2 * aops.c - NTFS kernel address space operations and page cache handling.
3 * Part of the Linux-NTFS project.
5 * Copyright (c) 2001-2005 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>
29 #include <linux/writeback.h>
41 * ntfs_end_buffer_async_read - async io completion for reading attributes
42 * @bh: buffer head on which io is completed
43 * @uptodate: whether @bh is now uptodate or not
45 * Asynchronous I/O completion handler for reading pages belonging to the
46 * attribute address space of an inode. The inodes can either be files or
47 * directories or they can be fake inodes describing some attribute.
49 * If NInoMstProtected(), perform the post read mst fixups when all IO on the
50 * page has been completed and mark the page uptodate or set the error bit on
51 * the page. To determine the size of the records that need fixing up, we
52 * cheat a little bit by setting the index_block_size in ntfs_inode to the ntfs
53 * record size, and index_block_size_bits, to the log(base 2) of the ntfs
56 static void ntfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
58 static DEFINE_SPINLOCK(page_uptodate_lock);
60 struct buffer_head *tmp;
63 int page_uptodate = 1;
66 ni = NTFS_I(page->mapping->host);
68 if (likely(uptodate)) {
69 s64 file_ofs, initialized_size;
71 set_buffer_uptodate(bh);
73 file_ofs = ((s64)page->index << PAGE_CACHE_SHIFT) +
75 read_lock_irqsave(&ni->size_lock, flags);
76 initialized_size = ni->initialized_size;
77 read_unlock_irqrestore(&ni->size_lock, flags);
78 /* Check for the current buffer head overflowing. */
79 if (file_ofs + bh->b_size > initialized_size) {
83 if (file_ofs < initialized_size)
84 ofs = initialized_size - file_ofs;
85 addr = kmap_atomic(page, KM_BIO_SRC_IRQ);
86 memset(addr + bh_offset(bh) + ofs, 0, bh->b_size - ofs);
87 flush_dcache_page(page);
88 kunmap_atomic(addr, KM_BIO_SRC_IRQ);
91 clear_buffer_uptodate(bh);
92 ntfs_error(ni->vol->sb, "Buffer I/O error, logical block %llu.",
93 (unsigned long long)bh->b_blocknr);
96 spin_lock_irqsave(&page_uptodate_lock, flags);
97 clear_buffer_async_read(bh);
101 if (!buffer_uptodate(tmp))
103 if (buffer_async_read(tmp)) {
104 if (likely(buffer_locked(tmp)))
106 /* Async buffers must be locked. */
109 tmp = tmp->b_this_page;
111 spin_unlock_irqrestore(&page_uptodate_lock, flags);
113 * If none of the buffers had errors then we can set the page uptodate,
114 * but we first have to perform the post read mst fixups, if the
115 * attribute is mst protected, i.e. if NInoMstProteced(ni) is true.
116 * Note we ignore fixup errors as those are detected when
117 * map_mft_record() is called which gives us per record granularity
118 * rather than per page granularity.
120 if (!NInoMstProtected(ni)) {
121 if (likely(page_uptodate && !PageError(page)))
122 SetPageUptodate(page);
125 unsigned int i, recs;
128 rec_size = ni->itype.index.block_size;
129 recs = PAGE_CACHE_SIZE / rec_size;
130 /* Should have been verified before we got here... */
132 addr = kmap_atomic(page, KM_BIO_SRC_IRQ);
133 for (i = 0; i < recs; i++)
134 post_read_mst_fixup((NTFS_RECORD*)(addr +
135 i * rec_size), rec_size);
136 flush_dcache_page(page);
137 kunmap_atomic(addr, KM_BIO_SRC_IRQ);
138 if (likely(page_uptodate && !PageError(page)))
139 SetPageUptodate(page);
144 spin_unlock_irqrestore(&page_uptodate_lock, flags);
149 * ntfs_read_block - fill a @page of an address space with data
150 * @page: page cache page to fill with data
152 * Fill the page @page of the address space belonging to the @page->host inode.
153 * We read each buffer asynchronously and when all buffers are read in, our io
154 * completion handler ntfs_end_buffer_read_async(), if required, automatically
155 * applies the mst fixups to the page before finally marking it uptodate and
158 * We only enforce allocated_size limit because i_size is checked for in
159 * generic_file_read().
161 * Return 0 on success and -errno on error.
163 * Contains an adapted version of fs/buffer.c::block_read_full_page().
165 static int ntfs_read_block(struct page *page)
172 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
173 sector_t iblock, lblock, zblock;
175 unsigned int blocksize, vcn_ofs;
177 unsigned char blocksize_bits;
179 ni = NTFS_I(page->mapping->host);
182 /* $MFT/$DATA must have its complete runlist in memory at all times. */
183 BUG_ON(!ni->runlist.rl && !ni->mft_no && !NInoAttr(ni));
185 blocksize_bits = VFS_I(ni)->i_blkbits;
186 blocksize = 1 << blocksize_bits;
188 if (!page_has_buffers(page))
189 create_empty_buffers(page, blocksize, 0);
190 bh = head = page_buffers(page);
196 iblock = (s64)page->index << (PAGE_CACHE_SHIFT - blocksize_bits);
197 read_lock_irqsave(&ni->size_lock, flags);
198 lblock = (ni->allocated_size + blocksize - 1) >> blocksize_bits;
199 zblock = (ni->initialized_size + blocksize - 1) >> blocksize_bits;
200 read_unlock_irqrestore(&ni->size_lock, flags);
202 /* Loop through all the buffers in the page. */
208 if (unlikely(buffer_uptodate(bh)))
210 if (unlikely(buffer_mapped(bh))) {
214 bh->b_bdev = vol->sb->s_bdev;
215 /* Is the block within the allowed limits? */
216 if (iblock < lblock) {
217 BOOL is_retry = FALSE;
219 /* Convert iblock into corresponding vcn and offset. */
220 vcn = (VCN)iblock << blocksize_bits >>
221 vol->cluster_size_bits;
222 vcn_ofs = ((VCN)iblock << blocksize_bits) &
223 vol->cluster_size_mask;
226 down_read(&ni->runlist.lock);
229 if (likely(rl != NULL)) {
230 /* Seek to element containing target vcn. */
231 while (rl->length && rl[1].vcn <= vcn)
233 lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
235 lcn = LCN_RL_NOT_MAPPED;
236 /* Successful remap. */
238 /* Setup buffer head to correct block. */
239 bh->b_blocknr = ((lcn << vol->cluster_size_bits)
240 + vcn_ofs) >> blocksize_bits;
241 set_buffer_mapped(bh);
242 /* Only read initialized data blocks. */
243 if (iblock < zblock) {
247 /* Fully non-initialized data block, zero it. */
250 /* It is a hole, need to zero it. */
253 /* If first try and runlist unmapped, map and retry. */
254 if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
258 * Attempt to map runlist, dropping lock for
261 up_read(&ni->runlist.lock);
262 err = ntfs_map_runlist(ni, vcn);
264 goto lock_retry_remap;
268 /* Hard error, zero out region. */
271 ntfs_error(vol->sb, "Failed to read from inode 0x%lx, "
272 "attribute type 0x%x, vcn 0x%llx, "
273 "offset 0x%x because its location on "
274 "disk could not be determined%s "
275 "(error code %lli).", ni->mft_no,
276 ni->type, (unsigned long long)vcn,
277 vcn_ofs, is_retry ? " even after "
278 "retrying" : "", (long long)lcn);
281 * Either iblock was outside lblock limits or
282 * ntfs_rl_vcn_to_lcn() returned error. Just zero that portion
283 * of the page and set the buffer uptodate.
286 bh->b_blocknr = -1UL;
287 clear_buffer_mapped(bh);
289 kaddr = kmap_atomic(page, KM_USER0);
290 memset(kaddr + i * blocksize, 0, blocksize);
291 flush_dcache_page(page);
292 kunmap_atomic(kaddr, KM_USER0);
293 set_buffer_uptodate(bh);
294 } while (i++, iblock++, (bh = bh->b_this_page) != head);
296 /* Release the lock if we took it. */
298 up_read(&ni->runlist.lock);
300 /* Check we have at least one buffer ready for i/o. */
302 struct buffer_head *tbh;
304 /* Lock the buffers. */
305 for (i = 0; i < nr; i++) {
308 tbh->b_end_io = ntfs_end_buffer_async_read;
309 set_buffer_async_read(tbh);
311 /* Finally, start i/o on the buffers. */
312 for (i = 0; i < nr; i++) {
314 if (likely(!buffer_uptodate(tbh)))
315 submit_bh(READ, tbh);
317 ntfs_end_buffer_async_read(tbh, 1);
321 /* No i/o was scheduled on any of the buffers. */
322 if (likely(!PageError(page)))
323 SetPageUptodate(page);
324 else /* Signal synchronous i/o error. */
331 * ntfs_readpage - fill a @page of a @file with data from the device
332 * @file: open file to which the page @page belongs or NULL
333 * @page: page cache page to fill with data
335 * For non-resident attributes, ntfs_readpage() fills the @page of the open
336 * file @file by calling the ntfs version of the generic block_read_full_page()
337 * function, ntfs_read_block(), which in turn creates and reads in the buffers
338 * associated with the page asynchronously.
