2 * aops.c - NTFS kernel address space operations and page cache handling.
3 * Part of the Linux-NTFS project.
5 * Copyright (c) 2001-2004 Anton Altaparmakov
6 * Copyright (c) 2002 Richard Russon
8 * This program/include file is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as published
10 * by the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program/include file is distributed in the hope that it will be
14 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
15 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program (in the main directory of the Linux-NTFS
20 * distribution in the file COPYING); if not, write to the Free Software
21 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 #include <linux/errno.h>
26 #include <linux/pagemap.h>
27 #include <linux/swap.h>
28 #include <linux/buffer_head.h>
33 * ntfs_end_buffer_async_read - async io completion for reading attributes
34 * @bh: buffer head on which io is completed
35 * @uptodate: whether @bh is now uptodate or not
37 * Asynchronous I/O completion handler for reading pages belonging to the
38 * attribute address space of an inode. The inodes can either be files or
39 * directories or they can be fake inodes describing some attribute.
41 * If NInoMstProtected(), perform the post read mst fixups when all IO on the
42 * page has been completed and mark the page uptodate or set the error bit on
43 * the page. To determine the size of the records that need fixing up, we cheat
44 * a little bit by setting the index_block_size in ntfs_inode to the ntfs
45 * record size, and index_block_size_bits, to the log(base 2) of the ntfs
48 static void ntfs_end_buffer_async_read(struct buffer_head
*bh
, int uptodate
)
50 static spinlock_t page_uptodate_lock
= SPIN_LOCK_UNLOCKED
;
52 struct buffer_head
*tmp
;
55 int page_uptodate
= 1;
58 ni
= NTFS_I(page
->mapping
->host
);
60 if (likely(uptodate
)) {
63 set_buffer_uptodate(bh
);
65 file_ofs
= ((s64
)page
->index
<< PAGE_CACHE_SHIFT
) +
67 /* Check for the current buffer head overflowing. */
68 if (file_ofs
+ bh
->b_size
> ni
->initialized_size
) {
72 if (file_ofs
< ni
->initialized_size
)
73 ofs
= ni
->initialized_size
- file_ofs
;
74 addr
= kmap_atomic(page
, KM_BIO_SRC_IRQ
);
75 memset(addr
+ bh_offset(bh
) + ofs
, 0, bh
->b_size
- ofs
);
76 flush_dcache_page(page
);
77 kunmap_atomic(addr
, KM_BIO_SRC_IRQ
);
80 clear_buffer_uptodate(bh
);
81 ntfs_error(ni
->vol
->sb
, "Buffer I/O error, logical block %llu.",
82 (unsigned long long)bh
->b_blocknr
);
86 spin_lock_irqsave(&page_uptodate_lock
, flags
);
87 clear_buffer_async_read(bh
);
91 if (!buffer_uptodate(tmp
))
93 if (buffer_async_read(tmp
)) {
94 if (likely(buffer_locked(tmp
)))
96 /* Async buffers must be locked. */
99 tmp
= tmp
->b_this_page
;
101 spin_unlock_irqrestore(&page_uptodate_lock
, flags
);
103 * If none of the buffers had errors then we can set the page uptodate,
104 * but we first have to perform the post read mst fixups, if the
105 * attribute is mst protected, i.e. if NInoMstProteced(ni) is true.
107 if (!NInoMstProtected(ni
)) {
108 if (likely(page_uptodate
&& !PageError(page
)))
109 SetPageUptodate(page
);
112 unsigned int i
, recs
, nr_err
;
115 rec_size
= ni
->itype
.index
.block_size
;
116 recs
= PAGE_CACHE_SIZE
/ rec_size
;
117 addr
= kmap_atomic(page
, KM_BIO_SRC_IRQ
);
118 for (i
= nr_err
= 0; i
< recs
; i
++) {
119 if (likely(!post_read_mst_fixup((NTFS_RECORD
*)(addr
+
120 i
* rec_size
), rec_size
)))
123 ntfs_error(ni
->vol
->sb
, "post_read_mst_fixup() failed, "
124 "corrupt %s record 0x%llx. Run chkdsk.",
125 ni
->mft_no
? "index" : "mft",
126 (unsigned long long)(((s64
)page
->index
127 << PAGE_CACHE_SHIFT
>>
128 ni
->itype
.index
.block_size_bits
) + i
));
130 flush_dcache_page(page
);
131 kunmap_atomic(addr
, KM_BIO_SRC_IRQ
);
132 if (likely(!PageError(page
))) {
133 if (likely(!nr_err
&& recs
)) {
134 if (likely(page_uptodate
))
135 SetPageUptodate(page
);
137 ntfs_error(ni
->vol
->sb
, "Setting page error, "
138 "index 0x%lx.", page
->index
);
146 spin_unlock_irqrestore(&page_uptodate_lock
, flags
);
151 * ntfs_read_block - fill a @page of an address space with data
152 * @page: page cache page to fill with data
154 * Fill the page @page of the address space belonging to the @page->host inode.
155 * We read each buffer asynchronously and when all buffers are read in, our io
156 * completion handler ntfs_end_buffer_read_async(), if required, automatically
157 * applies the mst fixups to the page before finally marking it uptodate and
160 * We only enforce allocated_size limit because i_size is checked for in
161 * generic_file_read().
163 * Return 0 on success and -errno on error.
165 * Contains an adapted version of fs/buffer.c::block_read_full_page().
167 static int ntfs_read_block(struct page
*page
)
174 struct buffer_head
*bh
, *head
, *arr
[MAX_BUF_PER_PAGE
];
175 sector_t iblock
, lblock
, zblock
;
176 unsigned int blocksize
, vcn_ofs
;
178 unsigned char blocksize_bits
;
180 ni
= NTFS_I(page
->mapping
->host
);
183 blocksize_bits
= VFS_I(ni
)->i_blkbits
;
184 blocksize
= 1 << blocksize_bits
;
186 if (!page_has_buffers(page
))
187 create_empty_buffers(page
, blocksize
, 0);
188 bh
= head
= page_buffers(page
);
194 iblock
= (s64
)page
->index
<< (PAGE_CACHE_SHIFT
- blocksize_bits
);
195 lblock
= (ni
->allocated_size
+ blocksize
- 1) >> blocksize_bits
;
196 zblock
= (ni
->initialized_size
+ blocksize
- 1) >> blocksize_bits
;
199 if (unlikely(!ni
->runlist
.rl
&& !ni
->mft_no
&& !NInoAttr(ni
)))
200 panic("NTFS: $MFT/$DATA runlist has been unmapped! This is a "
201 "very serious bug! Cannot continue...");
204 /* Loop through all the buffers in the page. */
210 if (unlikely(buffer_uptodate(bh
)))
212 if (unlikely(buffer_mapped(bh
))) {
216 bh
->b_bdev
= vol
->sb
->s_bdev
;
217 /* Is the block within the allowed limits? */
218 if (iblock
< lblock
) {
219 BOOL is_retry
= FALSE
;
221 /* Convert iblock into corresponding vcn and offset. */
222 vcn
= (VCN
)iblock
<< blocksize_bits
>>
223 vol
->cluster_size_bits
;
224 vcn_ofs
= ((VCN
)iblock
<< blocksize_bits
) &
225 vol
->cluster_size_mask
;
228 down_read(&ni
->runlist
.lock
);
231 if (likely(rl
!= NULL
)) {
232 /* Seek to element containing target vcn. */
233 while (rl
->length
&& rl
[1].vcn
<= vcn
)
235 lcn
= ntfs_vcn_to_lcn(rl
, vcn
);
237 lcn
= (LCN
)LCN_RL_NOT_MAPPED
;
238 /* Successful remap. */
240 /* Setup buffer head to correct block. */
241 bh
->b_blocknr
= ((lcn
<< vol
->cluster_size_bits
)
242 + vcn_ofs
) >> blocksize_bits
;
243 set_buffer_mapped(bh
);
244 /* Only read initialized data blocks. */
245 if (iblock
< zblock
) {
249 /* Fully non-initialized data block, zero it. */
252 /* It is a hole, need to zero it. */
255 /* If first try and runlist unmapped, map and retry. */
256 if (!is_retry
&& lcn
== LCN_RL_NOT_MAPPED
) {
259 * Attempt to map runlist, dropping lock for
262 up_read(&ni
->runlist
.lock
);
263 if (!ntfs_map_runlist(ni
, vcn
))
264 goto lock_retry_remap
;
267 /* Hard error, zero out region. */
269 ntfs_error(vol
->sb
, "ntfs_vcn_to_lcn(vcn = 0x%llx) "
270 "failed with error code 0x%llx%s.",
271 (unsigned long long)vcn
,
272 (unsigned long long)-lcn
,
273 is_retry
? " even after retrying" : "");
274 // FIXME: Depending on vol->on_errors, do something.
