2 * file.c - NTFS kernel file operations. Part of the Linux-NTFS project.
4 * Copyright (c) 2001-2005 Anton Altaparmakov
6 * This program/include file is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as published
8 * by the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program/include file is distributed in the hope that it will be
12 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
13 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program (in the main directory of the Linux-NTFS
18 * distribution in the file COPYING); if not, write to the Free Software
19 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 #include <linux/buffer_head.h>
23 #include <linux/pagemap.h>
24 #include <linux/pagevec.h>
25 #include <linux/sched.h>
26 #include <linux/swap.h>
27 #include <linux/uio.h>
28 #include <linux/writeback.h>
31 #include <asm/uaccess.h>
43 * ntfs_file_open - called when an inode is about to be opened
44 * @vi: inode to be opened
45 * @filp: file structure describing the inode
47 * Limit file size to the page cache limit on architectures where unsigned long
48 * is 32-bits. This is the most we can do for now without overflowing the page
49 * cache page index. Doing it this way means we don't run into problems because
50 * of existing too large files. It would be better to allow the user to read
51 * the beginning of the file but I doubt very much anyone is going to hit this
52 * check on a 32-bit architecture, so there is no point in adding the extra
53 * complexity required to support this.
55 * On 64-bit architectures, the check is hopefully optimized away by the
58 * After the check passes, just call generic_file_open() to do its work.
60 static int ntfs_file_open(struct inode *vi, struct file *filp)
62 if (sizeof(unsigned long) < 8) {
63 if (i_size_read(vi) > MAX_LFS_FILESIZE)
66 return generic_file_open(vi, filp);
72 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
73 * @ni: ntfs inode of the attribute to extend
74 * @new_init_size: requested new initialized size in bytes
75 * @cached_page: store any allocated but unused page here
76 * @lru_pvec: lru-buffering pagevec of the caller
78 * Extend the initialized size of an attribute described by the ntfs inode @ni
79 * to @new_init_size bytes. This involves zeroing any non-sparse space between
80 * the old initialized size and @new_init_size both in the page cache and on
81 * disk (if relevant complete pages are zeroed in the page cache then these may
82 * simply be marked dirty for later writeout). There is one caveat and that is
83 * that if any uptodate page cache pages between the old initialized size and
84 * the smaller of @new_init_size and the file size (vfs inode->i_size) are in
85 * memory, these need to be marked dirty without being zeroed since they could
86 * be non-zero due to mmap() based writes.
88 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
89 * in the resident attribute case, it is tied to the initialized size and, in
90 * the non-resident attribute case, it may not fall below the initialized size.
92 * Note that if the attribute is resident, we do not need to touch the page
93 * cache at all. This is because if the page cache page is not uptodate we
94 * bring it uptodate later, when doing the write to the mft record since we
95 * then already have the page mapped. And if the page is uptodate, the
96 * non-initialized region will already have been zeroed when the page was
97 * brought uptodate and the region may in fact already have been overwritten
98 * with new data via mmap() based writes, so we cannot just zero it. And since
99 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
100 * is unspecified, we choose not to do zeroing and thus we do not need to touch
101 * the page at all. For a more detailed explanation see ntfs_truncate() which
102 * is in fs/ntfs/inode.c.
104 * @cached_page and @lru_pvec are just optimisations for dealing with multiple
107 * Return 0 on success and -errno on error. In the case that an error is
108 * encountered it is possible that the initialized size will already have been
109 * incremented some way towards @new_init_size but it is guaranteed that if
110 * this is the case, the necessary zeroing will also have happened and that all
111 * metadata is self-consistent.
113 * Locking: This function locks the mft record of the base ntfs inode and
114 * maintains the lock throughout execution of the function. This is required
115 * so that the initialized size of the attribute can be modified safely.
117 static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size,
118 struct page **cached_page, struct pagevec *lru_pvec)
122 pgoff_t index, end_index;
124 struct inode *vi = VFS_I(ni);
126 MFT_RECORD *m = NULL;
128 ntfs_attr_search_ctx *ctx = NULL;
129 struct address_space *mapping;
130 struct page *page = NULL;
135 read_lock_irqsave(&ni->size_lock, flags);
136 old_init_size = ni->initialized_size;
137 old_i_size = i_size_read(vi);
138 BUG_ON(new_init_size > ni->allocated_size);
139 read_unlock_irqrestore(&ni->size_lock, flags);
140 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
141 "old_initialized_size 0x%llx, "
142 "new_initialized_size 0x%llx, i_size 0x%llx.",
143 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
144 (unsigned long long)old_init_size,
145 (unsigned long long)new_init_size, old_i_size);
149 base_ni = ni->ext.base_ntfs_ino;
150 /* Use goto to reduce indentation and we need the label below anyway. */
151 if (NInoNonResident(ni))
152 goto do_non_resident_extend;
153 BUG_ON(old_init_size != old_i_size);
154 m = map_mft_record(base_ni);
160 ctx = ntfs_attr_get_search_ctx(base_ni, m);
161 if (unlikely(!ctx)) {
165 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
166 CASE_SENSITIVE, 0, NULL, 0, ctx);
174 BUG_ON(a->non_resident);
175 /* The total length of the attribute value. */
176 attr_len = le32_to_cpu(a->data.resident.value_length);
177 BUG_ON(old_i_size != (loff_t)attr_len);
179 * Do the zeroing in the mft record and update the attribute size in
182 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
183 memset(kattr + attr_len, 0, new_init_size - attr_len);
184 a->data.resident.value_length = cpu_to_le32((u32)new_init_size);
185 /* Finally, update the sizes in the vfs and ntfs inodes. */
186 write_lock_irqsave(&ni->size_lock, flags);
187 i_size_write(vi, new_init_size);
188 ni->initialized_size = new_init_size;
189 write_unlock_irqrestore(&ni->size_lock, flags);
191 do_non_resident_extend:
193 * If the new initialized size @new_init_size exceeds the current file
194 * size (vfs inode->i_size), we need to extend the file size to the
195 * new initialized size.
