1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * file.c - NTFS kernel file operations. Part of the Linux-NTFS project.
5 * Copyright (c) 2001-2015 Anton Altaparmakov and Tuxera Inc.
8 #include <linux/blkdev.h>
9 #include <linux/backing-dev.h>
10 #include <linux/buffer_head.h>
11 #include <linux/gfp.h>
12 #include <linux/pagemap.h>
13 #include <linux/pagevec.h>
14 #include <linux/sched/signal.h>
15 #include <linux/swap.h>
16 #include <linux/uio.h>
17 #include <linux/writeback.h>
20 #include <linux/uaccess.h>
32 * ntfs_file_open - called when an inode is about to be opened
33 * @vi: inode to be opened
34 * @filp: file structure describing the inode
36 * Limit file size to the page cache limit on architectures where unsigned long
37 * is 32-bits. This is the most we can do for now without overflowing the page
38 * cache page index. Doing it this way means we don't run into problems because
39 * of existing too large files. It would be better to allow the user to read
40 * the beginning of the file but I doubt very much anyone is going to hit this
41 * check on a 32-bit architecture, so there is no point in adding the extra
42 * complexity required to support this.
44 * On 64-bit architectures, the check is hopefully optimized away by the
47 * After the check passes, just call generic_file_open() to do its work.
49 static int ntfs_file_open(struct inode *vi, struct file *filp)
51 if (sizeof(unsigned long) < 8) {
52 if (i_size_read(vi) > MAX_LFS_FILESIZE)
55 return generic_file_open(vi, filp);
61 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
62 * @ni: ntfs inode of the attribute to extend
63 * @new_init_size: requested new initialized size in bytes
65 * Extend the initialized size of an attribute described by the ntfs inode @ni
66 * to @new_init_size bytes. This involves zeroing any non-sparse space between
67 * the old initialized size and @new_init_size both in the page cache and on
68 * disk (if relevant complete pages are already uptodate in the page cache then
69 * these are simply marked dirty).
71 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
72 * in the resident attribute case, it is tied to the initialized size and, in
73 * the non-resident attribute case, it may not fall below the initialized size.
75 * Note that if the attribute is resident, we do not need to touch the page
76 * cache at all. This is because if the page cache page is not uptodate we
77 * bring it uptodate later, when doing the write to the mft record since we
78 * then already have the page mapped. And if the page is uptodate, the
79 * non-initialized region will already have been zeroed when the page was
80 * brought uptodate and the region may in fact already have been overwritten
81 * with new data via mmap() based writes, so we cannot just zero it. And since
82 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
83 * is unspecified, we choose not to do zeroing and thus we do not need to touch
84 * the page at all. For a more detailed explanation see ntfs_truncate() in
87 * Return 0 on success and -errno on error. In the case that an error is
88 * encountered it is possible that the initialized size will already have been
89 * incremented some way towards @new_init_size but it is guaranteed that if
90 * this is the case, the necessary zeroing will also have happened and that all
91 * metadata is self-consistent.
93 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
96 static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size)
100 pgoff_t index, end_index;
102 struct inode *vi = VFS_I(ni);
104 MFT_RECORD *m = NULL;
106 ntfs_attr_search_ctx *ctx = NULL;
107 struct address_space *mapping;
108 struct page *page = NULL;
113 read_lock_irqsave(&ni->size_lock, flags);
114 old_init_size = ni->initialized_size;
115 old_i_size = i_size_read(vi);
116 BUG_ON(new_init_size > ni->allocated_size);
117 read_unlock_irqrestore(&ni->size_lock, flags);
118 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
119 "old_initialized_size 0x%llx, "
120 "new_initialized_size 0x%llx, i_size 0x%llx.",
121 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
122 (unsigned long long)old_init_size,
123 (unsigned long long)new_init_size, old_i_size);
127 base_ni = ni->ext.base_ntfs_ino;
128 /* Use goto to reduce indentation and we need the label below anyway. */
129 if (NInoNonResident(ni))
130 goto do_non_resident_extend;
131 BUG_ON(old_init_size != old_i_size);
132 m = map_mft_record(base_ni);
138 ctx = ntfs_attr_get_search_ctx(base_ni, m);
139 if (unlikely(!ctx)) {
143 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
144 CASE_SENSITIVE, 0, NULL, 0, ctx);
152 BUG_ON(a->non_resident);
153 /* The total length of the attribute value. */
154 attr_len = le32_to_cpu(a->data.resident.value_length);
155 BUG_ON(old_i_size != (loff_t)attr_len);
157 * Do the zeroing in the mft record and update the attribute size in
160 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
161 memset(kattr + attr_len, 0, new_init_size - attr_len);
162 a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
163 /* Finally, update the sizes in the vfs and ntfs inodes. */
164 write_lock_irqsave(&ni->size_lock, flags);
165 i_size_write(vi, new_init_size);
166 ni->initialized_size = new_init_size;
167 write_unlock_irqrestore(&ni->size_lock, flags);
169 do_non_resident_extend:
171 * If the new initialized size @new_init_size exceeds the current file
172 * size (vfs inode->i_size), we need to extend the file size to the
173 * new initialized size.
175 if (new_init_size > old_i_size) {
176 m = map_mft_record(base_ni);
182 ctx = ntfs_attr_get_search_ctx(base_ni, m);
183 if (unlikely(!ctx)) {
187 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
188 CASE_SENSITIVE, 0, NULL, 0, ctx);
196 BUG_ON(!a->non_resident);
197 BUG_ON(old_i_size != (loff_t)
198 sle64_to_cpu(a->data.non_resident.data_size));
199 a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
200 flush_dcache_mft_record_page(ctx->ntfs_ino);
201 mark_mft_record_dirty(ctx->ntfs_ino);
202 /* Update the file size in the vfs inode. */
203 i_size_write(vi, new_init_size);
204 ntfs_attr_put_search_ctx(ctx);
206 unmap_mft_record(base_ni);
209 mapping = vi->i_mapping;
210 index = old_init_size >> PAGE_SHIFT;
211 end_index = (new_init_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
214 * Read the page. If the page is not present, this will zero
215 * the uninitialized regions for us.
217 page = read_mapping_page(mapping, index, NULL);
223 * Update the initialized size in the ntfs inode. This is
224 * enough to make ntfs_writepage() work.
226 write_lock_irqsave(&ni->size_lock, flags);
227 ni->initialized_size = (s64)(index + 1) << PAGE_SHIFT;
228 if (ni->initialized_size > new_init_size)
229 ni->initialized_size = new_init_size;
230 write_unlock_irqrestore(&ni->size_lock, flags);
231 /* Set the page dirty so it gets written out. */
232 set_page_dirty(page);
235 * Play nice with the vm and the rest of the system. This is
236 * very much needed as we can potentially be modifying the
237 * initialised size from a very small value to a really huge
239 * f = open(somefile, O_TRUNC);
240 * truncate(f, 10GiB);
243 * And this would mean we would be marking dirty hundreds of
244 * thousands of pages or as in the above example more than
245 * two and a half million pages!