277 * Either iblock was outside lblock limits or ntfs_vcn_to_lcn()
278 * returned error. Just zero that portion of the page and set
279 * the buffer uptodate.
282 bh
->b_blocknr
= -1UL;
283 clear_buffer_mapped(bh
);
285 kaddr
= kmap_atomic(page
, KM_USER0
);
286 memset(kaddr
+ i
* blocksize
, 0, blocksize
);
287 flush_dcache_page(page
);
288 kunmap_atomic(kaddr
, KM_USER0
);
289 set_buffer_uptodate(bh
);
290 } while (i
++, iblock
++, (bh
= bh
->b_this_page
) != head
);
292 /* Release the lock if we took it. */
294 up_read(&ni
->runlist
.lock
);
296 /* Check we have at least one buffer ready for i/o. */
298 struct buffer_head
*tbh
;
300 /* Lock the buffers. */
301 for (i
= 0; i
< nr
; i
++) {
304 tbh
->b_end_io
= ntfs_end_buffer_async_read
;
305 set_buffer_async_read(tbh
);
307 /* Finally, start i/o on the buffers. */
308 for (i
= 0; i
< nr
; i
++) {
310 if (likely(!buffer_uptodate(tbh
)))
311 submit_bh(READ
, tbh
);
313 ntfs_end_buffer_async_read(tbh
, 1);
317 /* No i/o was scheduled on any of the buffers. */
318 if (likely(!PageError(page
)))
319 SetPageUptodate(page
);
320 else /* Signal synchronous i/o error. */
327 * ntfs_readpage - fill a @page of a @file with data from the device
328 * @file: open file to which the page @page belongs or NULL
329 * @page: page cache page to fill with data
331 * For non-resident attributes, ntfs_readpage() fills the @page of the open
332 * file @file by calling the ntfs version of the generic block_read_full_page()
333 * function, ntfs_read_block(), which in turn creates and reads in the buffers
334 * associated with the page asynchronously.
336 * For resident attributes, OTOH, ntfs_readpage() fills @page by copying the
337 * data from the mft record (which at this stage is most likely in memory) and
338 * fills the remainder with zeroes. Thus, in this case, I/O is synchronous, as
339 * even if the mft record is not cached at this point in time, we need to wait
340 * for it to be read in before we can do the copy.
342 * Return 0 on success and -errno on error.
344 * WARNING: Do not make this function static! It is used by mft.c!
346 int ntfs_readpage(struct file
*file
, struct page
*page
)
349 ntfs_inode
*ni
, *base_ni
;
351 ntfs_attr_search_ctx
*ctx
;
356 BUG_ON(!PageLocked(page
));
359 * This can potentially happen because we clear PageUptodate() during
360 * ntfs_writepage() of MstProtected() attributes.
362 if (PageUptodate(page
)) {
367 ni
= NTFS_I(page
->mapping
->host
);
369 /* NInoNonResident() == NInoIndexAllocPresent() */
370 if (NInoNonResident(ni
)) {
372 * Only unnamed $DATA attributes can be compressed or
375 if (ni
->type
== AT_DATA
&& !ni
->name_len
) {
376 /* If file is encrypted, deny access, just like NT4. */
377 if (NInoEncrypted(ni
)) {
381 /* Compressed data streams are handled in compress.c. */
382 if (NInoCompressed(ni
))
383 return ntfs_read_compressed_block(page
);
385 /* Normal data stream. */
386 return ntfs_read_block(page
);
388 /* Attribute is resident, implying it is not compressed or encrypted. */
392 base_ni
= ni
->ext
.base_ntfs_ino
;
394 /* Map, pin, and lock the mft record. */
395 mrec
= map_mft_record(base_ni
);
400 ctx
= ntfs_attr_get_search_ctx(base_ni
, mrec
);
401 if (unlikely(!ctx
)) {
405 err
= ntfs_attr_lookup(ni
->type
, ni
->name
, ni
->name_len
,
406 CASE_SENSITIVE
, 0, NULL
, 0, ctx
);
408 goto put_unm_err_out
;
410 /* Starting position of the page within the attribute value. */
411 attr_pos
= page
->index
<< PAGE_CACHE_SHIFT
;
413 /* The total length of the attribute value. */
414 attr_len
= le32_to_cpu(ctx
->attr
->data
.resident
.value_length
);
416 kaddr
= kmap_atomic(page
, KM_USER0
);
417 /* Copy over in bounds data, zeroing the remainder of the page. */
418 if (attr_pos
< attr_len
) {
419 u32 bytes
= attr_len
- attr_pos
;
420 if (bytes
> PAGE_CACHE_SIZE
)
421 bytes
= PAGE_CACHE_SIZE
;
422 else if (bytes
< PAGE_CACHE_SIZE
)
423 memset(kaddr
+ bytes
, 0, PAGE_CACHE_SIZE
- bytes
);
424 /* Copy the data to the page. */
425 memcpy(kaddr
, attr_pos
+ (char*)ctx
->attr
+
427 ctx
->attr
->data
.resident
.value_offset
), bytes
);
429 memset(kaddr
, 0, PAGE_CACHE_SIZE
);
430 flush_dcache_page(page
);
431 kunmap_atomic(kaddr
, KM_USER0
);
433 SetPageUptodate(page
);
435 ntfs_attr_put_search_ctx(ctx
);
437 unmap_mft_record(base_ni
);
446 * ntfs_write_block - write a @page to the backing store
447 * @wbc: writeback control structure
448 * @page: page cache page to write out
450 * This function is for writing pages belonging to non-resident, non-mst
451 * protected attributes to their backing store.
453 * For a page with buffers, map and write the dirty buffers asynchronously
454 * under page writeback. For a page without buffers, create buffers for the
455 * page, then proceed as above.
457 * If a page doesn't have buffers the page dirty state is definitive. If a page
458 * does have buffers, the page dirty state is just a hint, and the buffer dirty
459 * state is definitive. (A hint which has rules: dirty buffers against a clean
460 * page is illegal. Other combinations are legal and need to be handled. In
461 * particular a dirty page containing clean buffers for example.)
463 * Return 0 on success and -errno on error.
465 * Based on ntfs_read_block() and __block_write_full_page().
467 static int ntfs_write_block(struct writeback_control
*wbc
, struct page
*page
)
471 sector_t block
, dblock
, iblock
;
476 struct buffer_head
*bh
, *head
;
477 unsigned int blocksize
, vcn_ofs
;
479 BOOL need_end_writeback
;
480 unsigned char blocksize_bits
;
482 vi
= page
->mapping
->host
;
486 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
487 "0x%lx.", vi
->i_ino
, ni
->type
, page
->index
);
489 BUG_ON(!NInoNonResident(ni
));
490 BUG_ON(NInoMstProtected(ni
));
492 blocksize_bits
= vi
->i_blkbits
;
493 blocksize
= 1 << blocksize_bits
;
495 if (!page_has_buffers(page
)) {
496 BUG_ON(!PageUptodate(page
));
497 create_empty_buffers(page
, blocksize
,
498 (1 << BH_Uptodate
) | (1 << BH_Dirty
));
500 bh
= head
= page_buffers(page
);
502 ntfs_warning(vol
->sb
, "Error allocating page buffers. "
503 "Redirtying page so we try again later.");
505 * Put the page back on mapping->dirty_pages, but leave its
506 * buffer's dirty state as-is.