197 if (new_init_size > old_i_size) {
198 m = map_mft_record(base_ni);
204 ctx = ntfs_attr_get_search_ctx(base_ni, m);
205 if (unlikely(!ctx)) {
209 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
210 CASE_SENSITIVE, 0, NULL, 0, ctx);
218 BUG_ON(!a->non_resident);
219 BUG_ON(old_i_size != (loff_t)
220 sle64_to_cpu(a->data.non_resident.data_size));
221 a->data.non_resident.data_size = cpu_to_sle64(new_init_size);
222 flush_dcache_mft_record_page(ctx->ntfs_ino);
223 mark_mft_record_dirty(ctx->ntfs_ino);
224 /* Update the file size in the vfs inode. */
225 i_size_write(vi, new_init_size);
226 ntfs_attr_put_search_ctx(ctx);
228 unmap_mft_record(base_ni);
231 mapping = vi->i_mapping;
232 index = old_init_size >> PAGE_CACHE_SHIFT;
233 end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
236 * Read the page. If the page is not present, this will zero
237 * the uninitialized regions for us.
239 page = read_cache_page(mapping, index,
240 (filler_t*)mapping->a_ops->readpage, NULL);
245 wait_on_page_locked(page);
246 if (unlikely(!PageUptodate(page) || PageError(page))) {
247 page_cache_release(page);
252 * Update the initialized size in the ntfs inode. This is
253 * enough to make ntfs_writepage() work.
255 write_lock_irqsave(&ni->size_lock, flags);
256 ni->initialized_size = (index + 1) << PAGE_CACHE_SHIFT;
257 if (ni->initialized_size > new_init_size)
258 ni->initialized_size = new_init_size;
259 write_unlock_irqrestore(&ni->size_lock, flags);
260 /* Set the page dirty so it gets written out. */
261 set_page_dirty(page);
262 page_cache_release(page);
264 * Play nice with the vm and the rest of the system. This is
265 * very much needed as we can potentially be modifying the
266 * initialised size from a very small value to a really huge
268 * f = open(somefile, O_TRUNC);
269 * truncate(f, 10GiB);
272 * And this would mean we would be marking dirty hundreds of
273 * thousands of pages or as in the above example more than
274 * two and a half million pages!
276 * TODO: For sparse pages could optimize this workload by using
277 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This
278 * would be set in readpage for sparse pages and here we would
279 * not need to mark dirty any pages which have this bit set.
280 * The only caveat is that we have to clear the bit everywhere
281 * where we allocate any clusters that lie in the page or that
284 * TODO: An even greater optimization would be for us to only
285 * call readpage() on pages which are not in sparse regions as
286 * determined from the runlist. This would greatly reduce the
287 * number of pages we read and make dirty in the case of sparse
290 balance_dirty_pages_ratelimited(mapping);
292 } while (++index < end_index);
293 read_lock_irqsave(&ni->size_lock, flags);
294 BUG_ON(ni->initialized_size != new_init_size);
295 read_unlock_irqrestore(&ni->size_lock, flags);
296 /* Now bring in sync the initialized_size in the mft record. */
297 m = map_mft_record(base_ni);
303 ctx = ntfs_attr_get_search_ctx(base_ni, m);
304 if (unlikely(!ctx)) {
308 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
309 CASE_SENSITIVE, 0, NULL, 0, ctx);
317 BUG_ON(!a->non_resident);
318 a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size);
320 flush_dcache_mft_record_page(ctx->ntfs_ino);
321 mark_mft_record_dirty(ctx->ntfs_ino);
323 ntfs_attr_put_search_ctx(ctx);
325 unmap_mft_record(base_ni);
326 ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.",
327 (unsigned long long)new_init_size, i_size_read(vi));
330 write_lock_irqsave(&ni->size_lock, flags);
331 ni->initialized_size = old_init_size;
332 write_unlock_irqrestore(&ni->size_lock, flags);
335 ntfs_attr_put_search_ctx(ctx);
337 unmap_mft_record(base_ni);
338 ntfs_debug("Failed. Returning error code %i.", err);
343 * ntfs_fault_in_pages_readable -
345 * Fault a number of userspace pages into pagetables.
347 * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes
348 * with more than two userspace pages as well as handling the single page case
351 * If you find this difficult to understand, then think of the while loop being
352 * the following code, except that we do without the integer variable ret:
355 * ret = __get_user(c, uaddr);
356 * uaddr += PAGE_SIZE;
357 * } while (!ret && uaddr < end);
359 * Note, the final __get_user() may well run out-of-bounds of the user buffer,
360 * but _not_ out-of-bounds of the page the user buffer belongs to, and since
361 * this is only a read and not a write, and since it is still in the same page,
362 * it should not matter and this makes the code much simpler.
364 static inline void ntfs_fault_in_pages_readable(const char __user *uaddr,
367 const char __user *end;
370 /* Set @end to the first byte outside the last page we care about. */
371 end = (const char __user*)PAGE_ALIGN((ptrdiff_t __user)uaddr + bytes);
373 while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end))
378 * ntfs_fault_in_pages_readable_iovec -
380 * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs.
