Error allocating memory. Redirtying "
1517 				"page so we try again later.");
1518 		/*
1519 		 * Put the page back on mapping->dirty_pages, but leave its
1520 		 * buffers' dirty state as-is.
1521 		 */
1522 		redirty_page_for_writepage(wbc, page);
1523 		err = 0;
1524 	} else {
1525 		ntfs_error(vi->i_sb, "Resident attribute write failed with "
1526 				"error %i.", err);
1527 		SetPageError(page);
1528 		NVolSetErrors(ni->vol);
1529 	}
1530 	unlock_page(page);
1531 	if (ctx)
1532 		ntfs_attr_put_search_ctx(ctx);
1533 	if (m)
1534 		unmap_mft_record(base_ni);
1535 	return err;
1536 }
1537 
1538 #endif	/* NTFS_RW */
1539 
1540 /**
1541  * ntfs_bmap - map logical file block to physical device block
1542  * @mapping:	address space mapping to which the block to be mapped belongs
1543  * @block:	logical block to map to its physical device block
1544  *
1545  * For regular, non-resident files (i.e. not compressed and not encrypted), map
1546  * the logical @block belonging to the file described by the address space
1547  * mapping @mapping to its physical device block.
1548  *
1549  * The size of the block is equal to the @s_blocksize field of the super block
1550  * of the mounted file system which is guaranteed to be smaller than or equal
1551  * to the cluster size thus the block is guaranteed to fit entirely inside the
1552  * cluster which means we do not need to care how many contiguous bytes are
1553  * available after the beginning of the block.
1554  *
1555  * Return the physical device block if the mapping succeeded or 0 if the block
1556  * is sparse or there was an error.
1557  *
1558  * Note: This is a problem if someone tries to run bmap() on $Boot system file
1559  * as that really is in block zero but there is nothing we can do.  bmap() is
1560  * just broken in that respect (just like it cannot distinguish sparse from
1561  * not available or error).
1562  */
ntfs_bmap(struct address_space * mapping,sector_t block)1563 static sector_t ntfs_bmap(struct address_space *mapping, sector_t block)
1564 {
1565 	s64 ofs, size;
1566 	loff_t i_size;
1567 	LCN lcn;
1568 	unsigned long blocksize, flags;
1569 	ntfs_inode *ni = NTFS_I(mapping->host);
1570 	ntfs_volume *vol = ni->vol;
1571 	unsigned delta;
1572 	unsigned char blocksize_bits, cluster_size_shift;
1573 
1574 	ntfs_debug("Entering for mft_no 0x%lx, logical block 0x%llx.",
1575 			ni->mft_no, (unsigned long long)block);
1576 	if (ni->type != AT_DATA || !NInoNonResident(ni) || NInoEncrypted(ni)) {
1577 		ntfs_error(vol->sb, "BMAP does not make sense for %s "
1578 				"attributes, returning 0.",
1579 				(ni->type != AT_DATA) ? "non-data" :
1580 				(!NInoNonResident(ni) ? "resident" :
1581 				"encrypted"));
1582 		return 0;
1583 	}
1584 	/* None of these can happen. */
1585 	BUG_ON(NInoCompressed(ni));
1586 	BUG_ON(NInoMstProtected(ni));
1587 	blocksize = vol->sb->s_blocksize;
1588 	blocksize_bits = vol->sb->s_blocksize_bits;
1589 	ofs = (s64)block << blocksize_bits;
1590 	read_lock_irqsave(&ni->size_lock, flags);
1591 	size = ni->initialized_size;
1592 	i_size = i_size_read(VFS_I(ni));
1593 	read_unlock_irqrestore(&ni->size_lock, flags);
1594 	/*
1595 	 * If the offset is outside the initialized size or the block straddles
1596 	 * the initialized size then pretend it is a hole unless the
1597 	 * initialized size equals the file size.