508 redirty_page_for_writepage(wbc
, page
);
513 /* NOTE: Different naming scheme to ntfs_read_block()! */
515 /* The first block in the page. */
516 block
= (s64
)page
->index
<< (PAGE_CACHE_SHIFT
- blocksize_bits
);
518 /* The first out of bounds block for the data size. */
519 dblock
= (vi
->i_size
+ blocksize
- 1) >> blocksize_bits
;
521 /* The last (fully or partially) initialized block. */
522 iblock
= ni
->initialized_size
>> blocksize_bits
;
525 * Be very careful. We have no exclusion from __set_page_dirty_buffers
526 * here, and the (potentially unmapped) buffers may become dirty at
527 * any time. If a buffer becomes dirty here after we've inspected it
528 * then we just miss that fact, and the page stays dirty.
530 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
531 * handle that here by just cleaning them.
535 * Loop through all the buffers in the page, mapping all the dirty
536 * buffers to disk addresses and handling any aliases from the
537 * underlying block device's mapping.
542 BOOL is_retry
= FALSE
;
544 if (unlikely(block
>= dblock
)) {
546 * Mapped buffers outside i_size will occur, because
547 * this page can be outside i_size when there is a
548 * truncate in progress. The contents of such buffers
549 * were zeroed by ntfs_writepage().
551 * FIXME: What about the small race window where
552 * ntfs_writepage() has not done any clearing because
553 * the page was within i_size but before we get here,
554 * vmtruncate() modifies i_size?
556 clear_buffer_dirty(bh
);
557 set_buffer_uptodate(bh
);
561 /* Clean buffers are not written out, so no need to map them. */
562 if (!buffer_dirty(bh
))
565 /* Make sure we have enough initialized size. */
566 if (unlikely((block
>= iblock
) &&
567 (ni
->initialized_size
< vi
->i_size
))) {
569 * If this page is fully outside initialized size, zero
570 * out all pages between the current initialized size
571 * and the current page. Just use ntfs_readpage() to do
572 * the zeroing transparently.
574 if (block
> iblock
) {
577 // - read_cache_page()
578 // Again for each page do:
579 // - wait_on_page_locked()
580 // - Check (PageUptodate(page) &&
582 // Update initialized size in the attribute and
584 // Again, for each page do:
585 // __set_page_dirty_buffers();
586 // page_cache_release()
587 // We don't need to wait on the writes.
591 * The current page straddles initialized size. Zero
592 * all non-uptodate buffers and set them uptodate (and
593 * dirty?). Note, there aren't any non-uptodate buffers
594 * if the page is uptodate.
595 * FIXME: For an uptodate page, the buffers may need to
596 * be written out because they were not initialized on
599 if (!PageUptodate(page
)) {
601 // Zero any non-uptodate buffers up to i_size.
602 // Set them uptodate and dirty.
605 // Update initialized size in the attribute and in the
606 // inode (up to i_size).
608 // FIXME: This is inefficient. Try to batch the two
609 // size changes to happen in one go.
610 ntfs_error(vol
->sb
, "Writing beyond initialized size "
611 "is not supported yet. Sorry.");
614 // Do NOT set_buffer_new() BUT DO clear buffer range
615 // outside write request range.
616 // set_buffer_uptodate() on complete buffers as well as
617 // set_buffer_dirty().
620 /* No need to map buffers that are already mapped. */
621 if (buffer_mapped(bh
))
624 /* Unmapped, dirty buffer. Need to map it. */
625 bh
->b_bdev
= vol
->sb
->s_bdev
;
627 /* Convert block into corresponding vcn and offset. */
628 vcn
= (VCN
)block
<< blocksize_bits
>> vol
->cluster_size_bits
;
629 vcn_ofs
= ((VCN
)block
<< blocksize_bits
) &
630 vol
->cluster_size_mask
;
633 down_read(&ni
->runlist
.lock
);
636 if (likely(rl
!= NULL
)) {
637 /* Seek to element containing target vcn. */
638 while (rl
->length
&& rl
[1].vcn
<= vcn
)
640 lcn
= ntfs_vcn_to_lcn(rl
, vcn
);
642 lcn
= (LCN
)LCN_RL_NOT_MAPPED
;
643 /* Successful remap. */
645 /* Setup buffer head to point to correct block. */
646 bh
->b_blocknr
= ((lcn
<< vol
->cluster_size_bits
) +
647 vcn_ofs
) >> blocksize_bits
;
648 set_buffer_mapped(bh
);
651 /* It is a hole, need to instantiate it. */
652 if (lcn
== LCN_HOLE
) {
653 // TODO: Instantiate the hole.
654 // clear_buffer_new(bh);
655 // unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
656 ntfs_error(vol
->sb
, "Writing into sparse regions is "
657 "not supported yet. Sorry.");
661 /* If first try and runlist unmapped, map and retry. */
662 if (!is_retry
&& lcn
== LCN_RL_NOT_MAPPED
) {
665 * Attempt to map runlist, dropping lock for
668 up_read(&ni
->runlist
.lock
);
669 err
= ntfs_map_runlist(ni
, vcn
);
671 goto lock_retry_remap
;
674 /* Failed to map the buffer, even after retrying. */
675 bh
->b_blocknr
= -1UL;
676 ntfs_error(vol
->sb
, "ntfs_vcn_to_lcn(vcn = 0x%llx) failed "
677 "with error code 0x%llx%s.",
678 (unsigned long long)vcn
,
679 (unsigned long long)-lcn
,
680 is_retry
? " even after retrying" : "");
681 // FIXME: Depending on vol->on_errors, do something.
685 } while (block
++, (bh
= bh
->b_this_page
) != head
);
687 /* Release the lock if we took it. */
689 up_read(&ni
->runlist
.lock
);
691 /* For the error case, need to reset bh to the beginning. */
694 /* Just an optimization, so ->readpage() isn't called later. */
695 if (unlikely(!PageUptodate(page
))) {
698 if (!buffer_uptodate(bh
)) {
703 } while ((bh
= bh
->b_this_page
) != head
);
705 SetPageUptodate(page
);
708 /* Setup all mapped, dirty buffers for async write i/o. */
711 if (buffer_mapped(bh
) && buffer_dirty(bh
)) {
713 if (test_clear_buffer_dirty(bh
)) {
714 BUG_ON(!buffer_uptodate(bh
));
715 mark_buffer_async_write(bh
);
718 } else if (unlikely(err
)) {
720 * For the error case. The buffer may have been set
721 * dirty during attachment to a dirty page.
724 clear_buffer_dirty(bh
);
726 } while ((bh
= bh
->b_this_page
) != head
);
729 // TODO: Remove the -EOPNOTSUPP check later on...
730 if (unlikely(err
== -EOPNOTSUPP
))
732 else if (err
== -ENOMEM
) {
733 ntfs_warning(vol
->sb
, "Error allocating memory. "
734 "Redirtying page so we try again "
737 * Put the page back on mapping->dirty_pages, but
738 * leave its buffer's dirty state as-is.
740 redirty_page_for_writepage(wbc
, page
);
746 BUG_ON(PageWriteback(page
));
747 set_page_writeback(page
); /* Keeps try_to_free_buffers() away. */
751 * Submit the prepared buffers for i/o. Note the page is unlocked,
752 * and the async write i/o completion handler can end_page_writeback()
753 * at any time after the *first* submit_bh(). So the buffers can then
756 need_end_writeback
= TRUE
;
758 struct buffer_head
*next
= bh
->b_this_page
;
759 if (buffer_async_write(bh
)) {
760 submit_bh(WRITE
, bh
);
761 need_end_writeback
= FALSE
;
765 } while (bh
!= head
);
767 /* If no i/o was started, need to end_page_writeback(). */
768 if (unlikely(need_end_writeback
))
769 end_page_writeback(page
);
775 static const char *ntfs_please_email
= "Please email "
776 "linux-ntfs-dev@lists.sourceforge.net and say that you saw "
777 "this message. Thank you.";
780 * ntfs_write_mst_block - write a @page to the backing store
781 * @wbc: writeback control structure
782 * @page: page cache page to write out
784 * This function is for writing pages belonging to non-resident, mst protected
785 * attributes to their backing store. The only supported attribute is the
786 * index allocation attribute. Both directory inodes and index inodes are
789 * The page must remain locked for the duration of the write because we apply
790 * the mst fixups, write, and then undo the fixups, so if we were to unlock the
791 * page before undoing the fixups, any other user of the page will see the
792 * page contents as corrupt.