247 * TODO: For sparse pages could optimize this workload by using
248 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This
249 * would be set in read_folio for sparse pages and here we would
250 * not need to mark dirty any pages which have this bit set.
251 * The only caveat is that we have to clear the bit everywhere
252 * where we allocate any clusters that lie in the page or that
255 * TODO: An even greater optimization would be for us to only
256 * call read_folio() on pages which are not in sparse regions as
257 * determined from the runlist. This would greatly reduce the
258 * number of pages we read and make dirty in the case of sparse
261 balance_dirty_pages_ratelimited(mapping);
263 } while (++index < end_index);
264 read_lock_irqsave(&ni->size_lock, flags);
265 BUG_ON(ni->initialized_size != new_init_size);
266 read_unlock_irqrestore(&ni->size_lock, flags);
267 /* Now bring in sync the initialized_size in the mft record. */
268 m = map_mft_record(base_ni);
274 ctx = ntfs_attr_get_search_ctx(base_ni, m);
275 if (unlikely(!ctx)) {
279 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
280 CASE_SENSITIVE, 0, NULL, 0, ctx);
288 BUG_ON(!a->non_resident);
289 a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
291 flush_dcache_mft_record_page(ctx->ntfs_ino);
292 mark_mft_record_dirty(ctx->ntfs_ino);
294 ntfs_attr_put_search_ctx(ctx);
296 unmap_mft_record(base_ni);
297 ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
298 (unsigned long long)new_init_size, i_size_read(vi));
301 write_lock_irqsave(&ni->size_lock, flags);
302 ni->initialized_size = old_init_size;
303 write_unlock_irqrestore(&ni->size_lock, flags);
306 ntfs_attr_put_search_ctx(ctx);
308 unmap_mft_record(base_ni);
309 ntfs_debug("Failed. Returning error code %i.", err);
313 static ssize_t ntfs_prepare_file_for_write(struct kiocb *iocb,
314 struct iov_iter *from)
320 struct file *file = iocb->ki_filp;
321 struct inode *vi = file_inode(file);
322 ntfs_inode *ni = NTFS_I(vi);
323 ntfs_volume *vol = ni->vol;
325 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
326 "0x%llx, count 0x%zx.", vi->i_ino,
327 (unsigned)le32_to_cpu(ni->type),
328 (unsigned long long)iocb->ki_pos,
329 iov_iter_count(from));
330 err = generic_write_checks(iocb, from);
331 if (unlikely(err <= 0))
334 * All checks have passed. Before we start doing any writing we want
335 * to abort any totally illegal writes.
337 BUG_ON(NInoMstProtected(ni));
338 BUG_ON(ni->type != AT_DATA);
339 /* If file is encrypted, deny access, just like NT4. */
340 if (NInoEncrypted(ni)) {
341 /* Only $DATA attributes can be encrypted. */
343 * Reminder for later: Encrypted files are _always_
344 * non-resident so that the content can always be encrypted.
346 ntfs_debug("Denying write access to encrypted file.");
350 if (NInoCompressed(ni)) {
351 /* Only unnamed $DATA attribute can be compressed. */
352 BUG_ON(ni->name_len);
354 * Reminder for later: If resident, the data is not actually
355 * compressed. Only on the switch to non-resident does
356 * compression kick in. This is in contrast to encrypted files
359 ntfs_error(vi->i_sb, "Writing to compressed files is not "
360 "implemented yet. Sorry.");
364 err = file_remove_privs(file);
368 * Our ->update_time method always succeeds thus file_update_time()
369 * cannot fail either so there is no need to check the return code.
371 file_update_time(file);
373 /* The first byte after the last cluster being written to. */
374 end = (pos + iov_iter_count(from) + vol->cluster_size_mask) &
375 ~(u64)vol->cluster_size_mask;
377 * If the write goes beyond the allocated size, extend the allocation
378 * to cover the whole of the write, rounded up to the nearest cluster.
380 read_lock_irqsave(&ni->size_lock, flags);
381 ll = ni->allocated_size;
382 read_unlock_irqrestore(&ni->size_lock, flags);
385 * Extend the allocation without changing the data size.
387 * Note we ensure the allocation is big enough to at least
388 * write some data but we do not require the allocation to be
389 * complete, i.e. it may be partial.
391 ll = ntfs_attr_extend_allocation(ni, end, -1, pos);
392 if (likely(ll >= 0)) {
394 /* If the extension was partial truncate the write. */
396 ntfs_debug("Truncating write to inode 0x%lx, "
397 "attribute type 0x%x, because "
398 "the allocation was only "
399 "partially extended.",
400 vi->i_ino, (unsigned)
401 le32_to_cpu(ni->type));
402 iov_iter_truncate(from, ll - pos);
406 read_lock_irqsave(&ni->size_lock, flags);
407 ll = ni->allocated_size;
408 read_unlock_irqrestore(&ni->size_lock, flags);
409 /* Perform a partial write if possible or fail. */
411 ntfs_debug("Truncating write to inode 0x%lx "
412 "attribute type 0x%x, because "
413 "extending the allocation "
414 "failed (error %d).",
415 vi->i_ino, (unsigned)
416 le32_to_cpu(ni->type),
418 iov_iter_truncate(from, ll - pos);
421 ntfs_error(vi->i_sb, "Cannot perform "
424 "type 0x%x, because "
428 vi->i_ino, (unsigned)
429 le32_to_cpu(ni->type),
432 ntfs_debug("Cannot perform write to "
434 "attribute type 0x%x, "
435 "because there is not "
437 vi->i_ino, (unsigned)
438 le32_to_cpu(ni->type));
444 * If the write starts beyond the initialized size, extend it up to the
445 * beginning of the write and initialize all non-sparse space between
446 * the old initialized size and the new one. This automatically also
447 * increments the vfs inode->i_size to keep it above or equal to the
450 read_lock_irqsave(&ni->size_lock, flags);
451 ll = ni->initialized_size;
452 read_unlock_irqrestore(&ni->size_lock, flags);
455 * Wait for ongoing direct i/o to complete before proceeding.
456 * New direct i/o cannot start as we hold i_mutex.
459 err = ntfs_attr_extend_initialized(ni, pos);
460 if (unlikely(err < 0))
461 ntfs_error(vi->i_sb, "Cannot perform write to inode "
462 "0x%lx, attribute type 0x%x, because "
463 "extending the initialized size "
464 "failed (error %d).", vi->i_ino,
465 (unsigned)le32_to_cpu(ni->type),
473 * __ntfs_grab_cache_pages - obtain a number of locked pages
474 * @mapping: address space mapping from which to obtain page cache pages
475 * @index: starting index in @mapping at which to begin obtaining pages
476 * @nr_pages: number of page cache pages to obtain
477 * @pages: array of pages in which to return the obtained page cache pages
478 * @cached_page: allocated but as yet unused page
480 * Obtain @nr_pages locked page cache pages from the mapping @mapping and
481 * starting at index @index.
483 * If a page is newly created, add it to lru list
485 * Note, the page locks are obtained in ascending page index order.