340 * For resident attributes, OTOH, ntfs_readpage() fills @page by copying the
341 * data from the mft record (which at this stage is most likely in memory) and
342 * fills the remainder with zeroes. Thus, in this case, I/O is synchronous, as
343 * even if the mft record is not cached at this point in time, we need to wait
344 * for it to be read in before we can do the copy.
346 * Return 0 on success and -errno on error.
348 static int ntfs_readpage(struct file *file, struct page *page)
350 ntfs_inode *ni, *base_ni;
352 ntfs_attr_search_ctx *ctx;
358 BUG_ON(!PageLocked(page));
360 * This can potentially happen because we clear PageUptodate() during
361 * ntfs_writepage() of MstProtected() attributes.
363 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);
389 * Attribute is resident, implying it is not compressed or encrypted.
390 * This also means the attribute is smaller than an mft record and
391 * hence smaller than a page, so can simply zero out any pages with
394 if (unlikely(page->index > 0)) {
395 kaddr = kmap_atomic(page, KM_USER0);
396 memset(kaddr, 0, PAGE_CACHE_SIZE);
397 flush_dcache_page(page);
398 kunmap_atomic(kaddr, KM_USER0);
404 base_ni = ni->ext.base_ntfs_ino;
405 /* Map, pin, and lock the mft record. */
406 mrec = map_mft_record(base_ni);
411 ctx = ntfs_attr_get_search_ctx(base_ni, mrec);
412 if (unlikely(!ctx)) {
416 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
417 CASE_SENSITIVE, 0, NULL, 0, ctx);
419 goto put_unm_err_out;
420 attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
421 read_lock_irqsave(&ni->size_lock, flags);
422 if (unlikely(attr_len > ni->initialized_size))
423 attr_len = ni->initialized_size;
424 read_unlock_irqrestore(&ni->size_lock, flags);
425 kaddr = kmap_atomic(page, KM_USER0);
426 /* Copy the data to the page. */
427 memcpy(kaddr, (u8*)ctx->attr +
428 le16_to_cpu(ctx->attr->data.resident.value_offset),
430 /* Zero the remainder of the page. */
431 memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len);
432 flush_dcache_page(page);
433 kunmap_atomic(kaddr, KM_USER0);
435 ntfs_attr_put_search_ctx(ctx);
437 unmap_mft_record(base_ni);
439 SetPageUptodate(page);
448 * ntfs_write_block - write a @page to the backing store
449 * @page: page cache page to write out
450 * @wbc: writeback control structure
452 * This function is for writing pages belonging to non-resident, non-mst
453 * protected attributes to their backing store.
455 * For a page with buffers, map and write the dirty buffers asynchronously
456 * under page writeback. For a page without buffers, create buffers for the
457 * page, then proceed as above.
459 * If a page doesn't have buffers the page dirty state is definitive. If a page
460 * does have buffers, the page dirty state is just a hint, and the buffer dirty
461 * state is definitive. (A hint which has rules: dirty buffers against a clean
462 * page is illegal. Other combinations are legal and need to be handled. In
463 * particular a dirty page containing clean buffers for example.)
465 * Return 0 on success and -errno on error.
467 * Based on ntfs_read_block() and __block_write_full_page().
469 static int ntfs_write_block(struct page *page, struct writeback_control *wbc)
473 s64 initialized_size;
475 sector_t block, dblock, iblock;
480 struct buffer_head *bh, *head;
482 unsigned int blocksize, vcn_ofs;
484 BOOL need_end_writeback;
485 unsigned char blocksize_bits;
487 vi = page->mapping->host;
491 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
492 "0x%lx.", ni->mft_no, ni->type, page->index);
494 BUG_ON(!NInoNonResident(ni));
495 BUG_ON(NInoMstProtected(ni));
497 blocksize_bits = vi->i_blkbits;
498 blocksize = 1 << blocksize_bits;
500 if (!page_has_buffers(page)) {
501 BUG_ON(!PageUptodate(page));
502 create_empty_buffers(page, blocksize,
503 (1 << BH_Uptodate) | (1 << BH_Dirty));
505 bh = head = page_buffers(page);
507 ntfs_warning(vol->sb, "Error allocating page buffers. "
508 "Redirtying page so we try again later.");
510 * Put the page back on mapping->dirty_pages, but leave its
511 * buffer's dirty state as-is.
513 redirty_page_for_writepage(wbc, page);
518 /* NOTE: Different naming scheme to ntfs_read_block()! */
520 /* The first block in the page. */
521 block = (s64)page->index << (PAGE_CACHE_SHIFT - blocksize_bits);
523 read_lock_irqsave(&ni->size_lock, flags);
524 i_size = i_size_read(vi);
525 initialized_size = ni->initialized_size;
526 read_unlock_irqrestore(&ni->size_lock, flags);
528 /* The first out of bounds block for the data size. */
529 dblock = (i_size + blocksize - 1) >> blocksize_bits;
531 /* The last (fully or partially) initialized block. */
532 iblock = initialized_size >> blocksize_bits;
535 * Be very careful. We have no exclusion from __set_page_dirty_buffers
536 * here, and the (potentially unmapped) buffers may become dirty at
537 * any time. If a buffer becomes dirty here after we've inspected it
538 * then we just miss that fact, and the page stays dirty.
540 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
541 * handle that here by just cleaning them.
545 * Loop through all the buffers in the page, mapping all the dirty
546 * buffers to disk addresses and handling any aliases from the
547 * underlying block device's mapping.
552 BOOL is_retry = FALSE;
554 if (unlikely(block >= dblock)) {
556 * Mapped buffers outside i_size will occur, because
557 * this page can be outside i_size when there is a
558 * truncate in progress. The contents of such buffers
559 * were zeroed by ntfs_writepage().
561 * FIXME: What about the small race window where
562 * ntfs_writepage() has not done any clearing because
563 * the page was within i_size but before we get here,
564 * vmtruncate() modifies i_size?
566 clear_buffer_dirty(bh);
567 set_buffer_uptodate(bh);
571 /* Clean buffers are not written out, so no need to map them. */
572 if (!buffer_dirty(bh))
575 /* Make sure we have enough initialized size. */
576 if (unlikely((block >= iblock) &&
577 (initialized_size < i_size))) {
579 * If this page is fully outside initialized size, zero
580 * out all pages between the current initialized size
581 * and the current page. Just use ntfs_readpage() to do
582 * the zeroing transparently.
584 if (block > iblock) {
587 // - read_cache_page()
588 // Again for each page do:
589 // - wait_on_page_locked()
590 // - Check (PageUptodate(page) &&
592 // Update initialized size in the attribute and
594 // Again, for each page do:
595 // __set_page_dirty_buffers();
596 // page_cache_release()
597 // We don't need to wait on the writes.
601 * The current page straddles initialized size. Zero
602 * all non-uptodate buffers and set them uptodate (and
603 * dirty?). Note, there aren't any non-uptodate buffers
604 * if the page is uptodate.
605 * FIXME: For an uptodate page, the buffers may need to
606 * be written out because they were not initialized on
609 if (!PageUptodate(page)) {
611 // Zero any non-uptodate buffers up to i_size.
612 // Set them uptodate and dirty.
615 // Update initialized size in the attribute and in the
616 // inode (up to i_size).
618 // FIXME: This is inefficient. Try to batch the two
619 // size changes to happen in one go.
620 ntfs_error(vol->sb, "Writing beyond initialized size "
621 "is not supported yet. Sorry.");
624 // Do NOT set_buffer_new() BUT DO clear buffer range
625 // outside write request range.
626 // set_buffer_uptodate() on complete buffers as well as
627 // set_buffer_dirty().
630 /* No need to map buffers that are already mapped. */
631 if (buffer_mapped(bh))
634 /* Unmapped, dirty buffer. Need to map it. */
635 bh->b_bdev = vol->sb->s_bdev;
637 /* Convert block into corresponding vcn and offset. */
638 vcn = (VCN)block << blocksize_bits;
639 vcn_ofs = vcn & vol->cluster_size_mask;
640 vcn >>= vol->cluster_size_bits;
643 down_read(&ni->runlist.lock);
646 if (likely(rl != NULL)) {
647 /* Seek to element containing target vcn. */
648 while (rl->length && rl[1].vcn <= vcn)
650 lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
652 lcn = LCN_RL_NOT_MAPPED;
653 /* Successful remap. */
655 /* Setup buffer head to point to correct block. */
656 bh->b_blocknr = ((lcn << vol->cluster_size_bits) +
657 vcn_ofs) >> blocksize_bits;
658 set_buffer_mapped(bh);
661 /* It is a hole, need to instantiate it. */
662 if (lcn == LCN_HOLE) {
663 // TODO: Instantiate the hole.