794 * Return 0 on success and -errno on error.
796 * Based on ntfs_write_block(), ntfs_mft_writepage(), and
797 * write_mft_record_nolock().
799 static int ntfs_write_mst_block(struct writeback_control
*wbc
,
802 sector_t block
, dblock
, rec_block
;
803 struct inode
*vi
= page
->mapping
->host
;
804 ntfs_inode
*ni
= NTFS_I(vi
);
805 ntfs_volume
*vol
= ni
->vol
;
807 unsigned int bh_size
= 1 << vi
->i_blkbits
;
808 unsigned int rec_size
;
809 struct buffer_head
*bh
, *head
;
810 int max_bhs
= PAGE_CACHE_SIZE
/ bh_size
;
811 struct buffer_head
*bhs
[max_bhs
];
812 int i
, nr_recs
, nr_bhs
, bhs_per_rec
, err
;
813 unsigned char bh_size_bits
;
816 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
817 "0x%lx.", vi
->i_ino
, ni
->type
, page
->index
);
818 BUG_ON(!NInoNonResident(ni
));
819 BUG_ON(!NInoMstProtected(ni
));
820 BUG_ON(!(S_ISDIR(vi
->i_mode
) ||
821 (NInoAttr(ni
) && ni
->type
== AT_INDEX_ALLOCATION
)));
822 BUG_ON(PageWriteback(page
));
823 BUG_ON(!PageUptodate(page
));
826 /* Make sure we have mapped buffers. */
827 if (unlikely(!page_has_buffers(page
))) {
829 ntfs_error(vol
->sb
, "Writing ntfs records without existing "
830 "buffers is not implemented yet. %s",
835 bh
= head
= page_buffers(page
);
837 goto no_buffers_err_out
;
839 bh_size_bits
= vi
->i_blkbits
;
840 rec_size
= ni
->itype
.index
.block_size
;
841 nr_recs
= PAGE_CACHE_SIZE
/ rec_size
;
843 bhs_per_rec
= rec_size
>> bh_size_bits
;
844 BUG_ON(!bhs_per_rec
);
846 /* The first block in the page. */
847 rec_block
= block
= (s64
)page
->index
<<
848 (PAGE_CACHE_SHIFT
- bh_size_bits
);
850 /* The first out of bounds block for the data size. */
851 dblock
= (vi
->i_size
+ bh_size
- 1) >> bh_size_bits
;
854 /* Need this to silence a stupid gcc warning. */
855 rec_is_dirty
= FALSE
;
857 if (unlikely(block
>= dblock
)) {
859 * Mapped buffers outside i_size will occur, because
860 * this page can be outside i_size when there is a
861 * truncate in progress. The contents of such buffers
862 * were zeroed by ntfs_writepage().
864 * FIXME: What about the small race window where
865 * ntfs_writepage() has not done any clearing because
866 * the page was within i_size but before we get here,
867 * vmtruncate() modifies i_size?
869 clear_buffer_dirty(bh
);
872 if (rec_block
== block
) {
873 /* This block is the first one in the record. */
874 rec_block
+= rec_size
>> bh_size_bits
;
875 if (!buffer_dirty(bh
)) {
876 /* Clean buffers are not written out. */
877 rec_is_dirty
= FALSE
;
882 /* This block is not the first one in the record. */
883 if (!buffer_dirty(bh
)) {
884 /* Clean buffers are not written out. */
885 BUG_ON(rec_is_dirty
);
888 BUG_ON(!rec_is_dirty
);
890 /* Attempting to write outside the initialized size is a bug. */
891 BUG_ON(((block
+ 1) << bh_size_bits
) > ni
->initialized_size
);
892 if (!buffer_mapped(bh
)) {
893 ntfs_error(vol
->sb
, "Writing ntfs records without "
894 "existing mapped buffers is not "
895 "implemented yet. %s",
897 clear_buffer_dirty(bh
);
901 if (!buffer_uptodate(bh
)) {
902 ntfs_error(vol
->sb
, "Writing ntfs records without "
903 "existing uptodate buffers is not "
904 "implemented yet. %s",
906 clear_buffer_dirty(bh
);
911 BUG_ON(nr_bhs
> max_bhs
);
912 } while (block
++, (bh
= bh
->b_this_page
) != head
);
913 /* If there were no dirty buffers, we are done. */
916 /* Apply the mst protection fixups. */
917 kaddr
= page_address(page
);
918 for (i
= 0; i
< nr_bhs
; i
++) {
919 if (!(i
% bhs_per_rec
)) {
920 err
= pre_write_mst_fixup((NTFS_RECORD
*)(kaddr
+
921 bh_offset(bhs
[i
])), rec_size
);
923 ntfs_error(vol
->sb
, "Failed to apply mst "
924 "fixups (inode 0x%lx, "
925 "attribute type 0x%x, page "
926 "index 0x%lx)! Umount and "
927 "run chkdsk.", vi
->i_ino
,
931 goto mst_cleanup_out
;
935 flush_dcache_page(page
);
936 /* Lock buffers and start synchronous write i/o on them. */
937 for (i
= 0; i
< nr_bhs
; i
++) {
938 struct buffer_head
*tbh
= bhs
[i
];
940 if (unlikely(test_set_buffer_locked(tbh
)))
942 if (unlikely(!test_clear_buffer_dirty(tbh
))) {
946 BUG_ON(!buffer_uptodate(tbh
));
947 BUG_ON(!buffer_mapped(tbh
));
949 tbh
->b_end_io
= end_buffer_write_sync
;
950 submit_bh(WRITE
, tbh
);
952 /* Wait on i/o completion of buffers. */
953 for (i
= 0; i
< nr_bhs
; i
++) {
954 struct buffer_head
*tbh
= bhs
[i
];
957 if (unlikely(!buffer_uptodate(tbh
))) {
960 * Set the buffer uptodate so the page & buffer states
961 * don't become out of sync.
963 if (PageUptodate(page
))
964 set_buffer_uptodate(tbh
);
967 /* Remove the mst protection fixups again. */
968 for (i
= 0; i
< nr_bhs
; i
++) {
969 if (!(i
% bhs_per_rec
))
970 post_write_mst_fixup((NTFS_RECORD
*)(kaddr
+
973 flush_dcache_page(page
);
975 /* I/O error during writing. This is really bad! */
976 ntfs_error(vol
->sb
, "I/O error while writing ntfs record "
977 "(inode 0x%lx, attribute type 0x%x, page "
978 "index 0x%lx)! Umount and run chkdsk.",
979 vi
->i_ino
, ni
->type
, page
->index
);
983 set_page_writeback(page
);
985 end_page_writeback(page
);
990 /* Remove the mst protection fixups again. */
991 for (i
= 0; i
< nr_bhs
; i
++) {
992 if (!(i
% bhs_per_rec
))
993 post_write_mst_fixup((NTFS_RECORD
*)(kaddr
+
997 /* Clean the buffers. */
998 for (i
= 0; i
< nr_bhs
; i
++)
999 clear_buffer_dirty(bhs
[i
]);
1006 * ntfs_writepage - write a @page to the backing store
1007 * @page: page cache page to write out
1008 * @wbc: writeback control structure
1010 * For non-resident attributes, ntfs_writepage() writes the @page by calling
1011 * the ntfs version of the generic block_write_full_page() function,
1012 * ntfs_write_block(), which in turn if necessary creates and writes the
1013 * buffers associated with the page asynchronously.
1015 * For resident attributes, OTOH, ntfs_writepage() writes the @page by copying
1016 * the data to the mft record (which at this stage is most likely in memory).
1017 * The mft record is then marked dirty and written out asynchronously via the
1018 * vfs inode dirty code path.
1020 * Note the caller clears the page dirty flag before calling ntfs_writepage().
1022 * Based on ntfs_readpage() and fs/buffer.c::block_write_full_page().
1024 * Return 0 on success and -errno on error.