487 static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
488 pgoff_t index, const unsigned nr_pages, struct page **pages,
489 struct page **cached_page)
496 pages[nr] = find_get_page_flags(mapping, index, FGP_LOCK |
500 *cached_page = page_cache_alloc(mapping);
501 if (unlikely(!*cached_page)) {
506 err = add_to_page_cache_lru(*cached_page, mapping,
508 mapping_gfp_constraint(mapping, GFP_KERNEL));
514 pages[nr] = *cached_page;
519 } while (nr < nr_pages);
524 unlock_page(pages[--nr]);
530 static inline void ntfs_submit_bh_for_read(struct buffer_head *bh)
534 bh->b_end_io = end_buffer_read_sync;
535 submit_bh(REQ_OP_READ, bh);
539 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
540 * @pages: array of destination pages
541 * @nr_pages: number of pages in @pages
542 * @pos: byte position in file at which the write begins
543 * @bytes: number of bytes to be written
545 * This is called for non-resident attributes from ntfs_file_buffered_write()
546 * with i_mutex held on the inode (@pages[0]->mapping->host). There are
547 * @nr_pages pages in @pages which are locked but not kmap()ped. The source
548 * data has not yet been copied into the @pages.
550 * Need to fill any holes with actual clusters, allocate buffers if necessary,
551 * ensure all the buffers are mapped, and bring uptodate any buffers that are
552 * only partially being written to.
554 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
555 * greater than PAGE_SIZE, that all pages in @pages are entirely inside
556 * the same cluster and that they are the entirety of that cluster, and that
557 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
559 * i_size is not to be modified yet.
561 * Return 0 on success or -errno on error.
563 static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
564 unsigned nr_pages, s64 pos, size_t bytes)
566 VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
568 s64 bh_pos, vcn_len, end, initialized_size;
572 ntfs_inode *ni, *base_ni = NULL;
574 runlist_element *rl, *rl2;
575 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
576 ntfs_attr_search_ctx *ctx = NULL;
577 MFT_RECORD *m = NULL;
578 ATTR_RECORD *a = NULL;
580 u32 attr_rec_len = 0;
581 unsigned blocksize, u;
583 bool rl_write_locked, was_hole, is_retry;
584 unsigned char blocksize_bits;
587 u8 mft_attr_mapped:1;
590 } status = { 0, 0, 0, 0 };
595 vi = pages[0]->mapping->host;
598 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
599 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
600 vi->i_ino, ni->type, pages[0]->index, nr_pages,
601 (long long)pos, bytes);
602 blocksize = vol->sb->s_blocksize;
603 blocksize_bits = vol->sb->s_blocksize_bits;
604 rl_write_locked = false;
611 cpos = pos >> vol->cluster_size_bits;
613 cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
615 * Loop over each buffer in each folio. Use goto to
616 * reduce indentation.
620 folio = page_folio(pages[u]);
621 bh_pos = folio_pos(folio);
622 head = folio_buffers(folio);
625 * create_empty_buffers() will create uptodate/dirty
626 * buffers if the folio is uptodate/dirty.
628 head = create_empty_buffers(folio, blocksize, 0);
635 /* Clear buffer_new on all buffers to reinitialise state. */
637 clear_buffer_new(bh);
638 bh_end = bh_pos + blocksize;
639 bh_cpos = bh_pos >> vol->cluster_size_bits;
640 bh_cofs = bh_pos & vol->cluster_size_mask;
641 if (buffer_mapped(bh)) {
643 * The buffer is already mapped. If it is uptodate,
646 if (buffer_uptodate(bh))
649 * The buffer is not uptodate. If the folio is uptodate
650 * set the buffer uptodate and otherwise ignore it.
652 if (folio_test_uptodate(folio)) {
653 set_buffer_uptodate(bh);
657 * Neither the folio nor the buffer are uptodate. If
658 * the buffer is only partially being written to, we
659 * need to read it in before the write, i.e. now.
661 if ((bh_pos < pos && bh_end > pos) ||
662 (bh_pos < end && bh_end > end)) {
664 * If the buffer is fully or partially within
665 * the initialized size, do an actual read.
666 * Otherwise, simply zero the buffer.
668 read_lock_irqsave(&ni->size_lock, flags);
669 initialized_size = ni->initialized_size;
670 read_unlock_irqrestore(&ni->size_lock, flags);
671 if (bh_pos < initialized_size) {
672 ntfs_submit_bh_for_read(bh);
675 folio_zero_range(folio, bh_offset(bh),
677 set_buffer_uptodate(bh);
682 /* Unmapped buffer. Need to map it. */
683 bh->b_bdev = vol->sb->s_bdev;
685 * If the current buffer is in the same clusters as the map
686 * cache, there is no need to check the runlist again. The
687 * map cache is made up of @vcn, which is the first cached file
688 * cluster, @vcn_len which is the number of cached file
689 * clusters, @lcn is the device cluster corresponding to @vcn,
690 * and @lcn_block is the block number corresponding to @lcn.
692 cdelta = bh_cpos - vcn;
693 if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
696 bh->b_blocknr = lcn_block +
697 (cdelta << (vol->cluster_size_bits -
699 (bh_cofs >> blocksize_bits);
700 set_buffer_mapped(bh);
702 * If the folio is uptodate so is the buffer. If the
703 * buffer is fully outside the write, we ignore it if
704 * it was already allocated and we mark it dirty so it
705 * gets written out if we allocated it. On the other
706 * hand, if we allocated the buffer but we are not
707 * marking it dirty we set buffer_new so we can do
710 if (folio_test_uptodate(folio)) {
711 if (!buffer_uptodate(bh))
712 set_buffer_uptodate(bh);
713 if (unlikely(was_hole)) {
714 /* We allocated the buffer. */
715 clean_bdev_bh_alias(bh);
716 if (bh_end <= pos || bh_pos >= end)
717 mark_buffer_dirty(bh);
723 /* Page is _not_ uptodate. */
724 if (likely(!was_hole)) {
726 * Buffer was already allocated. If it is not
727 * uptodate and is only partially being written
728 * to, we need to read it in before the write,
731 if (!buffer_uptodate(bh) && bh_pos < end &&
736 * If the buffer is fully or partially
737 * within the initialized size, do an
738 * actual read. Otherwise, simply zero
741 read_lock_irqsave(&ni->size_lock,
743 initialized_size = ni->initialized_size;
744 read_unlock_irqrestore(&ni->size_lock,
746 if (bh_pos < initialized_size) {
747 ntfs_submit_bh_for_read(bh);
750 folio_zero_range(folio,
753 set_buffer_uptodate(bh);
758 /* We allocated the buffer. */
759 clean_bdev_bh_alias(bh);
761 * If the buffer is fully outside the write, zero it,
762 * set it uptodate, and mark it dirty so it gets
763 * written out. If it is partially being written to,
764 * zero region surrounding the write but leave it to
765 * commit write to do anything else. Finally, if the
766 * buffer is fully being overwritten, do nothing.