664 // clear_buffer_new(bh);
665 // unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
666 ntfs_error(vol->sb, "Writing into sparse regions is "
667 "not supported yet. Sorry.");
671 /* If first try and runlist unmapped, map and retry. */
672 if (!is_retry && lcn == LCN_RL_NOT_MAPPED) {
675 * Attempt to map runlist, dropping lock for
678 up_read(&ni->runlist.lock);
679 err = ntfs_map_runlist(ni, vcn);
681 goto lock_retry_remap;
685 /* Failed to map the buffer, even after retrying. */
687 ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
688 "attribute type 0x%x, vcn 0x%llx, offset 0x%x "
689 "because its location on disk could not be "
690 "determined%s (error code %lli).", ni->mft_no,
691 ni->type, (unsigned long long)vcn,
692 vcn_ofs, is_retry ? " even after "
693 "retrying" : "", (long long)lcn);
697 } while (block++, (bh = bh->b_this_page) != head);
699 /* Release the lock if we took it. */
701 up_read(&ni->runlist.lock);
703 /* For the error case, need to reset bh to the beginning. */
706 /* Just an optimization, so ->readpage() isn't called later. */
707 if (unlikely(!PageUptodate(page))) {
710 if (!buffer_uptodate(bh)) {
715 } while ((bh = bh->b_this_page) != head);
717 SetPageUptodate(page);
720 /* Setup all mapped, dirty buffers for async write i/o. */
723 if (buffer_mapped(bh) && buffer_dirty(bh)) {
725 if (test_clear_buffer_dirty(bh)) {
726 BUG_ON(!buffer_uptodate(bh));
727 mark_buffer_async_write(bh);
730 } else if (unlikely(err)) {
732 * For the error case. The buffer may have been set
733 * dirty during attachment to a dirty page.
736 clear_buffer_dirty(bh);
738 } while ((bh = bh->b_this_page) != head);
741 // TODO: Remove the -EOPNOTSUPP check later on...
742 if (unlikely(err == -EOPNOTSUPP))
744 else if (err == -ENOMEM) {
745 ntfs_warning(vol->sb, "Error allocating memory. "
746 "Redirtying page so we try again "
749 * Put the page back on mapping->dirty_pages, but
750 * leave its buffer's dirty state as-is.
752 redirty_page_for_writepage(wbc, page);
758 BUG_ON(PageWriteback(page));
759 set_page_writeback(page); /* Keeps try_to_free_buffers() away. */
763 * Submit the prepared buffers for i/o. Note the page is unlocked,
764 * and the async write i/o completion handler can end_page_writeback()
765 * at any time after the *first* submit_bh(). So the buffers can then
768 need_end_writeback = TRUE;
770 struct buffer_head *next = bh->b_this_page;
771 if (buffer_async_write(bh)) {
772 submit_bh(WRITE, bh);
773 need_end_writeback = FALSE;
777 } while (bh != head);
779 /* If no i/o was started, need to end_page_writeback(). */
780 if (unlikely(need_end_writeback))
781 end_page_writeback(page);
788 * ntfs_write_mst_block - write a @page to the backing store
789 * @page: page cache page to write out
790 * @wbc: writeback control structure
792 * This function is for writing pages belonging to non-resident, mst protected
793 * attributes to their backing store. The only supported attributes are index
794 * allocation and $MFT/$DATA. Both directory inodes and index inodes are
795 * supported for the index allocation case.
797 * The page must remain locked for the duration of the write because we apply
798 * the mst fixups, write, and then undo the fixups, so if we were to unlock the
799 * page before undoing the fixups, any other user of the page will see the
800 * page contents as corrupt.
802 * We clear the page uptodate flag for the duration of the function to ensure
803 * exclusion for the $MFT/$DATA case against someone mapping an mft record we
804 * are about to apply the mst fixups to.
806 * Return 0 on success and -errno on error.
808 * Based on ntfs_write_block(), ntfs_mft_writepage(), and
809 * write_mft_record_nolock().
811 static int ntfs_write_mst_block(struct page *page,
812 struct writeback_control *wbc)
814 sector_t block, dblock, rec_block;
815 struct inode *vi = page->mapping->host;
816 ntfs_inode *ni = NTFS_I(vi);
817 ntfs_volume *vol = ni->vol;
819 unsigned char bh_size_bits = vi->i_blkbits;
820 unsigned int bh_size = 1 << bh_size_bits;
821 unsigned int rec_size = ni->itype.index.block_size;
822 ntfs_inode *locked_nis[PAGE_CACHE_SIZE / rec_size];
823 struct buffer_head *bh, *head, *tbh, *rec_start_bh;
824 int max_bhs = PAGE_CACHE_SIZE / bh_size;
825 struct buffer_head *bhs[max_bhs];
827 int i, nr_locked_nis, nr_recs, nr_bhs, bhs_per_rec, err, err2;
828 unsigned rec_size_bits;
829 BOOL sync, is_mft, page_is_dirty, rec_is_dirty;
831 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
832 "0x%lx.", vi->i_ino, ni->type, page->index);
833 BUG_ON(!NInoNonResident(ni));
834 BUG_ON(!NInoMstProtected(ni));
835 is_mft = (S_ISREG(vi->i_mode) && !vi->i_ino);
837 * NOTE: ntfs_write_mst_block() would be called for $MFTMirr if a page
838 * in its page cache were to be marked dirty. However this should
839 * never happen with the current driver and considering we do not
840 * handle this case here we do want to BUG(), at least for now.
842 BUG_ON(!(is_mft || S_ISDIR(vi->i_mode) ||
843 (NInoAttr(ni) && ni->type == AT_INDEX_ALLOCATION)));
846 /* Were we called for sync purposes? */
847 sync = (wbc->sync_mode == WB_SYNC_ALL);
849 /* Make sure we have mapped buffers. */
850 BUG_ON(!page_has_buffers(page));
851 bh = head = page_buffers(page);
854 rec_size_bits = ni->itype.index.block_size_bits;
855 BUG_ON(!(PAGE_CACHE_SIZE >> rec_size_bits));
856 bhs_per_rec = rec_size >> bh_size_bits;
857 BUG_ON(!bhs_per_rec);
859 /* The first block in the page. */
860 rec_block = block = (sector_t)page->index <<
861 (PAGE_CACHE_SHIFT - bh_size_bits);
863 /* The first out of bounds block for the data size. */
864 dblock = (i_size_read(vi) + bh_size - 1) >> bh_size_bits;
867 err = err2 = nr_bhs = nr_recs = nr_locked_nis = 0;
868 page_is_dirty = rec_is_dirty = FALSE;
871 BOOL is_retry = FALSE;
873 if (likely(block < rec_block)) {
874 if (unlikely(block >= dblock)) {
875 clear_buffer_dirty(bh);
876 set_buffer_uptodate(bh);
880 * This block is not the first one in the record. We
881 * ignore the buffer's dirty state because we could
882 * have raced with a parallel mark_ntfs_record_dirty().
886 if (unlikely(err2)) {
888 clear_buffer_dirty(bh);
891 } else /* if (block == rec_block) */ {
892 BUG_ON(block > rec_block);
893 /* This block is the first one in the record. */
894 rec_block += bhs_per_rec;
896 if (unlikely(block >= dblock)) {
897 clear_buffer_dirty(bh);
900 if (!buffer_dirty(bh)) {
901 /* Clean records are not written out. */
902 rec_is_dirty = FALSE;
908 /* Need to map the buffer if it is not mapped already. */
909 if (unlikely(!buffer_mapped(bh))) {
912 unsigned int vcn_ofs;
914 /* Obtain the vcn and offset of the current block. */
915 vcn = (VCN)block << bh_size_bits;
916 vcn_ofs = vcn & vol->cluster_size_mask;
917 vcn >>= vol->cluster_size_bits;
920 down_read(&ni->runlist.lock);
923 if (likely(rl != NULL)) {
924 /* Seek to element containing target vcn. */
925 while (rl->length && rl[1].vcn <= vcn)
927 lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
929 lcn = LCN_RL_NOT_MAPPED;
930 /* Successful remap. */
931 if (likely(lcn >= 0)) {
932 /* Setup buffer head to correct block. */
933 bh->b_blocknr = ((lcn <<
934 vol->cluster_size_bits) +
935 vcn_ofs) >> bh_size_bits;
936 set_buffer_mapped(bh);
939 * Remap failed. Retry to map the runlist once
940 * unless we are working on $MFT which always
941 * has the whole of its runlist in memory.