1026 static int ntfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
1030 ntfs_inode
*ni
, *base_ni
;
1032 ntfs_attr_search_ctx
*ctx
;
1034 u32 attr_len
, bytes
;
1037 BUG_ON(!PageLocked(page
));
1039 vi
= page
->mapping
->host
;
1041 /* Is the page fully outside i_size? (truncate in progress) */
1042 if (unlikely(page
->index
>= (vi
->i_size
+ PAGE_CACHE_SIZE
- 1) >>
1043 PAGE_CACHE_SHIFT
)) {
1045 ntfs_debug("Write outside i_size - truncated?");
1051 /* NInoNonResident() == NInoIndexAllocPresent() */
1052 if (NInoNonResident(ni
)) {
1054 * Only unnamed $DATA attributes can be compressed, encrypted,
1057 if (ni
->type
== AT_DATA
&& !ni
->name_len
) {
1058 /* If file is encrypted, deny access, just like NT4. */
1059 if (NInoEncrypted(ni
)) {
1061 ntfs_debug("Denying write access to encrypted "
1065 /* Compressed data streams are handled in compress.c. */
1066 if (NInoCompressed(ni
)) {
1067 // TODO: Implement and replace this check with
1068 // return ntfs_write_compressed_block(page);
1070 ntfs_error(vi
->i_sb
, "Writing to compressed "
1071 "files is not supported yet. "
1075 // TODO: Implement and remove this check.
1076 if (NInoSparse(ni
)) {
1078 ntfs_error(vi
->i_sb
, "Writing to sparse files "
1079 "is not supported yet. Sorry.");
1083 /* We have to zero every time due to mmap-at-end-of-file. */
1084 if (page
->index
>= (vi
->i_size
>> PAGE_CACHE_SHIFT
)) {
1085 /* The page straddles i_size. */
1086 unsigned int ofs
= vi
->i_size
& ~PAGE_CACHE_MASK
;
1087 kaddr
= kmap_atomic(page
, KM_USER0
);
1088 memset(kaddr
+ ofs
, 0, PAGE_CACHE_SIZE
- ofs
);
1089 flush_dcache_page(page
);
1090 kunmap_atomic(kaddr
, KM_USER0
);
1092 /* Handle mst protected attributes. */
1093 if (NInoMstProtected(ni
))
1094 return ntfs_write_mst_block(wbc
, page
);
1095 /* Normal data stream. */
1096 return ntfs_write_block(wbc
, page
);
1100 * Attribute is resident, implying it is not compressed, encrypted, or
1103 BUG_ON(page_has_buffers(page
));
1104 BUG_ON(!PageUptodate(page
));
1109 base_ni
= ni
->ext
.base_ntfs_ino
;
1111 /* Map, pin, and lock the mft record. */
1112 m
= map_mft_record(base_ni
);
1119 ctx
= ntfs_attr_get_search_ctx(base_ni
, m
);
1120 if (unlikely(!ctx
)) {
1124 err
= ntfs_attr_lookup(ni
->type
, ni
->name
, ni
->name_len
,
1125 CASE_SENSITIVE
, 0, NULL
, 0, ctx
);
1129 /* Starting position of the page within the attribute value. */
1130 attr_pos
= page
->index
<< PAGE_CACHE_SHIFT
;
1132 /* The total length of the attribute value. */
1133 attr_len
= le32_to_cpu(ctx
->attr
->data
.resident
.value_length
);
1135 if (unlikely(vi
->i_size
!= attr_len
)) {
1136 ntfs_error(vi
->i_sb
, "BUG()! i_size (0x%llx) doesn't match "
1137 "attr_len (0x%x). Aborting write.", vi
->i_size
,
1142 if (unlikely(attr_pos
>= attr_len
)) {
1143 ntfs_error(vi
->i_sb
, "BUG()! attr_pos (0x%llx) > attr_len "
1144 "(0x%x). Aborting write.",
1145 (unsigned long long)attr_pos
, attr_len
);
1150 bytes
= attr_len
- attr_pos
;
1151 if (unlikely(bytes
> PAGE_CACHE_SIZE
))
1152 bytes
= PAGE_CACHE_SIZE
;
1155 * Keep the VM happy. This must be done otherwise the radix-tree tag
1156 * PAGECACHE_TAG_DIRTY remains set even though the page is clean.
1158 BUG_ON(PageWriteback(page
));
1159 set_page_writeback(page
);
1163 * Here, we don't need to zero the out of bounds area everytime because
1164 * the below memcpy() already takes care of the mmap-at-end-of-file
1165 * requirements. If the file is converted to a non-resident one, then
1166 * the code path use is switched to the non-resident one where the
1167 * zeroing happens on each ntfs_writepage() invocation.
1169 * The above also applies nicely when i_size is decreased.
1171 * When i_size is increased, the memory between the old and new i_size
1172 * _must_ be zeroed (or overwritten with new data). Otherwise we will
1173 * expose data to userspace/disk which should never have been exposed.
1175 * FIXME: Ensure that i_size increases do the zeroing/overwriting and
1176 * if we cannot guarantee that, then enable the zeroing below. If the
1177 * zeroing below is enabled, we MUST move the unlock_page() from above
1178 * to after the kunmap_atomic(), i.e. just before the
1179 * end_page_writeback().
1182 kaddr
= kmap_atomic(page
, KM_USER0
);
1183 /* Copy the data from the page to the mft record. */
1184 memcpy((u8
*)ctx
->attr
+ le16_to_cpu(
1185 ctx
->attr
->data
.resident
.value_offset
) + attr_pos
,
1187 flush_dcache_mft_record_page(ctx
->ntfs_ino
);
1189 /* Zero out of bounds area. */
1190 if (likely(bytes
< PAGE_CACHE_SIZE
)) {
1191 memset(kaddr
+ bytes
, 0, PAGE_CACHE_SIZE
- bytes
);
1192 flush_dcache_page(page
);
1195 kunmap_atomic(kaddr
, KM_USER0
);
1197 end_page_writeback(page
);
1199 /* Mark the mft record dirty, so it gets written back. */
1200 mark_mft_record_dirty(ctx
->ntfs_ino
);
1202 ntfs_attr_put_search_ctx(ctx
);
1203 unmap_mft_record(base_ni
);
1206 if (err
== -ENOMEM
) {
1207 ntfs_warning(vi
->i_sb
, "Error allocating memory. Redirtying "
1208 "page so we try again later.");
1210 * Put the page back on mapping->dirty_pages, but leave its
1211 * buffer's dirty state as-is.
1213 redirty_page_for_writepage(wbc
, page
);
1216 ntfs_error(vi
->i_sb
, "Resident attribute write failed with "
1217 "error %i. Setting page error flag.", -err
);
1222 ntfs_attr_put_search_ctx(ctx
);
1224 unmap_mft_record(base_ni
);
1229 * ntfs_prepare_nonresident_write -
1232 static int ntfs_prepare_nonresident_write(struct page
*page
,
1233 unsigned from
, unsigned to
)
1237 sector_t block
, ablock
, iblock
;
1241 runlist_element
*rl
;
1242 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1243 unsigned int vcn_ofs
, block_start
, block_end
, blocksize
;
1246 unsigned char blocksize_bits
;
1248 vi
= page
->mapping
->host
;
1252 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1253 "0x%lx, from = %u, to = %u.", vi
->i_ino
, ni
->type
,
1254 page
->index
, from
, to
);
1256 BUG_ON(!NInoNonResident(ni
));
1257 BUG_ON(NInoMstProtected(ni
));
1259 blocksize_bits
= vi
->i_blkbits
;
1260 blocksize
= 1 << blocksize_bits
;
1263 * create_empty_buffers() will create uptodate/dirty buffers if the
1264 * page is uptodate/dirty.
1266 if (!page_has_buffers(page
))
1267 create_empty_buffers(page
, blocksize
, 0);
1268 bh
= head
= page_buffers(page
);
1272 /* The first block in the page. */
1273 block
= (s64
)page
->index
<< (PAGE_CACHE_SHIFT
- blocksize_bits
);
1276 * The first out of bounds block for the allocated size. No need to
1277 * round up as allocated_size is in multiples of cluster size and the
1278 * minimum cluster size is 512 bytes, which is equal to the smallest
1281 ablock
= ni
->allocated_size
>> blocksize_bits
;
1283 /* The last (fully or partially) initialized block. */
1284 iblock
= ni
->initialized_size
>> blocksize_bits
;
1286 /* Loop through all the buffers in the page. */
1291 block_end
= block_start
+ blocksize
;
1293 * If buffer @bh is outside the write, just mark it uptodate
1294 * if the page is uptodate and continue with the next buffer.