768 if (bh_end <= pos || bh_pos >= end) {
769 if (!buffer_uptodate(bh)) {
770 folio_zero_range(folio, bh_offset(bh),
772 set_buffer_uptodate(bh);
774 mark_buffer_dirty(bh);
778 if (!buffer_uptodate(bh) &&
779 (bh_pos < pos || bh_end > end)) {
783 kaddr = kmap_local_folio(folio, 0);
785 pofs = bh_pos & ~PAGE_MASK;
786 memset(kaddr + pofs, 0, pos - bh_pos);
789 pofs = end & ~PAGE_MASK;
790 memset(kaddr + pofs, 0, bh_end - end);
793 flush_dcache_folio(folio);
798 * Slow path: this is the first buffer in the cluster. If it
799 * is outside allocated size and is not uptodate, zero it and
802 read_lock_irqsave(&ni->size_lock, flags);
803 initialized_size = ni->allocated_size;
804 read_unlock_irqrestore(&ni->size_lock, flags);
805 if (bh_pos > initialized_size) {
806 if (folio_test_uptodate(folio)) {
807 if (!buffer_uptodate(bh))
808 set_buffer_uptodate(bh);
809 } else if (!buffer_uptodate(bh)) {
810 folio_zero_range(folio, bh_offset(bh),
812 set_buffer_uptodate(bh);
818 down_read(&ni->runlist.lock);
822 if (likely(rl != NULL)) {
823 /* Seek to element containing target cluster. */
824 while (rl->length && rl[1].vcn <= bh_cpos)
826 lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
827 if (likely(lcn >= 0)) {
829 * Successful remap, setup the map cache and
830 * use that to deal with the buffer.
834 vcn_len = rl[1].vcn - vcn;
835 lcn_block = lcn << (vol->cluster_size_bits -
839 * If the number of remaining clusters touched
840 * by the write is smaller or equal to the
841 * number of cached clusters, unlock the
842 * runlist as the map cache will be used from
845 if (likely(vcn + vcn_len >= cend)) {
846 if (rl_write_locked) {
847 up_write(&ni->runlist.lock);
848 rl_write_locked = false;
850 up_read(&ni->runlist.lock);
853 goto map_buffer_cached;
856 lcn = LCN_RL_NOT_MAPPED;
858 * If it is not a hole and not out of bounds, the runlist is
859 * probably unmapped so try to map it now.
861 if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
862 if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
863 /* Attempt to map runlist. */
864 if (!rl_write_locked) {
866 * We need the runlist locked for
867 * writing, so if it is locked for
868 * reading relock it now and retry in
869 * case it changed whilst we dropped
872 up_read(&ni->runlist.lock);
873 down_write(&ni->runlist.lock);
874 rl_write_locked = true;
877 err = ntfs_map_runlist_nolock(ni, bh_cpos,
884 * If @vcn is out of bounds, pretend @lcn is
885 * LCN_ENOENT. As long as the buffer is out
886 * of bounds this will work fine.
888 if (err == -ENOENT) {
891 goto rl_not_mapped_enoent;
895 /* Failed to map the buffer, even after retrying. */
897 ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
898 "attribute type 0x%x, vcn 0x%llx, "
899 "vcn offset 0x%x, because its "
900 "location on disk could not be "
901 "determined%s (error code %i).",
902 ni->mft_no, ni->type,
903 (unsigned long long)bh_cpos,
905 vol->cluster_size_mask,
906 is_retry ? " even after retrying" : "",
910 rl_not_mapped_enoent:
912 * The buffer is in a hole or out of bounds. We need to fill
913 * the hole, unless the buffer is in a cluster which is not
914 * touched by the write, in which case we just leave the buffer
915 * unmapped. This can only happen when the cluster size is
916 * less than the page cache size.
918 if (unlikely(vol->cluster_size < PAGE_SIZE)) {
919 bh_cend = (bh_end + vol->cluster_size - 1) >>
920 vol->cluster_size_bits;
921 if ((bh_cend <= cpos || bh_cpos >= cend)) {
924 * If the buffer is uptodate we skip it. If it
925 * is not but the folio is uptodate, we can set
926 * the buffer uptodate. If the folio is not
927 * uptodate, we can clear the buffer and set it
928 * uptodate. Whether this is worthwhile is
929 * debatable and this could be removed.
931 if (folio_test_uptodate(folio)) {
932 if (!buffer_uptodate(bh))
933 set_buffer_uptodate(bh);
934 } else if (!buffer_uptodate(bh)) {
935 folio_zero_range(folio, bh_offset(bh),
937 set_buffer_uptodate(bh);
943 * Out of bounds buffer is invalid if it was not really out of
946 BUG_ON(lcn != LCN_HOLE);
948 * We need the runlist locked for writing, so if it is locked
949 * for reading relock it now and retry in case it changed
950 * whilst we dropped the lock.
953 if (!rl_write_locked) {
954 up_read(&ni->runlist.lock);
955 down_write(&ni->runlist.lock);
956 rl_write_locked = true;
959 /* Find the previous last allocated cluster. */
960 BUG_ON(rl->lcn != LCN_HOLE);
963 while (--rl2 >= ni->runlist.rl) {
965 lcn = rl2->lcn + rl2->length;
969 rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
973 ntfs_debug("Failed to allocate cluster, error code %i.",
978 rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
983 if (ntfs_cluster_free_from_rl(vol, rl2)) {
984 ntfs_error(vol->sb, "Failed to release "
985 "allocated cluster in error "
986 "code path. Run chkdsk to "
987 "recover the lost cluster.");
994 status.runlist_merged = 1;
995 ntfs_debug("Allocated cluster, lcn 0x%llx.",
996 (unsigned long long)lcn);
997 /* Map and lock the mft record and get the attribute record. */
1001 base_ni = ni->ext.base_ntfs_ino;
1002 m = map_mft_record(base_ni);
1007 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1008 if (unlikely(!ctx)) {
1010 unmap_mft_record(base_ni);
1013 status.mft_attr_mapped = 1;
1014 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1015 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
1016 if (unlikely(err)) {
1024 * Find the runlist element with which the attribute extent
1025 * starts. Note, we cannot use the _attr_ version because we
1026 * have mapped the mft record. That is ok because we know the
1027 * runlist fragment must be mapped already to have ever gotten
1028 * here, so we can just use the _rl_ version.
1030 vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
1031 rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
1033 BUG_ON(!rl2->length);
1034 BUG_ON(rl2->lcn < LCN_HOLE);
1035 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
1037 * If @highest_vcn is zero, calculate the real highest_vcn
1038 * (which can really be zero).
1041 highest_vcn = (sle64_to_cpu(
1042 a->data.non_resident.allocated_size) >>
1043 vol->cluster_size_bits) - 1;
1045 * Determine the size of the mapping pairs array for the new
1046 * extent, i.e. the old extent with the hole filled.