1598 	 */
1599 	if (unlikely(ofs >= size || (ofs + blocksize > size && size < i_size)))
1600 		goto hole;
1601 	cluster_size_shift = vol->cluster_size_bits;
1602 	down_read(&ni->runlist.lock);
1603 	lcn = ntfs_attr_vcn_to_lcn_nolock(ni, ofs >> cluster_size_shift, false);
1604 	up_read(&ni->runlist.lock);
1605 	if (unlikely(lcn < LCN_HOLE)) {
1606 		/*
1607 		 * Step down to an integer to avoid gcc doing a long long
1608 		 * comparision in the switch when we know @lcn is between
1609 		 * LCN_HOLE and LCN_EIO (i.e. -1 to -5).
1610 		 *
1611 		 * Otherwise older gcc (at least on some architectures) will
1612 		 * try to use __cmpdi2() which is of course not available in
1613 		 * the kernel.
1614 		 */
1615 		switch ((int)lcn) {
1616 		case LCN_ENOENT:
1617 			/*
1618 			 * If the offset is out of bounds then pretend it is a
1619 			 * hole.
1620 			 */
1621 			goto hole;
1622 		case LCN_ENOMEM:
1623 			ntfs_error(vol->sb, "Not enough memory to complete "
1624 					"mapping for inode 0x%lx.  "
1625 					"Returning 0.", ni->mft_no);
1626 			break;
1627 		default:
1628 			ntfs_error(vol->sb, "Failed to complete mapping for "
1629 					"inode 0x%lx.  Run chkdsk.  "
1630 					"Returning 0.", ni->mft_no);
1631 			break;
1632 		}
1633 		return 0;
1634 	}
1635 	if (lcn < 0) {
1636 		/* It is a hole. */
1637 hole:
1638 		ntfs_debug("Done (returning hole).");
1639 		return 0;
1640 	}
1641 	/*
1642 	 * The block is really allocated and fullfils all our criteria.
1643 	 * Convert the cluster to units of block size and return the result.
1644 	 */
1645 	delta = ofs & vol->cluster_size_mask;
1646 	if (unlikely(sizeof(block) < sizeof(lcn))) {
1647 		block = lcn = ((lcn << cluster_size_shift) + delta) >>
1648 				blocksize_bits;
1649 		/* If the block number was truncated return 0. */
1650 		if (unlikely(block != lcn)) {
1651 			ntfs_error(vol->sb, "Physical block 0x%llx is too "
1652 					"large to be returned, returning 0.",
1653 					(long long)lcn);
1654 			return 0;
1655 		}
1656 	} else
1657 		block = ((lcn << cluster_size_shift) + delta) >>
1658 				blocksize_bits;
1659 	ntfs_debug("Done (returning block 0x%llx).", (unsigned long long)lcn);
1660 	return block;
1661 }
1662 
1663 /**
1664  * ntfs_normal_aops - address space operations for normal inodes and attributes
1665  *
1666  * Note these are not used for compressed or mst protected inodes and
1667  * attributes.