943 if (!is_mft && !is_retry &&
944 lcn == LCN_RL_NOT_MAPPED) {
947 * Attempt to map runlist, dropping
948 * lock for the duration.
950 up_read(&ni->runlist.lock);
951 err2 = ntfs_map_runlist(ni, vcn);
953 goto lock_retry_remap;
955 page_is_dirty = TRUE;
959 /* Hard error. Abort writing this record. */
960 if (!err || err == -ENOMEM)
963 ntfs_error(vol->sb, "Cannot write ntfs record "
964 "0x%llx (inode 0x%lx, "
965 "attribute type 0x%x) because "
966 "its location on disk could "
967 "not be determined (error "
968 "code %lli).", (s64)block <<
970 vol->mft_record_size_bits,
971 ni->mft_no, ni->type,
974 * If this is not the first buffer, remove the
975 * buffers in this record from the list of
976 * buffers to write and clear their dirty bit
977 * if not error -ENOMEM.
979 if (rec_start_bh != bh) {
980 while (bhs[--nr_bhs] != rec_start_bh)
982 if (err2 != -ENOMEM) {
986 } while ((rec_start_bh =
995 BUG_ON(!buffer_uptodate(bh));
996 BUG_ON(nr_bhs >= max_bhs);
998 } while (block++, (bh = bh->b_this_page) != head);
1000 up_read(&ni->runlist.lock);
1001 /* If there were no dirty buffers, we are done. */
1004 /* Map the page so we can access its contents. */
1006 /* Clear the page uptodate flag whilst the mst fixups are applied. */
1007 BUG_ON(!PageUptodate(page));
1008 ClearPageUptodate(page);
1009 for (i = 0; i < nr_bhs; i++) {
1012 /* Skip buffers which are not at the beginning of records. */
1013 if (i % bhs_per_rec)
1016 ofs = bh_offset(tbh);
1019 unsigned long mft_no;
1021 /* Get the mft record number. */
1022 mft_no = (((s64)page->index << PAGE_CACHE_SHIFT) + ofs)
1024 /* Check whether to write this mft record. */
1026 if (!ntfs_may_write_mft_record(vol, mft_no,
1027 (MFT_RECORD*)(kaddr + ofs), &tni)) {
1029 * The record should not be written. This
1030 * means we need to redirty the page before
1033 page_is_dirty = TRUE;
1035 * Remove the buffers in this mft record from
1036 * the list of buffers to write.
1040 } while (++i % bhs_per_rec);
1044 * The record should be written. If a locked ntfs
1045 * inode was returned, add it to the array of locked
1049 locked_nis[nr_locked_nis++] = tni;
1051 /* Apply the mst protection fixups. */
1052 err2 = pre_write_mst_fixup((NTFS_RECORD*)(kaddr + ofs),
1054 if (unlikely(err2)) {
1055 if (!err || err == -ENOMEM)
1057 ntfs_error(vol->sb, "Failed to apply mst fixups "
1058 "(inode 0x%lx, attribute type 0x%x, "
1059 "page index 0x%lx, page offset 0x%x)!"
1060 " Unmount and run chkdsk.", vi->i_ino,
1061 ni->type, page->index, ofs);
1063 * Mark all the buffers in this record clean as we do
1064 * not want to write corrupt data to disk.
1067 clear_buffer_dirty(bhs[i]);
1069 } while (++i % bhs_per_rec);
1074 /* If no records are to be written out, we are done. */
1077 flush_dcache_page(page);
1078 /* Lock buffers and start synchronous write i/o on them. */
1079 for (i = 0; i < nr_bhs; i++) {
1083 if (unlikely(test_set_buffer_locked(tbh)))
1085 /* The buffer dirty state is now irrelevant, just clean it. */
1086 clear_buffer_dirty(tbh);
1087 BUG_ON(!buffer_uptodate(tbh));
1088 BUG_ON(!buffer_mapped(tbh));
1090 tbh->b_end_io = end_buffer_write_sync;
1091 submit_bh(WRITE, tbh);
1093 /* Synchronize the mft mirror now if not @sync. */
1094 if (is_mft && !sync)
1097 /* Wait on i/o completion of buffers. */
1098 for (i = 0; i < nr_bhs; i++) {
1102 wait_on_buffer(tbh);
1103 if (unlikely(!buffer_uptodate(tbh))) {
1104 ntfs_error(vol->sb, "I/O error while writing ntfs "
1105 "record buffer (inode 0x%lx, "
1106 "attribute type 0x%x, page index "
1107 "0x%lx, page offset 0x%lx)! Unmount "
1108 "and run chkdsk.", vi->i_ino, ni->type,
1109 page->index, bh_offset(tbh));
1110 if (!err || err == -ENOMEM)
1113 * Set the buffer uptodate so the page and buffer
1114 * states do not become out of sync.
1116 set_buffer_uptodate(tbh);
1119 /* If @sync, now synchronize the mft mirror. */
1120 if (is_mft && sync) {
1122 for (i = 0; i < nr_bhs; i++) {
1123 unsigned long mft_no;
1127 * Skip buffers which are not at the beginning of
1130 if (i % bhs_per_rec)
1133 /* Skip removed buffers (and hence records). */
1136 ofs = bh_offset(tbh);
1137 /* Get the mft record number. */
1138 mft_no = (((s64)page->index << PAGE_CACHE_SHIFT) + ofs)
1140 if (mft_no < vol->mftmirr_size)
1141 ntfs_sync_mft_mirror(vol, mft_no,
1142 (MFT_RECORD*)(kaddr + ofs),
1148 /* Remove the mst protection fixups again. */
1149 for (i = 0; i < nr_bhs; i++) {
1150 if (!(i % bhs_per_rec)) {
1154 post_write_mst_fixup((NTFS_RECORD*)(kaddr +
1158 flush_dcache_page(page);
1160 /* Unlock any locked inodes. */
1161 while (nr_locked_nis-- > 0) {
1162 ntfs_inode *tni, *base_tni;
1164 tni = locked_nis[nr_locked_nis];
1165 /* Get the base inode. */
1166 down(&tni->extent_lock);
1167 if (tni->nr_extents >= 0)
1170 base_tni = tni->ext.base_ntfs_ino;
1173 up(&tni->extent_lock);
1174 ntfs_debug("Unlocking %s inode 0x%lx.",
1175 tni == base_tni ? "base" : "extent",
1177 up(&tni->mrec_lock);
1178 atomic_dec(&tni->count);
1179 iput(VFS_I(base_tni));
1181 SetPageUptodate(page);
1184 if (unlikely(err && err != -ENOMEM)) {
1186 * Set page error if there is only one ntfs record in the page.
1187 * Otherwise we would loose per-record granularity.
1189 if (ni->itype.index.block_size == PAGE_CACHE_SIZE)
1193 if (page_is_dirty) {
1194 ntfs_debug("Page still contains one or more dirty ntfs "
1195 "records. Redirtying the page starting at "
1196 "record 0x%lx.", page->index <<
1197 (PAGE_CACHE_SHIFT - rec_size_bits));
1198 redirty_page_for_writepage(wbc, page);
1202 * Keep the VM happy. This must be done otherwise the
1203 * radix-tree tag PAGECACHE_TAG_DIRTY remains set even though
1204 * the page is clean.
1206 BUG_ON(PageWriteback(page));
1207 set_page_writeback(page);
1209 end_page_writeback(page);
1212 ntfs_debug("Done.");
1217 * ntfs_writepage - write a @page to the backing store
1218 * @page: page cache page to write out
1219 * @wbc: writeback control structure
1221 * This is called from the VM when it wants to have a dirty ntfs page cache
1222 * page cleaned. The VM has already locked the page and marked it clean.
1224 * For non-resident attributes, ntfs_writepage() writes the @page by calling
1225 * the ntfs version of the generic block_write_full_page() function,
1226 * ntfs_write_block(), which in turn if necessary creates and writes the
1227 * buffers associated with the page asynchronously.
1229 * For resident attributes, OTOH, ntfs_writepage() writes the @page by copying
1230 * the data to the mft record (which at this stage is most likely in memory).
1231 * The mft record is then marked dirty and written out asynchronously via the
1232 * vfs inode dirty code path for the inode the mft record belongs to or via the
1233 * vm page dirty code path for the page the mft record is in.
1235 * Based on ntfs_readpage() and fs/buffer.c::block_write_full_page().
1237 * Return 0 on success and -errno on error.