1048 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
1050 if (unlikely(mp_size <= 0)) {
1051 if (!(err = mp_size))
1053 ntfs_debug("Failed to get size for mapping pairs "
1054 "array, error code %i.", err);
1058 * Resize the attribute record to fit the new mapping pairs
1061 attr_rec_len = le32_to_cpu(a->length);
1062 err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
1063 a->data.non_resident.mapping_pairs_offset));
1064 if (unlikely(err)) {
1065 BUG_ON(err != -ENOSPC);
1066 // TODO: Deal with this by using the current attribute
1067 // and fill it with as much of the mapping pairs
1068 // array as possible. Then loop over each attribute
1069 // extent rewriting the mapping pairs arrays as we go
1070 // along and if when we reach the end we have not
1071 // enough space, try to resize the last attribute
1072 // extent and if even that fails, add a new attribute
1074 // We could also try to resize at each step in the hope
1075 // that we will not need to rewrite every single extent.
1076 // Note, we may need to decompress some extents to fill
1077 // the runlist as we are walking the extents...
1078 ntfs_error(vol->sb, "Not enough space in the mft "
1079 "record for the extended attribute "
1080 "record. This case is not "
1081 "implemented yet.");
1085 status.mp_rebuilt = 1;
1087 * Generate the mapping pairs array directly into the attribute
1090 err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1091 a->data.non_resident.mapping_pairs_offset),
1092 mp_size, rl2, vcn, highest_vcn, NULL);
1093 if (unlikely(err)) {
1094 ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
1095 "attribute type 0x%x, because building "
1096 "the mapping pairs failed with error "
1097 "code %i.", vi->i_ino,
1098 (unsigned)le32_to_cpu(ni->type), err);
1102 /* Update the highest_vcn but only if it was not set. */
1103 if (unlikely(!a->data.non_resident.highest_vcn))
1104 a->data.non_resident.highest_vcn =
1105 cpu_to_sle64(highest_vcn);
1107 * If the attribute is sparse/compressed, update the compressed
1108 * size in the ntfs_inode structure and the attribute record.
1110 if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
1112 * If we are not in the first attribute extent, switch
1113 * to it, but first ensure the changes will make it to
1116 if (a->data.non_resident.lowest_vcn) {
1117 flush_dcache_mft_record_page(ctx->ntfs_ino);
1118 mark_mft_record_dirty(ctx->ntfs_ino);
1119 ntfs_attr_reinit_search_ctx(ctx);
1120 err = ntfs_attr_lookup(ni->type, ni->name,
1121 ni->name_len, CASE_SENSITIVE,
1123 if (unlikely(err)) {
1124 status.attr_switched = 1;
1127 /* @m is not used any more so do not set it. */
1130 write_lock_irqsave(&ni->size_lock, flags);
1131 ni->itype.compressed.size += vol->cluster_size;
1132 a->data.non_resident.compressed_size =
1133 cpu_to_sle64(ni->itype.compressed.size);
1134 write_unlock_irqrestore(&ni->size_lock, flags);
1136 /* Ensure the changes make it to disk. */
1137 flush_dcache_mft_record_page(ctx->ntfs_ino);
1138 mark_mft_record_dirty(ctx->ntfs_ino);
1139 ntfs_attr_put_search_ctx(ctx);
1140 unmap_mft_record(base_ni);
1141 /* Successfully filled the hole. */
1142 status.runlist_merged = 0;
1143 status.mft_attr_mapped = 0;
1144 status.mp_rebuilt = 0;
1145 /* Setup the map cache and use that to deal with the buffer. */
1149 lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
1152 * If the number of remaining clusters in the @pages is smaller
1153 * or equal to the number of cached clusters, unlock the
1154 * runlist as the map cache will be used from now on.
1156 if (likely(vcn + vcn_len >= cend)) {
1157 up_write(&ni->runlist.lock);
1158 rl_write_locked = false;
1161 goto map_buffer_cached;
1162 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1163 /* If there are no errors, do the next page. */
1164 if (likely(!err && ++u < nr_pages))
1166 /* If there are no errors, release the runlist lock if we took it. */
1168 if (unlikely(rl_write_locked)) {
1169 up_write(&ni->runlist.lock);
1170 rl_write_locked = false;
1171 } else if (unlikely(rl))
1172 up_read(&ni->runlist.lock);
1175 /* If we issued read requests, let them complete. */
1176 read_lock_irqsave(&ni->size_lock, flags);
1177 initialized_size = ni->initialized_size;
1178 read_unlock_irqrestore(&ni->size_lock, flags);
1179 while (wait_bh > wait) {
1182 if (likely(buffer_uptodate(bh))) {
1183 folio = bh->b_folio;
1184 bh_pos = folio_pos(folio) + bh_offset(bh);
1186 * If the buffer overflows the initialized size, need
1187 * to zero the overflowing region.
1189 if (unlikely(bh_pos + blocksize > initialized_size)) {
1192 if (likely(bh_pos < initialized_size))
1193 ofs = initialized_size - bh_pos;
1194 folio_zero_segment(folio, bh_offset(bh) + ofs,
1197 } else /* if (unlikely(!buffer_uptodate(bh))) */
1201 /* Clear buffer_new on all buffers. */
1204 bh = head = page_buffers(pages[u]);
1207 clear_buffer_new(bh);
1208 } while ((bh = bh->b_this_page) != head);
1209 } while (++u < nr_pages);
1210 ntfs_debug("Done.");
1213 if (status.attr_switched) {
1214 /* Get back to the attribute extent we modified. */
1215 ntfs_attr_reinit_search_ctx(ctx);
1216 if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1217 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
1218 ntfs_error(vol->sb, "Failed to find required "
1219 "attribute extent of attribute in "
1220 "error code path. Run chkdsk to "
1222 write_lock_irqsave(&ni->size_lock, flags);
1223 ni->itype.compressed.size += vol->cluster_size;
1224 write_unlock_irqrestore(&ni->size_lock, flags);
1225 flush_dcache_mft_record_page(ctx->ntfs_ino);
1226 mark_mft_record_dirty(ctx->ntfs_ino);
1228 * The only thing that is now wrong is the compressed
1229 * size of the base attribute extent which chkdsk
1230 * should be able to fix.
1236 status.attr_switched = 0;
1240 * If the runlist has been modified, need to restore it by punching a
1241 * hole into it and we then need to deallocate the on-disk cluster as
1242 * well. Note, we only modify the runlist if we are able to generate a
1243 * new mapping pairs array, i.e. only when the mapped attribute extent
1246 if (status.runlist_merged && !status.attr_switched) {
1247 BUG_ON(!rl_write_locked);
1248 /* Make the file cluster we allocated sparse in the runlist. */
1249 if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
1250 ntfs_error(vol->sb, "Failed to punch hole into "
1251 "attribute runlist in error code "
1252 "path. Run chkdsk to recover the "
1255 } else /* if (success) */ {
1256 status.runlist_merged = 0;
1258 * Deallocate the on-disk cluster we allocated but only
1259 * if we succeeded in punching its vcn out of the
1262 down_write(&vol->lcnbmp_lock);
1263 if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1264 ntfs_error(vol->sb, "Failed to release "
1265 "allocated cluster in error "
1266 "code path. Run chkdsk to "
1267 "recover the lost cluster.");
1270 up_write(&vol->lcnbmp_lock);
1274 * Resize the attribute record to its old size and rebuild the mapping
1275 * pairs array. Note, we only can do this if the runlist has been
1276 * restored to its old state which also implies that the mapped
1277 * attribute extent is not switched.