1668  */
1669 const struct address_space_operations ntfs_normal_aops = {
1670 	.readpage	= ntfs_readpage,
1671 #ifdef NTFS_RW
1672 	.writepage	= ntfs_writepage,
1673 	.set_page_dirty	= __set_page_dirty_buffers,
1674 #endif /* NTFS_RW */
1675 	.bmap		= ntfs_bmap,
1676 	.migratepage	= buffer_migrate_page,
1677 	.is_partially_uptodate = block_is_partially_uptodate,
1678 	.error_remove_page = generic_error_remove_page,
1679 };
1680 
1681 /**
1682  * ntfs_compressed_aops - address space operations for compressed inodes
1683  */
1684 const struct address_space_operations ntfs_compressed_aops = {
1685 	.readpage	= ntfs_readpage,
1686 #ifdef NTFS_RW
1687 	.writepage	= ntfs_writepage,
1688 	.set_page_dirty	= __set_page_dirty_buffers,
1689 #endif /* NTFS_RW */
1690 	.migratepage	= buffer_migrate_page,
1691 	.is_partially_uptodate = block_is_partially_uptodate,
1692 	.error_remove_page = generic_error_remove_page,
1693 };
1694 
1695 /**
1696  * ntfs_mst_aops - general address space operations for mst protecteed inodes
1697  *		   and attributes
1698  */
1699 const struct address_space_operations ntfs_mst_aops = {
1700 	.readpage	= ntfs_readpage,	/* Fill page with data. */
1701 #ifdef NTFS_RW
1702 	.writepage	= ntfs_writepage,	/* Write dirty page to disk. */
1703 	.set_page_dirty	= __set_page_dirty_nobuffers,	/* Set the page dirty
1704 						   without touching the buffers
1705 						   belonging to the page. */
1706 #endif /* NTFS_RW */
1707 	.migratepage	= buffer_migrate_page,
1708 	.is_partially_uptodate	= block_is_partially_uptodate,
1709 	.error_remove_page = generic_error_remove_page,
1710 };
1711 
1712 #ifdef NTFS_RW
1713 
1714 /**
1715  * mark_ntfs_record_dirty - mark an ntfs record dirty
1716  * @page:	page containing the ntfs record to mark dirty
1717  * @ofs:	byte offset within @page at which the ntfs record begins
1718  *
1719  * Set the buffers and the page in which the ntfs record is located dirty.
1720  *
1721  * The latter also marks the vfs inode the ntfs record belongs to dirty
1722  * (I_DIRTY_PAGES only).
1723  *
1724  * If the page does not have buffers, we create them and set them uptodate.
1725  * The page may not be locked which is why we need to handle the buffers under
1726  * the mapping->private_lock.  Once the buffers are marked dirty we no longer
1727  * need the lock since try_to_free_buffers() does not free dirty buffers.
1728  */
mark_ntfs_record_dirty(struct page * page,const unsigned int ofs)1729 void mark_ntfs_record_dirty(struct page *page, const unsigned int ofs) {
1730 	struct address_space *mapping = page->mapping;
1731 	ntfs_inode *ni = NTFS_I(mapping->host);
1732 	struct buffer_head *bh, *head, *buffers_to_free = NULL;
1733 	unsigned int end, bh_size, bh_ofs;
1734 
1735 	BUG_ON(!PageUptodate(page));
1736 	end = ofs + ni->itype.index.block_size;
1737 	bh_size = VFS_I(ni)->i_sb->s_blocksize;
1738 	spin_lock(&mapping->private_lock);
1739 	if (unlikely(!page_has_buffers(page))) {
1740 		spin_unlock(&mapping->private_lock);
1741 		bh = head = alloc_page_buffers(page, bh_size, true);
1742 		spin_lock(&mapping->private_lock);
1743 		if (likely(!page_has_buffers(page))) {
1744 			struct buffer_head *tail;
1745 
1746 			do {
1747 				set_buffer_uptodate(bh);
1748 				tail = bh;
1749 				bh = bh->b_this_page;
1750 			} while (bh);
1751 			tail->b_this_page = head;
1752 			attach_page_buffers(page, head);
1753 		} else
1754 			buffers_to_free = bh;
1755 	}
1756 	bh = head = page_buffers(page);
1757 	BUG_ON(!bh);
1758 	do {
1759 		bh_ofs = bh_offset(bh);
1760 		if (bh_ofs + bh_size <= ofs)
1761 			continue;
1762 		if (unlikely(bh_ofs >= end))
1763 			break;
1764 		set_buffer_dirty(bh);
1765 	} while ((bh = bh->b_this_page) != head);
1766 	spin_unlock(&mapping->private_lock);
1767 	__set_page_dirty_nobuffers(page);
1768 	if (unlikely(buffers_to_free)) {
1769 		do {
1770 			bh = buffers_to_free->b_this_page;
1771 			free_buffer_head(buffers_to_free);
1772 			buffers_to_free = bh;
1773 		} while (buffers_to_free);
1774 	}
1775 }
1776 
1777 #endif /* NTFS_RW */
1778