382 static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov,
383 size_t iov_ofs, int bytes)
386 const char __user *buf;
389 buf = iov->iov_base + iov_ofs;
390 len = iov->iov_len - iov_ofs;
393 ntfs_fault_in_pages_readable(buf, len);
401 * __ntfs_grab_cache_pages - obtain a number of locked pages
402 * @mapping: address space mapping from which to obtain page cache pages
403 * @index: starting index in @mapping at which to begin obtaining pages
404 * @nr_pages: number of page cache pages to obtain
405 * @pages: array of pages in which to return the obtained page cache pages
406 * @cached_page: allocated but as yet unused page
407 * @lru_pvec: lru-buffering pagevec of caller
409 * Obtain @nr_pages locked page cache pages from the mapping @maping and
410 * starting at index @index.
412 * If a page is newly created, increment its refcount and add it to the
413 * caller's lru-buffering pagevec @lru_pvec.
415 * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages
416 * are obtained at once instead of just one page and that 0 is returned on
417 * success and -errno on error.
419 * Note, the page locks are obtained in ascending page index order.
421 static inline int __ntfs_grab_cache_pages(struct address_space *mapping,
422 pgoff_t index, const unsigned nr_pages, struct page **pages,
423 struct page **cached_page, struct pagevec *lru_pvec)
430 pages[nr] = find_lock_page(mapping, index);
433 *cached_page = page_cache_alloc(mapping);
434 if (unlikely(!*cached_page)) {
439 err = add_to_page_cache(*cached_page, mapping, index,
446 pages[nr] = *cached_page;
447 page_cache_get(*cached_page);
448 if (unlikely(!pagevec_add(lru_pvec, *cached_page)))
449 __pagevec_lru_add(lru_pvec);
454 } while (nr < nr_pages);
459 unlock_page(pages[--nr]);
460 page_cache_release(pages[nr]);
465 static inline int ntfs_submit_bh_for_read(struct buffer_head *bh)
469 bh->b_end_io = end_buffer_read_sync;
470 return submit_bh(READ, bh);
474 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
475 * @pages: array of destination pages
476 * @nr_pages: number of pages in @pages
477 * @pos: byte position in file at which the write begins
478 * @bytes: number of bytes to be written
480 * This is called for non-resident attributes from ntfs_file_buffered_write()
481 * with i_sem held on the inode (@pages[0]->mapping->host). There are
482 * @nr_pages pages in @pages which are locked but not kmap()ped. The source
483 * data has not yet been copied into the @pages.
485 * Need to fill any holes with actual clusters, allocate buffers if necessary,
486 * ensure all the buffers are mapped, and bring uptodate any buffers that are
487 * only partially being written to.
489 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
490 * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside
491 * the same cluster and that they are the entirety of that cluster, and that
492 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
494 * i_size is not to be modified yet.
496 * Return 0 on success or -errno on error.
498 static int ntfs_prepare_pages_for_non_resident_write(struct page **pages,
499 unsigned nr_pages, s64 pos, size_t bytes)
501 VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend;
503 s64 bh_pos, vcn_len, end, initialized_size;
507 ntfs_inode *ni, *base_ni = NULL;
509 runlist_element *rl, *rl2;
510 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
511 ntfs_attr_search_ctx *ctx = NULL;
512 MFT_RECORD *m = NULL;
513 ATTR_RECORD *a = NULL;
515 u32 attr_rec_len = 0;
516 unsigned blocksize, u;
518 BOOL rl_write_locked, was_hole, is_retry;
519 unsigned char blocksize_bits;
522 u8 mft_attr_mapped:1;
525 } status = { 0, 0, 0, 0 };
530 vi = pages[0]->mapping->host;
533 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
534 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%x.",
535 vi->i_ino, ni->type, pages[0]->index, nr_pages,
536 (long long)pos, bytes);
537 blocksize_bits = vi->i_blkbits;
538 blocksize = 1 << blocksize_bits;
541 struct page *page = pages[u];
543 * create_empty_buffers() will create uptodate/dirty buffers if
544 * the page is uptodate/dirty.
546 if (!page_has_buffers(page)) {
547 create_empty_buffers(page, blocksize, 0);
548 if (unlikely(!page_has_buffers(page)))
551 } while (++u < nr_pages);
552 rl_write_locked = FALSE;
559 cpos = pos >> vol->cluster_size_bits;
561 cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits;
563 * Loop over each page and for each page over each buffer. Use goto to
564 * reduce indentation.
569 bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
570 bh = head = page_buffers(page);
576 /* Clear buffer_new on all buffers to reinitialise state. */
578 clear_buffer_new(bh);
579 bh_end = bh_pos + blocksize;
580 bh_cpos = bh_pos >> vol->cluster_size_bits;
581 bh_cofs = bh_pos & vol->cluster_size_mask;
582 if (buffer_mapped(bh)) {
584 * The buffer is already mapped. If it is uptodate,
587 if (buffer_uptodate(bh))
590 * The buffer is not uptodate. If the page is uptodate
591 * set the buffer uptodate and otherwise ignore it.
593 if (PageUptodate(page)) {
594 set_buffer_uptodate(bh);
598 * Neither the page nor the buffer are uptodate. If
599 * the buffer is only partially being written to, we
600 * need to read it in before the write, i.e. now.
602 if ((bh_pos < pos && bh_end > pos) ||
603 (bh_pos < end && bh_end > end)) {
605 * If the buffer is fully or partially within
606 * the initialized size, do an actual read.
607 * Otherwise, simply zero the buffer.