1279 if (status.mp_rebuilt && !status.runlist_merged) {
1280 if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
1281 ntfs_error(vol->sb, "Failed to restore attribute "
1282 "record in error code path. Run "
1283 "chkdsk to recover.");
1285 } else /* if (success) */ {
1286 if (ntfs_mapping_pairs_build(vol, (u8*)a +
1287 le16_to_cpu(a->data.non_resident.
1288 mapping_pairs_offset), attr_rec_len -
1289 le16_to_cpu(a->data.non_resident.
1290 mapping_pairs_offset), ni->runlist.rl,
1291 vcn, highest_vcn, NULL)) {
1292 ntfs_error(vol->sb, "Failed to restore "
1293 "mapping pairs array in error "
1294 "code path. Run chkdsk to "
1298 flush_dcache_mft_record_page(ctx->ntfs_ino);
1299 mark_mft_record_dirty(ctx->ntfs_ino);
1302 /* Release the mft record and the attribute. */
1303 if (status.mft_attr_mapped) {
1304 ntfs_attr_put_search_ctx(ctx);
1305 unmap_mft_record(base_ni);
1307 /* Release the runlist lock. */
1308 if (rl_write_locked)
1309 up_write(&ni->runlist.lock);
1311 up_read(&ni->runlist.lock);
1313 * Zero out any newly allocated blocks to avoid exposing stale data.
1314 * If BH_New is set, we know that the block was newly allocated above
1315 * and that it has not been fully zeroed and marked dirty yet.
1319 end = bh_cpos << vol->cluster_size_bits;
1321 folio = page_folio(pages[u]);
1322 bh = head = folio_buffers(folio);
1324 if (u == nr_pages &&
1325 folio_pos(folio) + bh_offset(bh) >= end)
1327 if (!buffer_new(bh))
1329 clear_buffer_new(bh);
1330 if (!buffer_uptodate(bh)) {
1331 if (folio_test_uptodate(folio))
1332 set_buffer_uptodate(bh);
1334 folio_zero_range(folio, bh_offset(bh),
1336 set_buffer_uptodate(bh);
1339 mark_buffer_dirty(bh);
1340 } while ((bh = bh->b_this_page) != head);
1341 } while (++u <= nr_pages);
1342 ntfs_error(vol->sb, "Failed. Returning error code %i.", err);
1346 static inline void ntfs_flush_dcache_pages(struct page **pages,
1351 * Warning: Do not do the decrement at the same time as the call to
1352 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1353 * decrement never happens so the loop never terminates.
1357 flush_dcache_page(pages[nr_pages]);
1358 } while (nr_pages > 0);
1362 * ntfs_commit_pages_after_non_resident_write - commit the received data
1363 * @pages: array of destination pages
1364 * @nr_pages: number of pages in @pages
1365 * @pos: byte position in file at which the write begins
1366 * @bytes: number of bytes to be written
1368 * See description of ntfs_commit_pages_after_write(), below.
1370 static inline int ntfs_commit_pages_after_non_resident_write(
1371 struct page **pages, const unsigned nr_pages,
1372 s64 pos, size_t bytes)
1374 s64 end, initialized_size;
1376 ntfs_inode *ni, *base_ni;
1377 struct buffer_head *bh, *head;
1378 ntfs_attr_search_ctx *ctx;
1381 unsigned long flags;
1382 unsigned blocksize, u;
1385 vi = pages[0]->mapping->host;
1387 blocksize = vi->i_sb->s_blocksize;
1396 bh_pos = (s64)page->index << PAGE_SHIFT;
1397 bh = head = page_buffers(page);
1402 bh_end = bh_pos + blocksize;
1403 if (bh_end <= pos || bh_pos >= end) {
1404 if (!buffer_uptodate(bh))
1407 set_buffer_uptodate(bh);
1408 mark_buffer_dirty(bh);
1410 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1412 * If all buffers are now uptodate but the page is not, set the
1415 if (!partial && !PageUptodate(page))
1416 SetPageUptodate(page);
1417 } while (++u < nr_pages);
1419 * Finally, if we do not need to update initialized_size or i_size we
1422 read_lock_irqsave(&ni->size_lock, flags);
1423 initialized_size = ni->initialized_size;
1424 read_unlock_irqrestore(&ni->size_lock, flags);
1425 if (end <= initialized_size) {
1426 ntfs_debug("Done.");
1430 * Update initialized_size/i_size as appropriate, both in the inode and
1436 base_ni = ni->ext.base_ntfs_ino;
1437 /* Map, pin, and lock the mft record. */
1438 m = map_mft_record(base_ni);
1445 BUG_ON(!NInoNonResident(ni));
1446 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1447 if (unlikely(!ctx)) {
1451 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1452 CASE_SENSITIVE, 0, NULL, 0, ctx);
1453 if (unlikely(err)) {
1459 BUG_ON(!a->non_resident);
1460 write_lock_irqsave(&ni->size_lock, flags);
1461 BUG_ON(end > ni->allocated_size);
1462 ni->initialized_size = end;
1463 a->data.non_resident.initialized_size = cpu_to_sle64(end);
1464 if (end > i_size_read(vi)) {
1465 i_size_write(vi, end);
1466 a->data.non_resident.data_size =
1467 a->data.non_resident.initialized_size;
1469 write_unlock_irqrestore(&ni->size_lock, flags);
1470 /* Mark the mft record dirty, so it gets written back. */
1471 flush_dcache_mft_record_page(ctx->ntfs_ino);
1472 mark_mft_record_dirty(ctx->ntfs_ino);
1473 ntfs_attr_put_search_ctx(ctx);
1474 unmap_mft_record(base_ni);
1475 ntfs_debug("Done.");
1479 ntfs_attr_put_search_ctx(ctx);
1481 unmap_mft_record(base_ni);
1482 ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
1485 NVolSetErrors(ni->vol);
1490 * ntfs_commit_pages_after_write - commit the received data
1491 * @pages: array of destination pages
1492 * @nr_pages: number of pages in @pages
1493 * @pos: byte position in file at which the write begins
1494 * @bytes: number of bytes to be written
1496 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1497 * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are
1498 * locked but not kmap()ped. The source data has already been copied into the
1499 * @page. ntfs_prepare_pages_for_non_resident_write() has been called before
1500 * the data was copied (for non-resident attributes only) and it returned
1503 * Need to set uptodate and mark dirty all buffers within the boundary of the
1504 * write. If all buffers in a page are uptodate we set the page uptodate, too.
1506 * Setting the buffers dirty ensures that they get written out later when
1507 * ntfs_writepage() is invoked by the VM.
1509 * Finally, we need to update i_size and initialized_size as appropriate both
1510 * in the inode and the mft record.