1296 if (block_end
<= from
|| block_start
>= to
) {
1297 if (PageUptodate(page
)) {
1298 if (!buffer_uptodate(bh
))
1299 set_buffer_uptodate(bh
);
1304 * @bh is at least partially being written to.
1305 * Make sure it is not marked as new.
1307 //if (buffer_new(bh))
1308 // clear_buffer_new(bh);
1310 if (block
>= ablock
) {
1311 // TODO: block is above allocated_size, need to
1312 // allocate it. Best done in one go to accommodate not
1313 // only block but all above blocks up to and including:
1314 // ((page->index << PAGE_CACHE_SHIFT) + to + blocksize
1315 // - 1) >> blobksize_bits. Obviously will need to round
1316 // up to next cluster boundary, too. This should be
1317 // done with a helper function, so it can be reused.
1318 ntfs_error(vol
->sb
, "Writing beyond allocated size "
1319 "is not supported yet. Sorry.");
1322 // Need to update ablock.
1323 // Need to set_buffer_new() on all block bhs that are
1327 * Now we have enough allocated size to fulfill the whole
1328 * request, i.e. block < ablock is true.
1330 if (unlikely((block
>= iblock
) &&
1331 (ni
->initialized_size
< vi
->i_size
))) {
1333 * If this page is fully outside initialized size, zero
1334 * out all pages between the current initialized size
1335 * and the current page. Just use ntfs_readpage() to do
1336 * the zeroing transparently.
1338 if (block
> iblock
) {
1340 // For each page do:
1341 // - read_cache_page()
1342 // Again for each page do:
1343 // - wait_on_page_locked()
1344 // - Check (PageUptodate(page) &&
1345 // !PageError(page))
1346 // Update initialized size in the attribute and
1348 // Again, for each page do:
1349 // __set_page_dirty_buffers();
1350 // page_cache_release()
1351 // We don't need to wait on the writes.
1355 * The current page straddles initialized size. Zero
1356 * all non-uptodate buffers and set them uptodate (and
1357 * dirty?). Note, there aren't any non-uptodate buffers
1358 * if the page is uptodate.
1359 * FIXME: For an uptodate page, the buffers may need to
1360 * be written out because they were not initialized on
1363 if (!PageUptodate(page
)) {
1365 // Zero any non-uptodate buffers up to i_size.
1366 // Set them uptodate and dirty.
1369 // Update initialized size in the attribute and in the
1370 // inode (up to i_size).
1372 // FIXME: This is inefficient. Try to batch the two
1373 // size changes to happen in one go.
1374 ntfs_error(vol
->sb
, "Writing beyond initialized size "
1375 "is not supported yet. Sorry.");
1378 // Do NOT set_buffer_new() BUT DO clear buffer range
1379 // outside write request range.
1380 // set_buffer_uptodate() on complete buffers as well as
1381 // set_buffer_dirty().
1384 /* Need to map unmapped buffers. */
1385 if (!buffer_mapped(bh
)) {
1386 /* Unmapped buffer. Need to map it. */
1387 bh
->b_bdev
= vol
->sb
->s_bdev
;
1389 /* Convert block into corresponding vcn and offset. */
1390 vcn
= (VCN
)block
<< blocksize_bits
>>
1391 vol
->cluster_size_bits
;
1392 vcn_ofs
= ((VCN
)block
<< blocksize_bits
) &
1393 vol
->cluster_size_mask
;
1398 down_read(&ni
->runlist
.lock
);
1399 rl
= ni
->runlist
.rl
;
1401 if (likely(rl
!= NULL
)) {
1402 /* Seek to element containing target vcn. */
1403 while (rl
->length
&& rl
[1].vcn
<= vcn
)
1405 lcn
= ntfs_vcn_to_lcn(rl
, vcn
);
1407 lcn
= (LCN
)LCN_RL_NOT_MAPPED
;
1408 if (unlikely(lcn
< 0)) {
1410 * We extended the attribute allocation above.
1411 * If we hit an ENOENT here it means that the
1412 * allocation was insufficient which is a bug.
1414 BUG_ON(lcn
== LCN_ENOENT
);
1416 /* It is a hole, need to instantiate it. */
1417 if (lcn
== LCN_HOLE
) {
1418 // TODO: Instantiate the hole.
1419 // clear_buffer_new(bh);
1420 // unmap_underlying_metadata(bh->b_bdev,
1422 // For non-uptodate buffers, need to
1423 // zero out the region outside the
1424 // request in this bh or all bhs,
1425 // depending on what we implemented
1427 // Need to flush_dcache_page().
1428 // Or could use set_buffer_new()
1430 ntfs_error(vol
->sb
, "Writing into "
1431 "sparse regions is "
1432 "not supported yet. "
1436 } else if (!is_retry
&&
1437 lcn
== LCN_RL_NOT_MAPPED
) {
1440 * Attempt to map runlist, dropping
1441 * lock for the duration.
1443 up_read(&ni
->runlist
.lock
);
1444 err
= ntfs_map_runlist(ni
, vcn
);
1446 goto lock_retry_remap
;
1450 * Failed to map the buffer, even after
1453 bh
->b_blocknr
= -1UL;
1454 ntfs_error(vol
->sb
, "ntfs_vcn_to_lcn(vcn = "
1455 "0x%llx) failed with error "
1457 (unsigned long long)vcn
,
1458 (unsigned long long)-lcn
,
1459 is_retry
? " even after "
1461 // FIXME: Depending on vol->on_errors, do
1467 /* We now have a successful remap, i.e. lcn >= 0. */
1469 /* Setup buffer head to correct block. */
1470 bh
->b_blocknr
= ((lcn
<< vol
->cluster_size_bits
)
1471 + vcn_ofs
) >> blocksize_bits
;
1472 set_buffer_mapped(bh
);
1474 // FIXME: Something analogous to this is needed for
1475 // each newly allocated block, i.e. BH_New.
1476 // FIXME: Might need to take this out of the
1477 // if (!buffer_mapped(bh)) {}, depending on how we
1478 // implement things during the allocated_size and
1479 // initialized_size extension code above.
1480 if (buffer_new(bh
)) {
1481 clear_buffer_new(bh
);
1482 unmap_underlying_metadata(bh
->b_bdev
,
1484 if (PageUptodate(page
)) {
1485 set_buffer_uptodate(bh
);
1489 * Page is _not_ uptodate, zero surrounding
1490 * region. NOTE: This is how we decide if to
1493 if (block_end
> to
|| block_start
< from
) {
1496 kaddr
= kmap_atomic(page
, KM_USER0
);
1498 memset(kaddr
+ to
, 0,
1500 if (block_start
< from
)
1501 memset(kaddr
+ block_start
, 0,
1504 flush_dcache_page(page
);
1505 kunmap_atomic(kaddr
, KM_USER0
);
1510 /* @bh is mapped, set it uptodate if the page is uptodate. */
1511 if (PageUptodate(page
)) {
1512 if (!buffer_uptodate(bh
))
1513 set_buffer_uptodate(bh
);
1517 * The page is not uptodate. The buffer is mapped. If it is not
1518 * uptodate, and it is only partially being written to, we need
1519 * to read the buffer in before the write, i.e. right now.
1521 if (!buffer_uptodate(bh
) &&
1522 (block_start
< from
|| block_end
> to
)) {
1523 ll_rw_block(READ
, 1, &bh
);
1526 } while (block
++, block_start
= block_end
,
1527 (bh
= bh
->b_this_page
) != head
);
1529 /* Release the lock if we took it. */
1531 up_read(&ni
->runlist
.lock
);
1535 /* If we issued read requests, let them complete. */
1536 while (wait_bh
> wait
) {
1537 wait_on_buffer(*--wait_bh
);
1538 if (!buffer_uptodate(*wait_bh
))
1542 ntfs_debug("Done.");
1546 * Zero out any newly allocated blocks to avoid exposing stale data.