609 read_lock_irqsave(&ni->size_lock, flags);
610 initialized_size = ni->initialized_size;
611 read_unlock_irqrestore(&ni->size_lock, flags);
612 if (bh_pos < initialized_size) {
613 ntfs_submit_bh_for_read(bh);
616 u8 *kaddr = kmap_atomic(page, KM_USER0);
617 memset(kaddr + bh_offset(bh), 0,
619 kunmap_atomic(kaddr, KM_USER0);
620 flush_dcache_page(page);
621 set_buffer_uptodate(bh);
626 /* Unmapped buffer. Need to map it. */
627 bh->b_bdev = vol->sb->s_bdev;
629 * If the current buffer is in the same clusters as the map
630 * cache, there is no need to check the runlist again. The
631 * map cache is made up of @vcn, which is the first cached file
632 * cluster, @vcn_len which is the number of cached file
633 * clusters, @lcn is the device cluster corresponding to @vcn,
634 * and @lcn_block is the block number corresponding to @lcn.
636 cdelta = bh_cpos - vcn;
637 if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) {
640 bh->b_blocknr = lcn_block +
641 (cdelta << (vol->cluster_size_bits -
643 (bh_cofs >> blocksize_bits);
644 set_buffer_mapped(bh);
646 * If the page is uptodate so is the buffer. If the
647 * buffer is fully outside the write, we ignore it if
648 * it was already allocated and we mark it dirty so it
649 * gets written out if we allocated it. On the other
650 * hand, if we allocated the buffer but we are not
651 * marking it dirty we set buffer_new so we can do
654 if (PageUptodate(page)) {
655 if (!buffer_uptodate(bh))
656 set_buffer_uptodate(bh);
657 if (unlikely(was_hole)) {
658 /* We allocated the buffer. */
659 unmap_underlying_metadata(bh->b_bdev,
661 if (bh_end <= pos || bh_pos >= end)
662 mark_buffer_dirty(bh);
668 /* Page is _not_ uptodate. */
669 if (likely(!was_hole)) {
671 * Buffer was already allocated. If it is not
672 * uptodate and is only partially being written
673 * to, we need to read it in before the write,
676 if (!buffer_uptodate(bh) && ((bh_pos < pos &&
681 * If the buffer is fully or partially
682 * within the initialized size, do an
683 * actual read. Otherwise, simply zero
686 read_lock_irqsave(&ni->size_lock,
688 initialized_size = ni->initialized_size;
689 read_unlock_irqrestore(&ni->size_lock,
691 if (bh_pos < initialized_size) {
692 ntfs_submit_bh_for_read(bh);
695 u8 *kaddr = kmap_atomic(page,
697 memset(kaddr + bh_offset(bh),
699 kunmap_atomic(kaddr, KM_USER0);
700 flush_dcache_page(page);
701 set_buffer_uptodate(bh);
706 /* We allocated the buffer. */
707 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
709 * If the buffer is fully outside the write, zero it,
710 * set it uptodate, and mark it dirty so it gets
711 * written out. If it is partially being written to,
712 * zero region surrounding the write but leave it to
713 * commit write to do anything else. Finally, if the
714 * buffer is fully being overwritten, do nothing.
716 if (bh_end <= pos || bh_pos >= end) {
717 if (!buffer_uptodate(bh)) {
718 u8 *kaddr = kmap_atomic(page, KM_USER0);
719 memset(kaddr + bh_offset(bh), 0,
721 kunmap_atomic(kaddr, KM_USER0);
722 flush_dcache_page(page);
723 set_buffer_uptodate(bh);
725 mark_buffer_dirty(bh);
729 if (!buffer_uptodate(bh) &&
730 (bh_pos < pos || bh_end > end)) {
734 kaddr = kmap_atomic(page, KM_USER0);
736 pofs = bh_pos & ~PAGE_CACHE_MASK;
737 memset(kaddr + pofs, 0, pos - bh_pos);
740 pofs = end & ~PAGE_CACHE_MASK;
741 memset(kaddr + pofs, 0, bh_end - end);
743 kunmap_atomic(kaddr, KM_USER0);
744 flush_dcache_page(page);
749 * Slow path: this is the first buffer in the cluster. If it
750 * is outside allocated size and is not uptodate, zero it and
753 read_lock_irqsave(&ni->size_lock, flags);
754 initialized_size = ni->allocated_size;
755 read_unlock_irqrestore(&ni->size_lock, flags);
756 if (bh_pos > initialized_size) {
757 if (PageUptodate(page)) {
758 if (!buffer_uptodate(bh))
759 set_buffer_uptodate(bh);
760 } else if (!buffer_uptodate(bh)) {
761 u8 *kaddr = kmap_atomic(page, KM_USER0);
762 memset(kaddr + bh_offset(bh), 0, blocksize);
763 kunmap_atomic(kaddr, KM_USER0);
764 flush_dcache_page(page);
765 set_buffer_uptodate(bh);
771 down_read(&ni->runlist.lock);
775 if (likely(rl != NULL)) {
776 /* Seek to element containing target cluster. */
777 while (rl->length && rl[1].vcn <= bh_cpos)
779 lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos);
780 if (likely(lcn >= 0)) {
782 * Successful remap, setup the map cache and
783 * use that to deal with the buffer.
787 vcn_len = rl[1].vcn - vcn;
788 lcn_block = lcn << (vol->cluster_size_bits -
791 * If the number of remaining clusters in the
792 * @pages is smaller or equal to the number of
793 * cached clusters, unlock the runlist as the
794 * map cache will be used from now on.
796 if (likely(vcn + vcn_len >= cend)) {
797 if (rl_write_locked) {
798 up_write(&ni->runlist.lock);
799 rl_write_locked = FALSE;
801 up_read(&ni->runlist.lock);
804 goto map_buffer_cached;
807 lcn = LCN_RL_NOT_MAPPED;
809 * If it is not a hole and not out of bounds, the runlist is
810 * probably unmapped so try to map it now.