1512 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1513 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1514 * page are uptodate, and updates i_size if the end of io is beyond i_size. In
1515 * that case, it also marks the inode dirty.
1517 * If things have gone as outlined in
1518 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1519 * content modifications here for non-resident attributes. For resident
1520 * attributes we need to do the uptodate bringing here which we combine with
1521 * the copying into the mft record which means we save one atomic kmap.
1523 * Return 0 on success or -errno on error.
1525 static int ntfs_commit_pages_after_write(struct page **pages,
1526 const unsigned nr_pages, s64 pos, size_t bytes)
1528 s64 end, initialized_size;
1531 ntfs_inode *ni, *base_ni;
1533 ntfs_attr_search_ctx *ctx;
1536 char *kattr, *kaddr;
1537 unsigned long flags;
1545 vi = page->mapping->host;
1547 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
1548 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.",
1549 vi->i_ino, ni->type, page->index, nr_pages,
1550 (long long)pos, bytes);
1551 if (NInoNonResident(ni))
1552 return ntfs_commit_pages_after_non_resident_write(pages,
1553 nr_pages, pos, bytes);
1554 BUG_ON(nr_pages > 1);
1556 * Attribute is resident, implying it is not compressed, encrypted, or
1562 base_ni = ni->ext.base_ntfs_ino;
1563 BUG_ON(NInoNonResident(ni));
1564 /* Map, pin, and lock the mft record. */
1565 m = map_mft_record(base_ni);
1572 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1573 if (unlikely(!ctx)) {
1577 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1578 CASE_SENSITIVE, 0, NULL, 0, ctx);
1579 if (unlikely(err)) {
1585 BUG_ON(a->non_resident);
1586 /* The total length of the attribute value. */
1587 attr_len = le32_to_cpu(a->data.resident.value_length);
1588 i_size = i_size_read(vi);
1589 BUG_ON(attr_len != i_size);
1590 BUG_ON(pos > attr_len);
1592 BUG_ON(end > le32_to_cpu(a->length) -
1593 le16_to_cpu(a->data.resident.value_offset));
1594 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
1595 kaddr = kmap_atomic(page);
1596 /* Copy the received data from the page to the mft record. */
1597 memcpy(kattr + pos, kaddr + pos, bytes);
1598 /* Update the attribute length if necessary. */
1599 if (end > attr_len) {
1601 a->data.resident.value_length = cpu_to_le32(attr_len);
1604 * If the page is not uptodate, bring the out of bounds area(s)
1605 * uptodate by copying data from the mft record to the page.
1607 if (!PageUptodate(page)) {
1609 memcpy(kaddr, kattr, pos);
1611 memcpy(kaddr + end, kattr + end, attr_len - end);
1612 /* Zero the region outside the end of the attribute value. */
1613 memset(kaddr + attr_len, 0, PAGE_SIZE - attr_len);
1614 flush_dcache_page(page);
1615 SetPageUptodate(page);
1617 kunmap_atomic(kaddr);
1618 /* Update initialized_size/i_size if necessary. */
1619 read_lock_irqsave(&ni->size_lock, flags);
1620 initialized_size = ni->initialized_size;
1621 BUG_ON(end > ni->allocated_size);
1622 read_unlock_irqrestore(&ni->size_lock, flags);
1623 BUG_ON(initialized_size != i_size);
1624 if (end > initialized_size) {
1625 write_lock_irqsave(&ni->size_lock, flags);
1626 ni->initialized_size = end;
1627 i_size_write(vi, end);
1628 write_unlock_irqrestore(&ni->size_lock, flags);
1630 /* Mark the mft record dirty, so it gets written back. */
1631 flush_dcache_mft_record_page(ctx->ntfs_ino);
1632 mark_mft_record_dirty(ctx->ntfs_ino);
1633 ntfs_attr_put_search_ctx(ctx);
1634 unmap_mft_record(base_ni);
1635 ntfs_debug("Done.");
1638 if (err == -ENOMEM) {
1639 ntfs_warning(vi->i_sb, "Error allocating memory required to "
1640 "commit the write.");
1641 if (PageUptodate(page)) {
1642 ntfs_warning(vi->i_sb, "Page is uptodate, setting "
1643 "dirty so the write will be retried "
1644 "later on by the VM.");
1646 * Put the page on mapping->dirty_pages, but leave its
1647 * buffers' dirty state as-is.
1649 __set_page_dirty_nobuffers(page);
1652 ntfs_error(vi->i_sb, "Page is not uptodate. Written "
1653 "data has been lost.");
1655 ntfs_error(vi->i_sb, "Resident attribute commit write failed "
1656 "with error %i.", err);
1657 NVolSetErrors(ni->vol);
1660 ntfs_attr_put_search_ctx(ctx);
1662 unmap_mft_record(base_ni);
1667 * Copy as much as we can into the pages and return the number of bytes which
1668 * were successfully copied. If a fault is encountered then clear the pages
1669 * out to (ofs + bytes) and return the number of bytes which were copied.
1671 static size_t ntfs_copy_from_user_iter(struct page **pages, unsigned nr_pages,
1672 unsigned ofs, struct iov_iter *i, size_t bytes)
1674 struct page **last_page = pages + nr_pages;
1676 unsigned len, copied;
1679 len = PAGE_SIZE - ofs;
1682 copied = copy_page_from_iter_atomic(*pages, ofs, len, i);
1690 } while (++pages < last_page);
1694 /* Zero the rest of the target like __copy_from_user(). */
1695 len = PAGE_SIZE - copied;
1699 zero_user(*pages, copied, len);
1703 } while (++pages < last_page);
1708 * ntfs_perform_write - perform buffered write to a file
1709 * @file: file to write to
1710 * @i: iov_iter with data to write
1711 * @pos: byte offset in file at which to begin writing to
1713 static ssize_t ntfs_perform_write(struct file *file, struct iov_iter *i,
1716 struct address_space *mapping = file->f_mapping;
1717 struct inode *vi = mapping->host;
1718 ntfs_inode *ni = NTFS_I(vi);
1719 ntfs_volume *vol = ni->vol;
1720 struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER];
1721 struct page *cached_page = NULL;
1725 ssize_t status, written = 0;
1728 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, pos "
1729 "0x%llx, count 0x%lx.", vi->i_ino,
1730 (unsigned)le32_to_cpu(ni->type),
1731 (unsigned long long)pos,
1732 (unsigned long)iov_iter_count(i));
1734 * If a previous ntfs_truncate() failed, repeat it and abort if it
1737 if (unlikely(NInoTruncateFailed(ni))) {
1741 err = ntfs_truncate(vi);
1742 if (err || NInoTruncateFailed(ni)) {
1745 ntfs_error(vol->sb, "Cannot perform write to inode "
1746 "0x%lx, attribute type 0x%x, because "
1747 "ntfs_truncate() failed (error code "
1749 (unsigned)le32_to_cpu(ni->type), err);
1754 * Determine the number of pages per cluster for non-resident
1758 if (vol->cluster_size > PAGE_SIZE && NInoNonResident(ni))
1759 nr_pages = vol->cluster_size >> PAGE_SHIFT;
1764 unsigned ofs, do_pages, u;
1767 start_idx = pos >> PAGE_SHIFT;
1768 ofs = pos & ~PAGE_MASK;
1769 bytes = PAGE_SIZE - ofs;
1772 vcn = pos >> vol->cluster_size_bits;
1773 if (vcn != last_vcn) {
1776 * Get the lcn of the vcn the write is in. If
1777 * it is a hole, need to lock down all pages in
1780 down_read(&ni->runlist.lock);
1781 lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
1782 vol->cluster_size_bits, false);
1783 up_read(&ni->runlist.lock);
1784 if (unlikely(lcn < LCN_HOLE)) {
1785 if (lcn == LCN_ENOMEM)
1789 ntfs_error(vol->sb, "Cannot "
1792 "attribute type 0x%x, "
1793 "because the attribute "
1795 vi->i_ino, (unsigned)
1796 le32_to_cpu(ni->type));
1800 if (lcn == LCN_HOLE) {
1801 start_idx = (pos & ~(s64)
1802 vol->cluster_size_mask)
1804 bytes = vol->cluster_size - (pos &
1805 vol->cluster_size_mask);
1806 do_pages = nr_pages;
1810 if (bytes > iov_iter_count(i))
1811 bytes = iov_iter_count(i);
1814 * Bring in the user page(s) that we will copy from _first_.