1239 static int ntfs_writepage(struct page *page, struct writeback_control *wbc)
1242 struct inode *vi = page->mapping->host;
1243 ntfs_inode *base_ni = NULL, *ni = NTFS_I(vi);
1245 ntfs_attr_search_ctx *ctx = NULL;
1246 MFT_RECORD *m = NULL;
1250 BUG_ON(!PageLocked(page));
1251 i_size = i_size_read(vi);
1252 /* Is the page fully outside i_size? (truncate in progress) */
1253 if (unlikely(page->index >= (i_size + PAGE_CACHE_SIZE - 1) >>
1254 PAGE_CACHE_SHIFT)) {
1256 * The page may have dirty, unmapped buffers. Make them
1257 * freeable here, so the page does not leak.
1259 block_invalidatepage(page, 0);
1261 ntfs_debug("Write outside i_size - truncated?");
1264 /* NInoNonResident() == NInoIndexAllocPresent() */
1265 if (NInoNonResident(ni)) {
1267 * Only unnamed $DATA attributes can be compressed, encrypted,
1270 if (ni->type == AT_DATA && !ni->name_len) {
1271 /* If file is encrypted, deny access, just like NT4. */
1272 if (NInoEncrypted(ni)) {
1274 ntfs_debug("Denying write access to encrypted "
1278 /* Compressed data streams are handled in compress.c. */
1279 if (NInoCompressed(ni)) {
1280 // TODO: Implement and replace this check with
1281 // return ntfs_write_compressed_block(page);
1283 ntfs_error(vi->i_sb, "Writing to compressed "
1284 "files is not supported yet. "
1288 // TODO: Implement and remove this check.
1289 if (NInoSparse(ni)) {
1291 ntfs_error(vi->i_sb, "Writing to sparse files "
1292 "is not supported yet. Sorry.");
1296 /* We have to zero every time due to mmap-at-end-of-file. */
1297 if (page->index >= (i_size >> PAGE_CACHE_SHIFT)) {
1298 /* The page straddles i_size. */
1299 unsigned int ofs = i_size & ~PAGE_CACHE_MASK;
1300 kaddr = kmap_atomic(page, KM_USER0);
1301 memset(kaddr + ofs, 0, PAGE_CACHE_SIZE - ofs);
1302 flush_dcache_page(page);
1303 kunmap_atomic(kaddr, KM_USER0);
1305 /* Handle mst protected attributes. */
1306 if (NInoMstProtected(ni))
1307 return ntfs_write_mst_block(page, wbc);
1308 /* Normal data stream. */
1309 return ntfs_write_block(page, wbc);
1312 * Attribute is resident, implying it is not compressed, encrypted,
1313 * sparse, or mst protected. This also means the attribute is smaller
1314 * than an mft record and hence smaller than a page, so can simply
1315 * return error on any pages with index above 0.
1317 BUG_ON(page_has_buffers(page));
1318 BUG_ON(!PageUptodate(page));
1319 if (unlikely(page->index > 0)) {
1320 ntfs_error(vi->i_sb, "BUG()! page->index (0x%lx) > 0. "
1321 "Aborting write.", page->index);
1322 BUG_ON(PageWriteback(page));
1323 set_page_writeback(page);
1325 end_page_writeback(page);
1331 base_ni = ni->ext.base_ntfs_ino;
1332 /* Map, pin, and lock the mft record. */
1333 m = map_mft_record(base_ni);
1340 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1341 if (unlikely(!ctx)) {
1345 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1346 CASE_SENSITIVE, 0, NULL, 0, ctx);
1350 * Keep the VM happy. This must be done otherwise the radix-tree tag
1351 * PAGECACHE_TAG_DIRTY remains set even though the page is clean.
1353 BUG_ON(PageWriteback(page));
1354 set_page_writeback(page);
1358 * Here, we don't need to zero the out of bounds area everytime because
1359 * the below memcpy() already takes care of the mmap-at-end-of-file
1360 * requirements. If the file is converted to a non-resident one, then
1361 * the code path use is switched to the non-resident one where the
1362 * zeroing happens on each ntfs_writepage() invocation.
1364 * The above also applies nicely when i_size is decreased.
1366 * When i_size is increased, the memory between the old and new i_size
1367 * _must_ be zeroed (or overwritten with new data). Otherwise we will
1368 * expose data to userspace/disk which should never have been exposed.
1370 * FIXME: Ensure that i_size increases do the zeroing/overwriting and
1371 * if we cannot guarantee that, then enable the zeroing below. If the
1372 * zeroing below is enabled, we MUST move the unlock_page() from above
1373 * to after the kunmap_atomic(), i.e. just before the
1374 * end_page_writeback().
1375 * UPDATE: ntfs_prepare/commit_write() do the zeroing on i_size
1376 * increases for resident attributes so those are ok.
1377 * TODO: ntfs_truncate(), others?
1380 attr_len = le32_to_cpu(ctx->attr->data.resident.value_length);
1381 i_size = i_size_read(vi);
1382 if (unlikely(attr_len > i_size)) {
1384 ctx->attr->data.resident.value_length = cpu_to_le32(attr_len);
1386 kaddr = kmap_atomic(page, KM_USER0);
1387 /* Copy the data from the page to the mft record. */
1388 memcpy((u8*)ctx->attr +
1389 le16_to_cpu(ctx->attr->data.resident.value_offset),
1391 flush_dcache_mft_record_page(ctx->ntfs_ino);
1392 /* Zero out of bounds area in the page cache page. */
1393 memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len);
1394 flush_dcache_page(page);
1395 kunmap_atomic(kaddr, KM_USER0);
1397 end_page_writeback(page);
1399 /* Mark the mft record dirty, so it gets written back. */
1400 mark_mft_record_dirty(ctx->ntfs_ino);
1401 ntfs_attr_put_search_ctx(ctx);
1402 unmap_mft_record(base_ni);
1405 if (err == -ENOMEM) {
1406 ntfs_warning(vi->i_sb, "Error allocating memory. Redirtying "
1407 "page so we try again later.");
1409 * Put the page back on mapping->dirty_pages, but leave its
1410 * buffers' dirty state as-is.
1412 redirty_page_for_writepage(wbc, page);
1415 ntfs_error(vi->i_sb, "Resident attribute write failed with "
1418 NVolSetErrors(ni->vol);
1423 ntfs_attr_put_search_ctx(ctx);
1425 unmap_mft_record(base_ni);
1430 * ntfs_prepare_nonresident_write -
1433 static int ntfs_prepare_nonresident_write(struct page *page,
1434 unsigned from, unsigned to)
1438 s64 initialized_size;
1440 sector_t block, ablock, iblock;
1444 runlist_element *rl;
1445 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
1446 unsigned long flags;
1447 unsigned int vcn_ofs, block_start, block_end, blocksize;
1450 unsigned char blocksize_bits;
1452 vi = page->mapping->host;
1456 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1457 "0x%lx, from = %u, to = %u.", ni->mft_no, ni->type,
1458 page->index, from, to);
1460 BUG_ON(!NInoNonResident(ni));
1462 blocksize_bits = vi->i_blkbits;
1463 blocksize = 1 << blocksize_bits;
1466 * create_empty_buffers() will create uptodate/dirty buffers if the
1467 * page is uptodate/dirty.
1469 if (!page_has_buffers(page))
1470 create_empty_buffers(page, blocksize, 0);
1471 bh = head = page_buffers(page);
1475 /* The first block in the page. */
1476 block = (s64)page->index << (PAGE_CACHE_SHIFT - blocksize_bits);
1478 read_lock_irqsave(&ni->size_lock, flags);
1480 * The first out of bounds block for the allocated size. No need to
1481 * round up as allocated_size is in multiples of cluster size and the
1482 * minimum cluster size is 512 bytes, which is equal to the smallest
1485 ablock = ni->allocated_size >> blocksize_bits;
1486 i_size = i_size_read(vi);
1487 initialized_size = ni->initialized_size;
1488 read_unlock_irqrestore(&ni->size_lock, flags);
1490 /* The last (fully or partially) initialized block. */
1491 iblock = initialized_size >> blocksize_bits;
1493 /* Loop through all the buffers in the page. */
1498 block_end = block_start + blocksize;
1500 * If buffer @bh is outside the write, just mark it uptodate
1501 * if the page is uptodate and continue with the next buffer.
1503 if (block_end <= from || block_start >= to) {
1504 if (PageUptodate(page)) {
1505 if (!buffer_uptodate(bh))
1506 set_buffer_uptodate(bh);
1511 * @bh is at least partially being written to.
1512 * Make sure it is not marked as new.
1514 //if (buffer_new(bh))
1515 // clear_buffer_new(bh);
1517 if (block >= ablock) {
1518 // TODO: block is above allocated_size, need to
1519 // allocate it. Best done in one go to accommodate not
1520 // only block but all above blocks up to and including:
1521 // ((page->index << PAGE_CACHE_SHIFT) + to + blocksize
1522 // - 1) >> blobksize_bits. Obviously will need to round
1523 // up to next cluster boundary, too. This should be
1524 // done with a helper function, so it can be reused.