1547 * If BH_New is set, we know that the block was newly allocated in the
1549 * FIXME: What about initialized_size increments? Have we done all the
1550 * required zeroing above? If not this error handling is broken, and
1551 * in particular the if (block_end <= from) check is completely bogus.
1557 block_end
= block_start
+ blocksize
;
1558 if (block_end
<= from
)
1560 if (block_start
>= to
)
1562 if (buffer_new(bh
)) {
1565 clear_buffer_new(bh
);
1566 kaddr
= kmap_atomic(page
, KM_USER0
);
1567 memset(kaddr
+ block_start
, 0, bh
->b_size
);
1568 kunmap_atomic(kaddr
, KM_USER0
);
1569 set_buffer_uptodate(bh
);
1570 mark_buffer_dirty(bh
);
1573 } while (block_start
= block_end
, (bh
= bh
->b_this_page
) != head
);
1575 flush_dcache_page(page
);
1577 up_read(&ni
->runlist
.lock
);
1582 * ntfs_prepare_write - prepare a page for receiving data
1584 * This is called from generic_file_write() with i_sem held on the inode
1585 * (@page->mapping->host). The @page is locked and kmap()ped so page_address()
1586 * can simply be used. The source data has not yet been copied into the @page.
1588 * Need to extend the attribute/fill in holes if necessary, create blocks and
1589 * make partially overwritten blocks uptodate,
1591 * i_size is not to be modified yet.
1593 * Return 0 on success or -errno on error.
1595 * Should be using block_prepare_write() [support for sparse files] or
1596 * cont_prepare_write() [no support for sparse files]. Can't do that due to
1597 * ntfs specifics but can look at them for implementation guidancea.
1599 * Note: In the range, @from is inclusive and @to is exclusive, i.e. @from is
1600 * the first byte in the page that will be written to and @to is the first byte
1601 * after the last byte that will be written to.
1603 static int ntfs_prepare_write(struct file
*file
, struct page
*page
,
1604 unsigned from
, unsigned to
)
1606 struct inode
*vi
= page
->mapping
->host
;
1607 ntfs_inode
*ni
= NTFS_I(vi
);
1609 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1610 "0x%lx, from = %u, to = %u.", vi
->i_ino
, ni
->type
,
1611 page
->index
, from
, to
);
1613 BUG_ON(!PageLocked(page
));
1614 BUG_ON(from
> PAGE_CACHE_SIZE
);
1615 BUG_ON(to
> PAGE_CACHE_SIZE
);
1618 if (NInoNonResident(ni
)) {
1620 * Only unnamed $DATA attributes can be compressed, encrypted,
1623 if (ni
->type
== AT_DATA
&& !ni
->name_len
) {
1624 /* If file is encrypted, deny access, just like NT4. */
1625 if (NInoEncrypted(ni
)) {
1626 ntfs_debug("Denying write access to encrypted "
1630 /* Compressed data streams are handled in compress.c. */
1631 if (NInoCompressed(ni
)) {
1632 // TODO: Implement and replace this check with
1633 // return ntfs_write_compressed_block(page);
1634 ntfs_error(vi
->i_sb
, "Writing to compressed "
1635 "files is not supported yet. "
1639 // TODO: Implement and remove this check.
1640 if (NInoSparse(ni
)) {
1641 ntfs_error(vi
->i_sb
, "Writing to sparse files "
1642 "is not supported yet. Sorry.");
1647 // TODO: Implement and remove this check.
1648 if (NInoMstProtected(ni
)) {
1649 ntfs_error(vi
->i_sb
, "Writing to MST protected "
1650 "attributes is not supported yet. "
1655 /* Normal data stream. */
1656 return ntfs_prepare_nonresident_write(page
, from
, to
);
1660 * Attribute is resident, implying it is not compressed, encrypted, or
1663 BUG_ON(page_has_buffers(page
));
1665 /* Do we need to resize the attribute? */
1666 if (((s64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
> vi
->i_size
) {
1667 // TODO: Implement resize...
1668 ntfs_error(vi
->i_sb
, "Writing beyond the existing file size is "
1669 "not supported yet. Sorry.");
1674 * Because resident attributes are handled by memcpy() to/from the
1675 * corresponding MFT record, and because this form of i/o is byte
1676 * aligned rather than block aligned, there is no need to bring the
1677 * page uptodate here as in the non-resident case where we need to
1678 * bring the buffers straddled by the write uptodate before
1679 * generic_file_write() does the copying from userspace.
1681 * We thus defer the uptodate bringing of the page region outside the
1682 * region written to to ntfs_commit_write(). The reason for doing this
1683 * is that we save one round of:
1684 * map_mft_record(), ntfs_attr_get_search_ctx(),
1685 * ntfs_attr_lookup(), kmap_atomic(), kunmap_atomic(),
1686 * ntfs_attr_put_search_ctx(), unmap_mft_record().
1687 * Which is obviously a very worthwhile save.
1689 * Thus we just return success now...
1691 ntfs_debug("Done.");
1696 * NOTES: There is a disparity between the apparent need to extend the
1697 * attribute in prepare write but to update i_size only in commit write.
1698 * Need to make sure i_sem protection is sufficient. And if not will need to
1699 * handle this in some way or another.
1703 * ntfs_commit_nonresident_write -
1706 static int ntfs_commit_nonresident_write(struct page
*page
,
1707 unsigned from
, unsigned to
)
1709 s64 pos
= ((s64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
1711 struct buffer_head
*bh
, *head
;
1712 unsigned int block_start
, block_end
, blocksize
;
1715 vi
= page
->mapping
->host
;
1717 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1718 "0x%lx, from = %u, to = %u.", vi
->i_ino
,
1719 NTFS_I(vi
)->type
, page
->index
, from
, to
);
1721 blocksize
= 1 << vi
->i_blkbits
;
1723 // FIXME: We need a whole slew of special cases in here for MST
1724 // protected attributes for example. For compressed files, too...
1725 // For now, we know ntfs_prepare_write() would have failed so we can't
1726 // get here in any of the cases which we have to special case, so we
1727 // are just a ripped off unrolled generic_commit_write() at present.
1729 bh
= head
= page_buffers(page
);
1733 block_end
= block_start
+ blocksize
;
1734 if (block_end
<= from
|| block_start
>= to
) {
1735 if (!buffer_uptodate(bh
))
1738 set_buffer_uptodate(bh
);
1739 mark_buffer_dirty(bh
);
1741 } while (block_start
= block_end
, (bh
= bh
->b_this_page
) != head
);
1744 * If this is a partial write which happened to make all buffers
1745 * uptodate then we can optimize away a bogus ->readpage() for the next
1746 * read(). Here we 'discover' whether the page went uptodate as a
1747 * result of this (potentially partial) write.
1750 SetPageUptodate(page
);
1753 * Not convinced about this at all. See disparity comment above. For
1754 * now we know ntfs_prepare_write() would have failed in the write
1755 * exceeds i_size case, so this will never trigger which is fine.
1757 if (pos
> vi
->i_size
) {
1758 ntfs_error(vi
->i_sb
, "Writing beyond the existing file size is "
1759 "not supported yet. Sorry.");
1761 // vi->i_size = pos;
1762 // mark_inode_dirty(vi);
1764 ntfs_debug("Done.");
1769 * ntfs_commit_write - commit the received data
1771 * This is called from generic_file_write() with i_sem held on the inode
1772 * (@page->mapping->host). The @page is locked and kmap()ped so page_address()
1773 * can simply be used. The source data has already been copied into the @page.
1775 * Need to mark modified blocks dirty so they get written out later when
1776 * ntfs_writepage() is invoked by the VM.
1778 * Return 0 on success or -errno on error.
1780 * Should be using generic_commit_write(). This marks buffers uptodate and
1781 * dirty, sets the page uptodate if all buffers in the page are uptodate, and
1782 * updates i_size if the end of io is beyond i_size. In that case, it also
1783 * marks the inode dirty. - We could still use this (obviously except for
1784 * NInoMstProtected() attributes, where we will need to duplicate the core code
1785 * because we need our own async_io completion handler) but we could just do
1786 * the i_size update in prepare write, when we resize the attribute. Then
1787 * we would avoid the i_size update and mark_inode_dirty() happening here.