1815 * Otherwise there is a nasty deadlock on copying from the same
1816 * page(s) as we are writing to, without it/them being marked
1817 * up-to-date. Note, at present there is nothing to stop the
1818 * pages being swapped out between us bringing them into memory
1819 * and doing the actual copying.
1821 if (unlikely(fault_in_iov_iter_readable(i, bytes))) {
1825 /* Get and lock @do_pages starting at index @start_idx. */
1826 status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages,
1827 pages, &cached_page);
1828 if (unlikely(status))
1831 * For non-resident attributes, we need to fill any holes with
1832 * actual clusters and ensure all bufferes are mapped. We also
1833 * need to bring uptodate any buffers that are only partially
1836 if (NInoNonResident(ni)) {
1837 status = ntfs_prepare_pages_for_non_resident_write(
1838 pages, do_pages, pos, bytes);
1839 if (unlikely(status)) {
1841 unlock_page(pages[--do_pages]);
1842 put_page(pages[do_pages]);
1847 u = (pos >> PAGE_SHIFT) - pages[0]->index;
1848 copied = ntfs_copy_from_user_iter(pages + u, do_pages - u, ofs,
1850 ntfs_flush_dcache_pages(pages + u, do_pages - u);
1852 if (likely(copied == bytes)) {
1853 status = ntfs_commit_pages_after_write(pages, do_pages,
1857 unlock_page(pages[--do_pages]);
1858 put_page(pages[do_pages]);
1860 if (unlikely(status < 0)) {
1861 iov_iter_revert(i, copied);
1865 if (unlikely(copied < bytes)) {
1866 iov_iter_revert(i, copied);
1869 else if (bytes > PAGE_SIZE - ofs)
1870 bytes = PAGE_SIZE - ofs;
1875 balance_dirty_pages_ratelimited(mapping);
1876 if (fatal_signal_pending(current)) {
1880 } while (iov_iter_count(i));
1882 put_page(cached_page);
1883 ntfs_debug("Done. Returning %s (written 0x%lx, status %li).",
1884 written ? "written" : "status", (unsigned long)written,
1886 return written ? written : status;
1890 * ntfs_file_write_iter - simple wrapper for ntfs_file_write_iter_nolock()
1891 * @iocb: IO state structure
1892 * @from: iov_iter with data to write
1894 * Basically the same as generic_file_write_iter() except that it ends up
1895 * up calling ntfs_perform_write() instead of generic_perform_write() and that
1896 * O_DIRECT is not implemented.
1898 static ssize_t ntfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1900 struct file *file = iocb->ki_filp;
1901 struct inode *vi = file_inode(file);
1902 ssize_t written = 0;
1906 /* We can write back this queue in page reclaim. */
1907 err = ntfs_prepare_file_for_write(iocb, from);
1908 if (iov_iter_count(from) && !err)
1909 written = ntfs_perform_write(file, from, iocb->ki_pos);
1911 iocb->ki_pos += written;
1912 if (likely(written > 0))
1913 written = generic_write_sync(iocb, written);
1914 return written ? written : err;
1918 * ntfs_file_fsync - sync a file to disk
1919 * @filp: file to be synced
1920 * @datasync: if non-zero only flush user data and not metadata
1922 * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync
1923 * system calls. This function is inspired by fs/buffer.c::file_fsync().
1925 * If @datasync is false, write the mft record and all associated extent mft
1926 * records as well as the $DATA attribute and then sync the block device.
1928 * If @datasync is true and the attribute is non-resident, we skip the writing
1929 * of the mft record and all associated extent mft records (this might still
1930 * happen due to the write_inode_now() call).
1932 * Also, if @datasync is true, we do not wait on the inode to be written out
1933 * but we always wait on the page cache pages to be written out.
1935 * Locking: Caller must hold i_mutex on the inode.
1937 * TODO: We should probably also write all attribute/index inodes associated
1938 * with this inode but since we have no simple way of getting to them we ignore
1939 * this problem for now.
1941 static int ntfs_file_fsync(struct file *filp, loff_t start, loff_t end,
1944 struct inode *vi = filp->f_mapping->host;
1947 ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
1949 err = file_write_and_wait_range(filp, start, end);
1954 BUG_ON(S_ISDIR(vi->i_mode));
1955 if (!datasync || !NInoNonResident(NTFS_I(vi)))
1956 ret = __ntfs_write_inode(vi, 1);
1957 write_inode_now(vi, !datasync);
1959 * NOTE: If we were to use mapping->private_list (see ext2 and
1960 * fs/buffer.c) for dirty blocks then we could optimize the below to be
1961 * sync_mapping_buffers(vi->i_mapping).
1963 err = sync_blockdev(vi->i_sb->s_bdev);
1964 if (unlikely(err && !ret))
1967 ntfs_debug("Done.");
1969 ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error "
1970 "%u.", datasync ? "data" : "", vi->i_ino, -ret);
1975 #endif /* NTFS_RW */
1977 const struct file_operations ntfs_file_ops = {
1978 .llseek = generic_file_llseek,
1979 .read_iter = generic_file_read_iter,
1981 .write_iter = ntfs_file_write_iter,
1982 .fsync = ntfs_file_fsync,
1983 #endif /* NTFS_RW */
1984 .mmap = generic_file_mmap,
1985 .open = ntfs_file_open,
1986 .splice_read = filemap_splice_read,
1989 const struct inode_operations ntfs_file_inode_ops = {
1991 .setattr = ntfs_setattr,
1992 #endif /* NTFS_RW */
1995 const struct file_operations ntfs_empty_file_ops = {};
1997 const struct inode_operations ntfs_empty_inode_ops = {};