812 if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) {
813 if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) {
814 /* Attempt to map runlist. */
815 if (!rl_write_locked) {
817 * We need the runlist locked for
818 * writing, so if it is locked for
819 * reading relock it now and retry in
820 * case it changed whilst we dropped
823 up_read(&ni->runlist.lock);
824 down_write(&ni->runlist.lock);
825 rl_write_locked = TRUE;
828 err = ntfs_map_runlist_nolock(ni, bh_cpos,
835 * If @vcn is out of bounds, pretend @lcn is
836 * LCN_ENOENT. As long as the buffer is out
837 * of bounds this will work fine.
839 if (err == -ENOENT) {
842 goto rl_not_mapped_enoent;
846 /* Failed to map the buffer, even after retrying. */
848 ntfs_error(vol->sb, "Failed to write to inode 0x%lx, "
849 "attribute type 0x%x, vcn 0x%llx, "
850 "vcn offset 0x%x, because its "
851 "location on disk could not be "
852 "determined%s (error code %i).",
853 ni->mft_no, ni->type,
854 (unsigned long long)bh_cpos,
856 vol->cluster_size_mask,
857 is_retry ? " even after retrying" : "",
861 rl_not_mapped_enoent:
863 * The buffer is in a hole or out of bounds. We need to fill
864 * the hole, unless the buffer is in a cluster which is not
865 * touched by the write, in which case we just leave the buffer
866 * unmapped. This can only happen when the cluster size is
867 * less than the page cache size.
869 if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) {
870 bh_cend = (bh_end + vol->cluster_size - 1) >>
871 vol->cluster_size_bits;
872 if ((bh_cend <= cpos || bh_cpos >= cend)) {
875 * If the buffer is uptodate we skip it. If it
876 * is not but the page is uptodate, we can set
877 * the buffer uptodate. If the page is not
878 * uptodate, we can clear the buffer and set it
879 * uptodate. Whether this is worthwhile is
880 * debatable and this could be removed.
882 if (PageUptodate(page)) {
883 if (!buffer_uptodate(bh))
884 set_buffer_uptodate(bh);
885 } else if (!buffer_uptodate(bh)) {
886 u8 *kaddr = kmap_atomic(page, KM_USER0);
887 memset(kaddr + bh_offset(bh), 0,
889 kunmap_atomic(kaddr, KM_USER0);
890 flush_dcache_page(page);
891 set_buffer_uptodate(bh);
897 * Out of bounds buffer is invalid if it was not really out of
900 BUG_ON(lcn != LCN_HOLE);
902 * We need the runlist locked for writing, so if it is locked
903 * for reading relock it now and retry in case it changed
904 * whilst we dropped the lock.
907 if (!rl_write_locked) {
908 up_read(&ni->runlist.lock);
909 down_write(&ni->runlist.lock);
910 rl_write_locked = TRUE;
913 /* Find the previous last allocated cluster. */
914 BUG_ON(rl->lcn != LCN_HOLE);
917 while (--rl2 >= ni->runlist.rl) {
919 lcn = rl2->lcn + rl2->length;
923 rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE,
927 ntfs_debug("Failed to allocate cluster, error code %i.",
932 rl = ntfs_runlists_merge(ni->runlist.rl, rl2);
937 if (ntfs_cluster_free_from_rl(vol, rl2)) {
938 ntfs_error(vol->sb, "Failed to release "
939 "allocated cluster in error "
940 "code path. Run chkdsk to "
941 "recover the lost cluster.");
948 status.runlist_merged = 1;
949 ntfs_debug("Allocated cluster, lcn 0x%llx.", lcn);
950 /* Map and lock the mft record and get the attribute record. */
954 base_ni = ni->ext.base_ntfs_ino;
955 m = map_mft_record(base_ni);
960 ctx = ntfs_attr_get_search_ctx(base_ni, m);
961 if (unlikely(!ctx)) {
963 unmap_mft_record(base_ni);
966 status.mft_attr_mapped = 1;
967 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
968 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx);
977 * Find the runlist element with which the attribute extent
978 * starts. Note, we cannot use the _attr_ version because we
979 * have mapped the mft record. That is ok because we know the
980 * runlist fragment must be mapped already to have ever gotten
981 * here, so we can just use the _rl_ version.
983 vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn);
984 rl2 = ntfs_rl_find_vcn_nolock(rl, vcn);
986 BUG_ON(!rl2->length);
987 BUG_ON(rl2->lcn < LCN_HOLE);
988 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
990 * If @highest_vcn is zero, calculate the real highest_vcn
991 * (which can really be zero).
994 highest_vcn = (sle64_to_cpu(
995 a->data.non_resident.allocated_size) >>
996 vol->cluster_size_bits) - 1;
998 * Determine the size of the mapping pairs array for the new
999 * extent, i.e. the old extent with the hole filled.
1001 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn,
1003 if (unlikely(mp_size <= 0)) {
1004 if (!(err = mp_size))
1006 ntfs_debug("Failed to get size for mapping pairs "
1007 "array, error code %i.", err);
1011 * Resize the attribute record to fit the new mapping pairs
1014 attr_rec_len = le32_to_cpu(a->length);
1015 err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu(
1016 a->data.non_resident.mapping_pairs_offset));
1017 if (unlikely(err)) {
1018 BUG_ON(err != -ENOSPC);
1019 // TODO: Deal with this by using the current attribute
1020 // and fill it with as much of the mapping pairs
1021 // array as possible. Then loop over each attribute
1022 // extent rewriting the mapping pairs arrays as we go
1023 // along and if when we reach the end we have not
1024 // enough space, try to resize the last attribute
1025 // extent and if even that fails, add a new attribute
1027 // We could also try to resize at each step in the hope
1028 // that we will not need to rewrite every single extent.