1789 * Can't use generic_commit_write() due to ntfs specialities but can look at
1790 * it for implementation guidance.
1792 * If things have gone as outlined in ntfs_prepare_write(), then we do not
1793 * need to do any page content modifications here at all, except in the write
1794 * to resident attribute case, where we need to do the uptodate bringing here
1795 * which we combine with the copying into the mft record which means we only
1796 * need to map the mft record and find the attribute record in it only once.
1798 static int ntfs_commit_write(struct file
*file
, struct page
*page
,
1799 unsigned from
, unsigned to
)
1803 ntfs_inode
*ni
, *base_ni
;
1804 char *kaddr
, *kattr
;
1805 ntfs_attr_search_ctx
*ctx
;
1807 u32 attr_len
, bytes
;
1810 vi
= page
->mapping
->host
;
1813 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1814 "0x%lx, from = %u, to = %u.", vi
->i_ino
, ni
->type
,
1815 page
->index
, from
, to
);
1817 if (NInoNonResident(ni
)) {
1819 * Only unnamed $DATA attributes can be compressed, encrypted,
1822 if (ni
->type
== AT_DATA
&& !ni
->name_len
) {
1823 /* If file is encrypted, deny access, just like NT4. */
1824 if (NInoEncrypted(ni
)) {
1825 // Should never get here!
1826 ntfs_debug("Denying write access to encrypted "
1830 /* Compressed data streams are handled in compress.c. */
1831 if (NInoCompressed(ni
)) {
1832 // TODO: Implement and replace this check with
1833 // return ntfs_write_compressed_block(page);
1834 // Should never get here!
1835 ntfs_error(vi
->i_sb
, "Writing to compressed "
1836 "files is not supported yet. "
1840 // TODO: Implement and remove this check.
1841 if (NInoSparse(ni
)) {
1842 // Should never get here!
1843 ntfs_error(vi
->i_sb
, "Writing to sparse files "
1844 "is not supported yet. Sorry.");
1849 // TODO: Implement and remove this check.
1850 if (NInoMstProtected(ni
)) {
1851 // Should never get here!
1852 ntfs_error(vi
->i_sb
, "Writing to MST protected "
1853 "attributes is not supported yet. "
1858 /* Normal data stream. */
1859 return ntfs_commit_nonresident_write(page
, from
, to
);
1863 * Attribute is resident, implying it is not compressed, encrypted, or
1867 /* Do we need to resize the attribute? */
1868 if (((s64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
> vi
->i_size
) {
1869 // TODO: Implement resize...
1870 // pos = ((s64)page->index << PAGE_CACHE_SHIFT) + to;
1871 // vi->i_size = pos;
1872 // mark_inode_dirty(vi);
1873 // Should never get here!
1874 ntfs_error(vi
->i_sb
, "Writing beyond the existing file size is "
1875 "not supported yet. Sorry.");
1882 base_ni
= ni
->ext
.base_ntfs_ino
;
1884 /* Map, pin, and lock the mft record. */
1885 m
= map_mft_record(base_ni
);
1892 ctx
= ntfs_attr_get_search_ctx(base_ni
, m
);
1893 if (unlikely(!ctx
)) {
1897 err
= ntfs_attr_lookup(ni
->type
, ni
->name
, ni
->name_len
,
1898 CASE_SENSITIVE
, 0, NULL
, 0, ctx
);
1902 /* Starting position of the page within the attribute value. */
1903 attr_pos
= page
->index
<< PAGE_CACHE_SHIFT
;
1905 /* The total length of the attribute value. */
1906 attr_len
= le32_to_cpu(ctx
->attr
->data
.resident
.value_length
);
1908 if (unlikely(vi
->i_size
!= attr_len
)) {
1909 ntfs_error(vi
->i_sb
, "BUG()! i_size (0x%llx) doesn't match "
1910 "attr_len (0x%x). Aborting write.", vi
->i_size
,
1915 if (unlikely(attr_pos
>= attr_len
)) {
1916 ntfs_error(vi
->i_sb
, "BUG()! attr_pos (0x%llx) > attr_len "
1917 "(0x%x). Aborting write.",
1918 (unsigned long long)attr_pos
, attr_len
);
1923 bytes
= attr_len
- attr_pos
;
1924 if (unlikely(bytes
> PAGE_CACHE_SIZE
))
1925 bytes
= PAGE_CACHE_SIZE
;
1928 * Calculate the address of the attribute value corresponding to the
1929 * beginning of the current data @page.
1931 kattr
= (u8
*)ctx
->attr
+ le16_to_cpu(
1932 ctx
->attr
->data
.resident
.value_offset
) + attr_pos
;
1934 kaddr
= kmap_atomic(page
, KM_USER0
);
1936 /* Copy the received data from the page to the mft record. */
1937 memcpy(kattr
+ from
, kaddr
+ from
, to
- from
);
1938 flush_dcache_mft_record_page(ctx
->ntfs_ino
);
1940 if (!PageUptodate(page
)) {
1942 * Bring the out of bounds area(s) uptodate by copying data
1943 * from the mft record to the page.
1946 memcpy(kaddr
, kattr
, from
);
1948 memcpy(kaddr
+ to
, kattr
+ to
, bytes
- to
);
1950 /* Zero the region outside the end of the attribute value. */
1951 if (likely(bytes
< PAGE_CACHE_SIZE
))
1952 memset(kaddr
+ bytes
, 0, PAGE_CACHE_SIZE
- bytes
);
1955 * The probability of not having done any of the above is
1956 * extremely small, so we just flush unconditionally.
1958 flush_dcache_page(page
);
1959 SetPageUptodate(page
);
1961 kunmap_atomic(kaddr
, KM_USER0
);
1963 /* Mark the mft record dirty, so it gets written back. */
1964 mark_mft_record_dirty(ctx
->ntfs_ino
);
1966 ntfs_attr_put_search_ctx(ctx
);
1967 unmap_mft_record(base_ni
);
1968 ntfs_debug("Done.");
1971 if (err
== -ENOMEM
) {
1972 ntfs_warning(vi
->i_sb
, "Error allocating memory required to "
1973 "commit the write.");
1974 if (PageUptodate(page
)) {
1975 ntfs_warning(vi
->i_sb
, "Page is uptodate, setting "
1976 "dirty so the write will be retried "
1977 "later on by the VM.");
1979 * Put the page on mapping->dirty_pages, but leave its
1980 * buffer's dirty state as-is.
1982 __set_page_dirty_nobuffers(page
);
1985 ntfs_error(vi
->i_sb
, "Page is not uptodate. Written "
1986 "data has been lost. )-:");
1988 ntfs_error(vi
->i_sb
, "Resident attribute write failed with "
1989 "error %i. Setting page error flag.", -err
);
1993 ntfs_attr_put_search_ctx(ctx
);
1995 unmap_mft_record(base_ni
);
1999 #endif /* NTFS_RW */
2002 * ntfs_aops - general address space operations for inodes and attributes
2004 struct address_space_operations ntfs_aops
= {
2005 .readpage
= ntfs_readpage
, /* Fill page with data. */
2006 .sync_page
= block_sync_page
, /* Currently, just unplugs the
2007 disk request queue. */
2009 .writepage
= ntfs_writepage
, /* Write dirty page to disk. */
2010 .prepare_write
= ntfs_prepare_write
, /* Prepare page and buffers
2011 ready to receive data. */
2012 .commit_write
= ntfs_commit_write
, /* Commit received data. */
2013 #endif /* NTFS_RW */
2017 * ntfs_mst_aops - general address space operations for mst protecteed inodes
2020 struct address_space_operations ntfs_mst_aops
= {
2021 .readpage
= ntfs_readpage
, /* Fill page with data. */
2022 .sync_page
= block_sync_page
, /* Currently, just unplugs the
2023 disk request queue. */
2025 .writepage
= ntfs_writepage
, /* Write dirty page to disk. */
2026 .set_page_dirty
= __set_page_dirty_nobuffers
, /* Set the page dirty
2027 without touching the buffers
2028 belonging to the page. */
2029 #endif /* NTFS_RW */