1525 ntfs_error(vol->sb, "Writing beyond allocated size "
1526 "is not supported yet. Sorry.");
1529 // Need to update ablock.
1530 // Need to set_buffer_new() on all block bhs that are
1534 * Now we have enough allocated size to fulfill the whole
1535 * request, i.e. block < ablock is true.
1537 if (unlikely((block >= iblock) &&
1538 (initialized_size < i_size))) {
1540 * If this page is fully outside initialized size, zero
1541 * out all pages between the current initialized size
1542 * and the current page. Just use ntfs_readpage() to do
1543 * the zeroing transparently.
1545 if (block > iblock) {
1547 // For each page do:
1548 // - read_cache_page()
1549 // Again for each page do:
1550 // - wait_on_page_locked()
1551 // - Check (PageUptodate(page) &&
1552 // !PageError(page))
1553 // Update initialized size in the attribute and
1555 // Again, for each page do:
1556 // __set_page_dirty_buffers();
1557 // page_cache_release()
1558 // We don't need to wait on the writes.
1562 * The current page straddles initialized size. Zero
1563 * all non-uptodate buffers and set them uptodate (and
1564 * dirty?). Note, there aren't any non-uptodate buffers
1565 * if the page is uptodate.
1566 * FIXME: For an uptodate page, the buffers may need to
1567 * be written out because they were not initialized on
1570 if (!PageUptodate(page)) {
1572 // Zero any non-uptodate buffers up to i_size.
1573 // Set them uptodate and dirty.
1576 // Update initialized size in the attribute and in the
1577 // inode (up to i_size).
1579 // FIXME: This is inefficient. Try to batch the two
1580 // size changes to happen in one go.
1581 ntfs_error(vol->sb, "Writing beyond initialized size "
1582 "is not supported yet. Sorry.");
1585 // Do NOT set_buffer_new() BUT DO clear buffer range
1586 // outside write request range.
1587 // set_buffer_uptodate() on complete buffers as well as
1588 // set_buffer_dirty().
1591 /* Need to map unmapped buffers. */
1592 if (!buffer_mapped(bh)) {
1593 /* Unmapped buffer. Need to map it. */
1594 bh->b_bdev = vol->sb->s_bdev;
1596 /* Convert block into corresponding vcn and offset. */
1597 vcn = (VCN)block << blocksize_bits >>
1598 vol->cluster_size_bits;
1599 vcn_ofs = ((VCN)block << blocksize_bits) &
1600 vol->cluster_size_mask;
1605 down_read(&ni->runlist.lock);
1606 rl = ni->runlist.rl;
1608 if (likely(rl != NULL)) {
1609 /* Seek to element containing target vcn. */
1610 while (rl->length && rl[1].vcn <= vcn)
1612 lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
1614 lcn = LCN_RL_NOT_MAPPED;
1615 if (unlikely(lcn < 0)) {
1617 * We extended the attribute allocation above.
1618 * If we hit an ENOENT here it means that the
1619 * allocation was insufficient which is a bug.
1621 BUG_ON(lcn == LCN_ENOENT);
1623 /* It is a hole, need to instantiate it. */
1624 if (lcn == LCN_HOLE) {
1625 // TODO: Instantiate the hole.
1626 // clear_buffer_new(bh);
1627 // unmap_underlying_metadata(bh->b_bdev,
1629 // For non-uptodate buffers, need to
1630 // zero out the region outside the
1631 // request in this bh or all bhs,
1632 // depending on what we implemented
1634 // Need to flush_dcache_page().
1635 // Or could use set_buffer_new()
1637 ntfs_error(vol->sb, "Writing into "
1638 "sparse regions is "
1639 "not supported yet. "
1643 } else if (!is_retry &&
1644 lcn == LCN_RL_NOT_MAPPED) {
1647 * Attempt to map runlist, dropping
1648 * lock for the duration.
1650 up_read(&ni->runlist.lock);
1651 err = ntfs_map_runlist(ni, vcn);
1653 goto lock_retry_remap;
1658 * Failed to map the buffer, even after
1662 ntfs_error(vol->sb, "Failed to write to inode "
1663 "0x%lx, attribute type 0x%x, "
1664 "vcn 0x%llx, offset 0x%x "
1665 "because its location on disk "
1666 "could not be determined%s "
1667 "(error code %lli).",
1668 ni->mft_no, ni->type,
1669 (unsigned long long)vcn,
1670 vcn_ofs, is_retry ? " even "
1671 "after retrying" : "",
1677 /* We now have a successful remap, i.e. lcn >= 0. */
1679 /* Setup buffer head to correct block. */
1680 bh->b_blocknr = ((lcn << vol->cluster_size_bits)
1681 + vcn_ofs) >> blocksize_bits;
1682 set_buffer_mapped(bh);
1684 // FIXME: Something analogous to this is needed for
1685 // each newly allocated block, i.e. BH_New.
1686 // FIXME: Might need to take this out of the
1687 // if (!buffer_mapped(bh)) {}, depending on how we
1688 // implement things during the allocated_size and
1689 // initialized_size extension code above.
1690 if (buffer_new(bh)) {
1691 clear_buffer_new(bh);
1692 unmap_underlying_metadata(bh->b_bdev,
1694 if (PageUptodate(page)) {
1695 set_buffer_uptodate(bh);
1699 * Page is _not_ uptodate, zero surrounding
1700 * region. NOTE: This is how we decide if to
1703 if (block_end > to || block_start < from) {
1706 kaddr = kmap_atomic(page, KM_USER0);
1708 memset(kaddr + to, 0,
1710 if (block_start < from)
1711 memset(kaddr + block_start, 0,
1714 flush_dcache_page(page);
1715 kunmap_atomic(kaddr, KM_USER0);
1720 /* @bh is mapped, set it uptodate if the page is uptodate. */
1721 if (PageUptodate(page)) {
1722 if (!buffer_uptodate(bh))
1723 set_buffer_uptodate(bh);
1727 * The page is not uptodate. The buffer is mapped. If it is not
1728 * uptodate, and it is only partially being written to, we need
1729 * to read the buffer in before the write, i.e. right now.
1731 if (!buffer_uptodate(bh) &&
1732 (block_start < from || block_end > to)) {
1733 ll_rw_block(READ, 1, &bh);
1736 } while (block++, block_start = block_end,
1737 (bh = bh->b_this_page) != head);
1739 /* Release the lock if we took it. */
1741 up_read(&ni->runlist.lock);
1745 /* If we issued read requests, let them complete. */
1746 while (wait_bh > wait) {
1747 wait_on_buffer(*--wait_bh);
1748 if (!buffer_uptodate(*wait_bh))
1752 ntfs_debug("Done.");
1756 * Zero out any newly allocated blocks to avoid exposing stale data.
1757 * If BH_New is set, we know that the block was newly allocated in the
1759 * FIXME: What about initialized_size increments? Have we done all the
1760 * required zeroing above? If not this error handling is broken, and
1761 * in particular the if (block_end <= from) check is completely bogus.
1767 block_end = block_start + blocksize;
1768 if (block_end <= from)
1770 if (block_start >= to)
1772 if (buffer_new(bh)) {
1775 clear_buffer_new(bh);
1776 kaddr = kmap_atomic(page, KM_USER0);
1777 memset(kaddr + block_start, 0, bh->b_size);
1778 kunmap_atomic(kaddr, KM_USER0);
1779 set_buffer_uptodate(bh);
1780 mark_buffer_dirty(bh);
1783 } while (block_start = block_end, (bh = bh->b_this_page) != head);
1785 flush_dcache_page(page);
1787 up_read(&ni->runlist.lock);
1792 * ntfs_prepare_write - prepare a page for receiving data
1794 * This is called from generic_file_write() with i_sem held on the inode
1795 * (@page->mapping->host). The @page is locked but not kmap()ped. The source
1796 * data has not yet been copied into the @page.
1798 * Need to extend the attribute/fill in holes if necessary, create blocks and
1799 * make partially overwritten blocks uptodate,
1801 * i_size is not to be modified yet.
1803 * Return 0 on success or -errno on error.
1805 * Should be using block_prepare_write() [support for sparse files] or
1806 * cont_prepare_write() [no support for sparse files]. Cannot do that due to
1807 * ntfs specifics but can look at them for implementation guidance.
1809 * Note: In the range, @from is inclusive and @to is exclusive, i.e. @from is
1810 * the first byte in the page that will be written to and @to is the first byte
1811 * after the last byte that will be written to.