1029 // Note, we may need to decompress some extents to fill
1030 // the runlist as we are walking the extents...
1031 ntfs_error(vol->sb, "Not enough space in the mft "
1032 "record for the extended attribute "
1033 "record. This case is not "
1034 "implemented yet.");
1038 status.mp_rebuilt = 1;
1040 * Generate the mapping pairs array directly into the attribute
1043 err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu(
1044 a->data.non_resident.mapping_pairs_offset),
1045 mp_size, rl2, vcn, highest_vcn, NULL);
1046 if (unlikely(err)) {
1047 ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, "
1048 "attribute type 0x%x, because building "
1049 "the mapping pairs failed with error "
1050 "code %i.", vi->i_ino,
1051 (unsigned)le32_to_cpu(ni->type), err);
1055 /* Update the highest_vcn but only if it was not set. */
1056 if (unlikely(!a->data.non_resident.highest_vcn))
1057 a->data.non_resident.highest_vcn =
1058 cpu_to_sle64(highest_vcn);
1060 * If the attribute is sparse/compressed, update the compressed
1061 * size in the ntfs_inode structure and the attribute record.
1063 if (likely(NInoSparse(ni) || NInoCompressed(ni))) {
1065 * If we are not in the first attribute extent, switch
1066 * to it, but first ensure the changes will make it to
1069 if (a->data.non_resident.lowest_vcn) {
1070 flush_dcache_mft_record_page(ctx->ntfs_ino);
1071 mark_mft_record_dirty(ctx->ntfs_ino);
1072 ntfs_attr_reinit_search_ctx(ctx);
1073 err = ntfs_attr_lookup(ni->type, ni->name,
1074 ni->name_len, CASE_SENSITIVE,
1076 if (unlikely(err)) {
1077 status.attr_switched = 1;
1080 /* @m is not used any more so do not set it. */
1083 write_lock_irqsave(&ni->size_lock, flags);
1084 ni->itype.compressed.size += vol->cluster_size;
1085 a->data.non_resident.compressed_size =
1086 cpu_to_sle64(ni->itype.compressed.size);
1087 write_unlock_irqrestore(&ni->size_lock, flags);
1089 /* Ensure the changes make it to disk. */
1090 flush_dcache_mft_record_page(ctx->ntfs_ino);
1091 mark_mft_record_dirty(ctx->ntfs_ino);
1092 ntfs_attr_put_search_ctx(ctx);
1093 unmap_mft_record(base_ni);
1094 /* Successfully filled the hole. */
1095 status.runlist_merged = 0;
1096 status.mft_attr_mapped = 0;
1097 status.mp_rebuilt = 0;
1098 /* Setup the map cache and use that to deal with the buffer. */
1102 lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits);
1105 * If the number of remaining clusters in the @pages is smaller
1106 * or equal to the number of cached clusters, unlock the
1107 * runlist as the map cache will be used from now on.
1109 if (likely(vcn + vcn_len >= cend)) {
1110 up_write(&ni->runlist.lock);
1111 rl_write_locked = FALSE;
1114 goto map_buffer_cached;
1115 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1116 /* If there are no errors, do the next page. */
1117 if (likely(!err && ++u < nr_pages))
1119 /* If there are no errors, release the runlist lock if we took it. */
1121 if (unlikely(rl_write_locked)) {
1122 up_write(&ni->runlist.lock);
1123 rl_write_locked = FALSE;
1124 } else if (unlikely(rl))
1125 up_read(&ni->runlist.lock);
1128 /* If we issued read requests, let them complete. */
1129 read_lock_irqsave(&ni->size_lock, flags);
1130 initialized_size = ni->initialized_size;
1131 read_unlock_irqrestore(&ni->size_lock, flags);
1132 while (wait_bh > wait) {
1135 if (likely(buffer_uptodate(bh))) {
1137 bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) +
1140 * If the buffer overflows the initialized size, need
1141 * to zero the overflowing region.
1143 if (unlikely(bh_pos + blocksize > initialized_size)) {
1147 if (likely(bh_pos < initialized_size))
1148 ofs = initialized_size - bh_pos;
1149 kaddr = kmap_atomic(page, KM_USER0);
1150 memset(kaddr + bh_offset(bh) + ofs, 0,
1152 kunmap_atomic(kaddr, KM_USER0);
1153 flush_dcache_page(page);
1155 } else /* if (unlikely(!buffer_uptodate(bh))) */
1159 /* Clear buffer_new on all buffers. */
1162 bh = head = page_buffers(pages[u]);
1165 clear_buffer_new(bh);
1166 } while ((bh = bh->b_this_page) != head);
1167 } while (++u < nr_pages);
1168 ntfs_debug("Done.");
1171 if (status.attr_switched) {
1172 /* Get back to the attribute extent we modified. */
1173 ntfs_attr_reinit_search_ctx(ctx);
1174 if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1175 CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) {
1176 ntfs_error(vol->sb, "Failed to find required "
1177 "attribute extent of attribute in "
1178 "error code path. Run chkdsk to "
1180 write_lock_irqsave(&ni->size_lock, flags);
1181 ni->itype.compressed.size += vol->cluster_size;
1182 write_unlock_irqrestore(&ni->size_lock, flags);
1183 flush_dcache_mft_record_page(ctx->ntfs_ino);
1184 mark_mft_record_dirty(ctx->ntfs_ino);
1186 * The only thing that is now wrong is the compressed
1187 * size of the base attribute extent which chkdsk
1188 * should be able to fix.
1194 status.attr_switched = 0;
1198 * If the runlist has been modified, need to restore it by punching a
1199 * hole into it and we then need to deallocate the on-disk cluster as
1200 * well. Note, we only modify the runlist if we are able to generate a
1201 * new mapping pairs array, i.e. only when the mapped attribute extent
1204 if (status.runlist_merged && !status.attr_switched) {
1205 BUG_ON(!rl_write_locked);
1206 /* Make the file cluster we allocated sparse in the runlist. */
1207 if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) {
1208 ntfs_error(vol->sb, "Failed to punch hole into "
1209 "attribute runlist in error code "
1210 "path. Run chkdsk to recover the "
1213 make_bad_inode(VFS_I(base_ni));
1215 } else /* if (success) */ {
1216 status.runlist_merged = 0;
1218 * Deallocate the on-disk cluster we allocated but only
1219 * if we succeeded in punching its vcn out of the
1222 down_write(&vol->lcnbmp_lock);
1223 if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) {
1224 ntfs_error(vol->sb, "Failed to release "
1225 "allocated cluster in error "
1226 "code path. Run chkdsk to "
1227 "recover the lost cluster.");
1230 up_write(&vol->lcnbmp_lock);
1234 * Resize the attribute record to its old size and rebuild the mapping
1235 * pairs array. Note, we only can do this if the runlist has been
1236 * restored to its old state which also implies that the mapped
1237 * attribute extent is not switched.
1239 if (status.mp_rebuilt && !status.runlist_merged) {
1240 if (ntfs_attr_record_resize(m, a, attr_rec_len)) {
1241 ntfs_error(vol->sb, "Failed to restore attribute "
1242 "record in error code path. Run "
1243 "chkdsk to recover.");
1245 make_bad_inode(VFS_I(base_ni));
1247 } else /* if (success) */ {
1248 if (ntfs_mapping_pairs_build(vol, (u8*)a +
1249 le16_to_cpu(a->data.non_resident.
1250 mapping_pairs_offset), attr_rec_len -
1251 le16_to_cpu(a->data.non_resident.
1252 mapping_pairs_offset), ni->runlist.rl,
1253 vcn, highest_vcn, NULL)) {
1254 ntfs_error(vol->sb, "Failed to restore "
1255 "mapping pairs array in error "
1256 "code path. Run chkdsk to "
1259 make_bad_inode(VFS_I(base_ni));
1262 flush_dcache_mft_record_page(ctx->ntfs_ino);
1263 mark_mft_record_dirty(ctx->ntfs_ino);
1266 /* Release the mft record and the attribute. */
1267 if (status.mft_attr_mapped) {
1268 ntfs_attr_put_search_ctx(ctx);
1269 unmap_mft_record(base_ni);
1271 /* Release the runlist lock. */
1272 if (rl_write_locked)
1273 up_write(&ni->runlist.lock);
1275 up_read(&ni->runlist.lock);
1277 * Zero out any newly allocated blocks to avoid exposing stale data.
1278 * If BH_New is set, we know that the block was newly allocated above
1279 * and that it has not been fully zeroed and marked dirty yet.
1283 end = bh_cpos << vol->cluster_size_bits;
1286 bh = head = page_buffers(page);
1288 if (u == nr_pages &&
1289 ((s64)page->index << PAGE_CACHE_SHIFT) +
1290 bh_offset(bh) >= end)
1292 if (!buffer_new(bh))
1294 clear_buffer_new(bh);
1295 if (!buffer_uptodate(bh)) {
1296 if (PageUptodate(page))
1297 set_buffer_uptodate(bh);
1299 u8 *kaddr = kmap_atomic(page, KM_USER0);
1300 memset(kaddr + bh_offset(bh), 0,
1302 kunmap_atomic(kaddr, KM_USER0);
1303 flush_dcache_page(page);
1304 set_buffer_uptodate(bh);
1307 mark_buffer_dirty(bh);
1308 } while ((bh = bh->b_this_page) != head);
1309 } while (++u <= nr_pages);
1310 ntfs_error(vol->sb, "Failed. Returning error code %i.", err);
1315 * Copy as much as we can into the pages and return the number of bytes which
1316 * were sucessfully copied. If a fault is encountered then clear the pages
1317 * out to (ofs + bytes) and return the number of bytes which were copied.
1319 static inline size_t ntfs_copy_from_user(struct page **pages,
1320 unsigned nr_pages, unsigned ofs, const char __user *buf,
1323 struct page **last_page = pages + nr_pages;
1330 len = PAGE_CACHE_SIZE - ofs;
1333 kaddr = kmap_atomic(*pages, KM_USER0);
1334 left = __copy_from_user_inatomic(kaddr + ofs, buf, len);
1335 kunmap_atomic(kaddr, KM_USER0);
1336 if (unlikely(left)) {
1337 /* Do it the slow way. */
1338 kaddr = kmap(*pages);
1339 left = __copy_from_user(kaddr + ofs, buf, len);
1350 } while (++pages < last_page);
1354 total += len - left;
1355 /* Zero the rest of the target like __copy_from_user(). */
1356 while (++pages < last_page) {
1360 len = PAGE_CACHE_SIZE;
1363 kaddr = kmap_atomic(*pages, KM_USER0);
1364 memset(kaddr, 0, len);
1365 kunmap_atomic(kaddr, KM_USER0);
1370 static size_t __ntfs_copy_from_user_iovec(char *vaddr,
1371 const struct iovec *iov, size_t iov_ofs, size_t bytes)
1376 const char __user *buf = iov->iov_base + iov_ofs;
1380 len = iov->iov_len - iov_ofs;
1383 left = __copy_from_user_inatomic(vaddr, buf, len);
1387 if (unlikely(left)) {
1389 * Zero the rest of the target like __copy_from_user().