1813 static int ntfs_prepare_write(struct file *file, struct page *page,
1814 unsigned from, unsigned to)
1818 struct inode *vi = page->mapping->host;
1819 ntfs_inode *base_ni = NULL, *ni = NTFS_I(vi);
1820 ntfs_volume *vol = ni->vol;
1821 ntfs_attr_search_ctx *ctx = NULL;
1822 MFT_RECORD *m = NULL;
1828 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
1829 "0x%lx, from = %u, to = %u.", vi->i_ino, ni->type,
1830 page->index, from, to);
1831 BUG_ON(!PageLocked(page));
1832 BUG_ON(from > PAGE_CACHE_SIZE);
1833 BUG_ON(to > PAGE_CACHE_SIZE);
1835 BUG_ON(NInoMstProtected(ni));
1837 * If a previous ntfs_truncate() failed, repeat it and abort if it
1840 if (unlikely(NInoTruncateFailed(ni))) {
1841 down_write(&vi->i_alloc_sem);
1842 err = ntfs_truncate(vi);
1843 up_write(&vi->i_alloc_sem);
1844 if (err || NInoTruncateFailed(ni)) {
1850 /* If the attribute is not resident, deal with it elsewhere. */
1851 if (NInoNonResident(ni)) {
1853 * Only unnamed $DATA attributes can be compressed, encrypted,
1856 if (ni->type == AT_DATA && !ni->name_len) {
1857 /* If file is encrypted, deny access, just like NT4. */
1858 if (NInoEncrypted(ni)) {
1859 ntfs_debug("Denying write access to encrypted "
1863 /* Compressed data streams are handled in compress.c. */
1864 if (NInoCompressed(ni)) {
1865 // TODO: Implement and replace this check with
1866 // return ntfs_write_compressed_block(page);
1867 ntfs_error(vi->i_sb, "Writing to compressed "
1868 "files is not supported yet. "
1872 // TODO: Implement and remove this check.
1873 if (NInoSparse(ni)) {
1874 ntfs_error(vi->i_sb, "Writing to sparse files "
1875 "is not supported yet. Sorry.");
1879 /* Normal data stream. */
1880 return ntfs_prepare_nonresident_write(page, from, to);
1883 * Attribute is resident, implying it is not compressed, encrypted, or
1886 BUG_ON(page_has_buffers(page));
1887 new_size = ((s64)page->index << PAGE_CACHE_SHIFT) + to;
1888 /* If we do not need to resize the attribute allocation we are done. */
1889 if (new_size <= i_size_read(vi))
1891 /* Map, pin, and lock the (base) mft record. */
1895 base_ni = ni->ext.base_ntfs_ino;
1896 m = map_mft_record(base_ni);
1903 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1904 if (unlikely(!ctx)) {
1908 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1909 CASE_SENSITIVE, 0, NULL, 0, ctx);
1910 if (unlikely(err)) {
1917 /* The total length of the attribute value. */
1918 attr_len = le32_to_cpu(a->data.resident.value_length);
1919 /* Fix an eventual previous failure of ntfs_commit_write(). */
1920 i_size = i_size_read(vi);
1921 if (unlikely(attr_len > i_size)) {
1923 a->data.resident.value_length = cpu_to_le32(attr_len);
1925 /* If we do not need to resize the attribute allocation we are done. */
1926 if (new_size <= attr_len)
1928 /* Check if new size is allowed in $AttrDef. */
1929 err = ntfs_attr_size_bounds_check(vol, ni->type, new_size);
1930 if (unlikely(err)) {
1931 if (err == -ERANGE) {
1932 ntfs_error(vol->sb, "Write would cause the inode "
1933 "0x%lx to exceed the maximum size for "
1934 "its attribute type (0x%x). Aborting "
1935 "write.", vi->i_ino,
1936 le32_to_cpu(ni->type));
1938 ntfs_error(vol->sb, "Inode 0x%lx has unknown "
1939 "attribute type 0x%x. Aborting "
1940 "write.", vi->i_ino,
1941 le32_to_cpu(ni->type));
1947 * Extend the attribute record to be able to store the new attribute
1950 if (new_size >= vol->mft_record_size || ntfs_attr_record_resize(m, a,
1951 le16_to_cpu(a->data.resident.value_offset) +
1953 /* Not enough space in the mft record. */
1954 ntfs_error(vol->sb, "Not enough space in the mft record for "
1955 "the resized attribute value. This is not "
1956 "supported yet. Aborting write.");
1961 * We have enough space in the mft record to fit the write. This
1962 * implies the attribute is smaller than the mft record and hence the
1963 * attribute must be in a single page and hence page->index must be 0.
1965 BUG_ON(page->index);
1967 * If the beginning of the write is past the old size, enlarge the
1968 * attribute value up to the beginning of the write and fill it with
1971 if (from > attr_len) {
1972 memset((u8*)a + le16_to_cpu(a->data.resident.value_offset) +
1973 attr_len, 0, from - attr_len);
1974 a->data.resident.value_length = cpu_to_le32(from);
1975 /* Zero the corresponding area in the page as well. */
1976 if (PageUptodate(page)) {
1977 kaddr = kmap_atomic(page, KM_USER0);
1978 memset(kaddr + attr_len, 0, from - attr_len);
1979 kunmap_atomic(kaddr, KM_USER0);
1980 flush_dcache_page(page);
1983 flush_dcache_mft_record_page(ctx->ntfs_ino);
1984 mark_mft_record_dirty(ctx->ntfs_ino);
1986 ntfs_attr_put_search_ctx(ctx);
1987 unmap_mft_record(base_ni);
1989 * Because resident attributes are handled by memcpy() to/from the
1990 * corresponding MFT record, and because this form of i/o is byte
1991 * aligned rather than block aligned, there is no need to bring the
1992 * page uptodate here as in the non-resident case where we need to
1993 * bring the buffers straddled by the write uptodate before
1994 * generic_file_write() does the copying from userspace.
1996 * We thus defer the uptodate bringing of the page region outside the
1997 * region written to to ntfs_commit_write(), which makes the code
1998 * simpler and saves one atomic kmap which is good.
2001 ntfs_debug("Done.");
2005 ntfs_warning(vi->i_sb, "Error allocating memory required to "
2006 "prepare the write.");
2008 ntfs_error(vi->i_sb, "Resident attribute prepare write failed "
2009 "with error %i.", err);
2015 ntfs_attr_put_search_ctx(ctx);
2017 unmap_mft_record(base_ni);
2022 * ntfs_commit_nonresident_write -
2025 static int ntfs_commit_nonresident_write(struct page *page,
2026 unsigned from, unsigned to)
2028 s64 pos = ((s64)page->index << PAGE_CACHE_SHIFT) + to;
2029 struct inode *vi = page->mapping->host;
2030 struct buffer_head *bh, *head;
2031 unsigned int block_start, block_end, blocksize;
2034 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
2035 "0x%lx, from = %u, to = %u.", vi->i_ino,
2036 NTFS_I(vi)->type, page->index, from, to);
2037 blocksize = 1 << vi->i_blkbits;
2039 // FIXME: We need a whole slew of special cases in here for compressed
2040 // files for example...
2041 // For now, we know ntfs_prepare_write() would have failed so we can't
2042 // get here in any of the cases which we have to special case, so we
2043 // are just a ripped off, unrolled generic_commit_write().
2045 bh = head = page_buffers(page);
2049 block_end = block_start + blocksize;
2050 if (block_end <= from || block_start >= to) {
2051 if (!buffer_uptodate(bh))
2054 set_buffer_uptodate(bh);
2055 mark_buffer_dirty(bh);
2057 } while (block_start = block_end, (bh = bh->b_this_page) != head);
2059 * If this is a partial write which happened to make all buffers
2060 * uptodate then we can optimize away a bogus ->readpage() for the next
2061 * read(). Here we 'discover' whether the page went uptodate as a
2062 * result of this (potentially partial) write.
2065 SetPageUptodate(page);
2067 * Not convinced about this at all. See disparity comment above. For
2068 * now we know ntfs_prepare_write() would have failed in the write
2069 * exceeds i_size case, so this will never trigger which is fine.
2071 if (pos > i_size_read(vi)) {
2072 ntfs_error(vi->i_sb, "Writing beyond the existing file size is "
2073 "not supported yet. Sorry.");
2075 // vi->i_size = pos;
2076 // mark_inode_dirty(vi);
2078 ntfs_debug("Done.");
2083 * ntfs_commit_write - commit the received data
2085 * This is called from generic_file_write() with i_sem held on the inode
2086 * (@page->mapping->host). The @page is locked but not kmap()ped. The source
2087 * data has already been copied into the @page. ntfs_prepare_write() has been
2088 * called before the data copied and it returned success so we can take the
2089 * results of various BUG checks and some error handling for granted.
2091 * Need to mark modified blocks dirty so they get written out later when
2092 * ntfs_writepage() is invoked by the VM.
2094 * Return 0 on success or -errno on error.
2096 * Should be using generic_commit_write(). This marks buffers uptodate and
2097 * dirty, sets the page uptodate if all buffers in the page are uptodate, and
2098 * updates i_size if the end of io is beyond i_size. In that case, it also
2099 * marks the inode dirty.
2101 * Cannot use generic_commit_write() due to ntfs specialities but can look at
2102 * it for implementation guidance.
2104 * If things have gone as outlined in ntfs_prepare_write(), then we do not
2105 * need to do any page content modifications here at all, except in the write
2106 * to resident attribute case, where we need to do the uptodate bringing here
2107 * which we combine with the copying into the mft record which means we save
2110 static int ntfs_commit_write(struct file *file, struct page *page,
2111 unsigned from, unsigned to)
2113 struct inode *vi = page->mapping->host;
2114 ntfs_inode *base_ni, *ni = NTFS_I(vi);
2115 char *kaddr, *kattr;
2116 ntfs_attr_search_ctx *ctx;
2122 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, page index "
2123 "0x%lx, from = %u, to = %u.", vi->i_ino, ni->type,
2124 page->index, from, to);
2125 /* If the attribute is not resident, deal with it elsewhere. */
2126 if (NInoNonResident(ni)) {
2127 /* Only unnamed $DATA attributes can be compressed/encrypted. */
2128 if (ni->type == AT_DATA && !ni->name_len) {
2129 /* Encrypted files need separate handling. */
2130 if (NInoEncrypted(ni)) {
2131 // We never get here at present!
2134 /* Compressed data streams are handled in compress.c. */
2135 if (NInoCompressed(ni)) {
2136 // TODO: Implement this!
2137 // return ntfs_write_compressed_block(page);
2138 // We never get here at present!
2142 /* Normal data stream. */
2143 return ntfs_commit_nonresident_write(page, from, to);
2146 * Attribute is resident, implying it is not compressed, encrypted, or
2152 base_ni = ni->ext.base_ntfs_ino;
2153 /* Map, pin, and lock the mft record. */
2154 m = map_mft_record(base_ni);
2161 ctx = ntfs_attr_get_search_ctx(base_ni, m);
2162 if (unlikely(!ctx)) {
2166 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
2167 CASE_SENSITIVE, 0, NULL, 0, ctx);
2168 if (unlikely(err)) {
2174 /* The total length of the attribute value. */
2175 attr_len = le32_to_cpu(a->data.resident.value_length);
2176 BUG_ON(from > attr_len);
2177 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
2178 kaddr = kmap_atomic(page, KM_USER0);
2179 /* Copy the received data from the page to the mft record. */
2180 memcpy(kattr + from, kaddr + from, to - from);
2181 /* Update the attribute length if necessary. */
2182 if (to > attr_len) {
2184 a->data.resident.value_length = cpu_to_le32(attr_len);
2187 * If the page is not uptodate, bring the out of bounds area(s)
2188 * uptodate by copying data from the mft record to the page.
2190 if (!PageUptodate(page)) {
2192 memcpy(kaddr, kattr, from);
2194 memcpy(kaddr + to, kattr + to, attr_len - to);
2195 /* Zero the region outside the end of the attribute value. */
2196 if (attr_len < PAGE_CACHE_SIZE)
2197 memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len);
2199 * The probability of not having done any of the above is
2200 * extremely small, so we just flush unconditionally.
2202 flush_dcache_page(page);
2203 SetPageUptodate(page);
2205 kunmap_atomic(kaddr, KM_USER0);
2206 /* Update i_size if necessary. */
2207 if (i_size_read(vi) < attr_len) {
2208 unsigned long flags;
2210 write_lock_irqsave(&ni->size_lock, flags);
2211 ni->allocated_size = ni->initialized_size = attr_len;
2212 i_size_write(vi, attr_len);
2213 write_unlock_irqrestore(&ni->size_lock, flags);
2215 /* Mark the mft record dirty, so it gets written back. */
2216 flush_dcache_mft_record_page(ctx->ntfs_ino);
2217 mark_mft_record_dirty(ctx->ntfs_ino);
2218 ntfs_attr_put_search_ctx(ctx);
2219 unmap_mft_record(base_ni);
2220 ntfs_debug("Done.");
2223 if (err == -ENOMEM) {
2224 ntfs_warning(vi->i_sb, "Error allocating memory required to "
2225 "commit the write.");
2226 if (PageUptodate(page)) {
2227 ntfs_warning(vi->i_sb, "Page is uptodate, setting "
2228 "dirty so the write will be retried "
2229 "later on by the VM.");
2231 * Put the page on mapping->dirty_pages, but leave its
2232 * buffers' dirty state as-is.
2234 __set_page_dirty_nobuffers(page);
2237 ntfs_error(vi->i_sb, "Page is not uptodate. Written "
2238 "data has been lost.");
2240 ntfs_error(vi->i_sb, "Resident attribute commit write failed "
2241 "with error %i.", err);
2242 NVolSetErrors(ni->vol);
2246 ntfs_attr_put_search_ctx(ctx);
2248 unmap_mft_record(base_ni);
2252 #endif /* NTFS_RW */
2255 * ntfs_aops - general address space operations for inodes and attributes
2257 struct address_space_operations ntfs_aops = {
2258 .readpage = ntfs_readpage, /* Fill page with data. */
2259 .sync_page = block_sync_page, /* Currently, just unplugs the
2260 disk request queue. */
2262 .writepage = ntfs_writepage, /* Write dirty page to disk. */
2263 .prepare_write = ntfs_prepare_write, /* Prepare page and buffers
2264 ready to receive data. */
2265 .commit_write = ntfs_commit_write, /* Commit received data. */
2266 #endif /* NTFS_RW */
2270 * ntfs_mst_aops - general address space operations for mst protecteed inodes
2273 struct address_space_operations ntfs_mst_aops = {
2274 .readpage = ntfs_readpage, /* Fill page with data. */
2275 .sync_page = block_sync_page, /* Currently, just unplugs the
2276 disk request queue. */
2278 .writepage = ntfs_writepage, /* Write dirty page to disk. */
2279 .set_page_dirty = __set_page_dirty_nobuffers, /* Set the page dirty
2280 without touching the buffers
2281 belonging to the page. */
2282 #endif /* NTFS_RW */
2288 * mark_ntfs_record_dirty - mark an ntfs record dirty
2289 * @page: page containing the ntfs record to mark dirty
2290 * @ofs: byte offset within @page at which the ntfs record begins
2292 * Set the buffers and the page in which the ntfs record is located dirty.
2294 * The latter also marks the vfs inode the ntfs record belongs to dirty
2295 * (I_DIRTY_PAGES only).
2297 * If the page does not have buffers, we create them and set them uptodate.
2298 * The page may not be locked which is why we need to handle the buffers under
2299 * the mapping->private_lock. Once the buffers are marked dirty we no longer
2300 * need the lock since try_to_free_buffers() does not free dirty buffers.
2302 void mark_ntfs_record_dirty(struct page *page, const unsigned int ofs) {
2303 struct address_space *mapping = page->mapping;
2304 ntfs_inode *ni = NTFS_I(mapping->host);
2305 struct buffer_head *bh, *head, *buffers_to_free = NULL;
2306 unsigned int end, bh_size, bh_ofs;
2308 BUG_ON(!PageUptodate(page));
2309 end = ofs + ni->itype.index.block_size;
2310 bh_size = 1 << VFS_I(ni)->i_blkbits;
2311 spin_lock(&mapping->private_lock);
2312 if (unlikely(!page_has_buffers(page))) {
2313 spin_unlock(&mapping->private_lock);
2314 bh = head = alloc_page_buffers(page, bh_size, 1);
2315 spin_lock(&mapping->private_lock);
2316 if (likely(!page_has_buffers(page))) {
2317 struct buffer_head *tail;
2320 set_buffer_uptodate(bh);
2322 bh = bh->b_this_page;
2324 tail->b_this_page = head;
2325 attach_page_buffers(page, head);
2327 buffers_to_free = bh;
2329 bh = head = page_buffers(page);
2331 bh_ofs = bh_offset(bh);
2332 if (bh_ofs + bh_size <= ofs)
2334 if (unlikely(bh_ofs >= end))
2336 set_buffer_dirty(bh);
2337 } while ((bh = bh->b_this_page) != head);
2338 spin_unlock(&mapping->private_lock);
2339 __set_page_dirty_nobuffers(page);
2340 if (unlikely(buffers_to_free)) {
2342 bh = buffers_to_free->b_this_page;
2343 free_buffer_head(buffers_to_free);
2344 buffers_to_free = bh;
2345 } while (buffers_to_free);
2349 #endif /* NTFS_RW */