1391 memset(vaddr, 0, bytes);
1403 static inline void ntfs_set_next_iovec(const struct iovec **iovp,
1404 size_t *iov_ofsp, size_t bytes)
1406 const struct iovec *iov = *iovp;
1407 size_t iov_ofs = *iov_ofsp;
1412 len = iov->iov_len - iov_ofs;
1417 if (iov->iov_len == iov_ofs) {
1423 *iov_ofsp = iov_ofs;
1427 * This has the same side-effects and return value as ntfs_copy_from_user().
1428 * The difference is that on a fault we need to memset the remainder of the
1429 * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s
1430 * single-segment behaviour.
1432 * We call the same helper (__ntfs_copy_from_user_iovec()) both when atomic and
1433 * when not atomic. This is ok because __ntfs_copy_from_user_iovec() calls
1434 * __copy_from_user_inatomic() and it is ok to call this when non-atomic. In
1435 * fact, the only difference between __copy_from_user_inatomic() and
1436 * __copy_from_user() is that the latter calls might_sleep(). And on many
1437 * architectures __copy_from_user_inatomic() is just defined to
1438 * __copy_from_user() so it makes no difference at all on those architectures.
1440 static inline size_t ntfs_copy_from_user_iovec(struct page **pages,
1441 unsigned nr_pages, unsigned ofs, const struct iovec **iov,
1442 size_t *iov_ofs, size_t bytes)
1444 struct page **last_page = pages + nr_pages;
1446 size_t copied, len, total = 0;
1449 len = PAGE_CACHE_SIZE - ofs;
1452 kaddr = kmap_atomic(*pages, KM_USER0);
1453 copied = __ntfs_copy_from_user_iovec(kaddr + ofs,
1454 *iov, *iov_ofs, len);
1455 kunmap_atomic(kaddr, KM_USER0);
1456 if (unlikely(copied != len)) {
1457 /* Do it the slow way. */
1458 kaddr = kmap(*pages);
1459 copied = __ntfs_copy_from_user_iovec(kaddr + ofs,
1460 *iov, *iov_ofs, len);
1462 if (unlikely(copied != len))
1469 ntfs_set_next_iovec(iov, iov_ofs, len);
1471 } while (++pages < last_page);
1476 /* Zero the rest of the target like __copy_from_user(). */
1477 while (++pages < last_page) {
1481 len = PAGE_CACHE_SIZE;
1484 kaddr = kmap_atomic(*pages, KM_USER0);
1485 memset(kaddr, 0, len);
1486 kunmap_atomic(kaddr, KM_USER0);
1491 static inline void ntfs_flush_dcache_pages(struct page **pages,
1497 * Warning: Do not do the decrement at the same time as the
1498 * call because flush_dcache_page() is a NULL macro on i386
1499 * and hence the decrement never happens.
1501 flush_dcache_page(pages[nr_pages]);
1502 } while (--nr_pages > 0);
1506 * ntfs_commit_pages_after_non_resident_write - commit the received data
1507 * @pages: array of destination pages
1508 * @nr_pages: number of pages in @pages
1509 * @pos: byte position in file at which the write begins
1510 * @bytes: number of bytes to be written
1512 * See description of ntfs_commit_pages_after_write(), below.
1514 static inline int ntfs_commit_pages_after_non_resident_write(
1515 struct page **pages, const unsigned nr_pages,
1516 s64 pos, size_t bytes)
1518 s64 end, initialized_size;
1520 ntfs_inode *ni, *base_ni;
1521 struct buffer_head *bh, *head;
1522 ntfs_attr_search_ctx *ctx;
1525 unsigned long flags;
1526 unsigned blocksize, u;
1529 vi = pages[0]->mapping->host;
1531 blocksize = 1 << vi->i_blkbits;
1540 bh_pos = (s64)page->index << PAGE_CACHE_SHIFT;
1541 bh = head = page_buffers(page);
1546 bh_end = bh_pos + blocksize;
1547 if (bh_end <= pos || bh_pos >= end) {
1548 if (!buffer_uptodate(bh))
1551 set_buffer_uptodate(bh);
1552 mark_buffer_dirty(bh);
1554 } while (bh_pos += blocksize, (bh = bh->b_this_page) != head);
1556 * If all buffers are now uptodate but the page is not, set the
1559 if (!partial && !PageUptodate(page))
1560 SetPageUptodate(page);
1561 } while (++u < nr_pages);
1563 * Finally, if we do not need to update initialized_size or i_size we
1566 read_lock_irqsave(&ni->size_lock, flags);
1567 initialized_size = ni->initialized_size;
1568 read_unlock_irqrestore(&ni->size_lock, flags);
1569 if (end <= initialized_size) {
1570 ntfs_debug("Done.");
1574 * Update initialized_size/i_size as appropriate, both in the inode and
1580 base_ni = ni->ext.base_ntfs_ino;
1581 /* Map, pin, and lock the mft record. */
1582 m = map_mft_record(base_ni);
1589 BUG_ON(!NInoNonResident(ni));
1590 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1591 if (unlikely(!ctx)) {
1595 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1596 CASE_SENSITIVE, 0, NULL, 0, ctx);
1597 if (unlikely(err)) {
1603 BUG_ON(!a->non_resident);
1604 write_lock_irqsave(&ni->size_lock, flags);
1605 BUG_ON(end > ni->allocated_size);
1606 ni->initialized_size = end;
1607 a->data.non_resident.initialized_size = cpu_to_sle64(end);
1608 if (end > i_size_read(vi)) {
1609 i_size_write(vi, end);
1610 a->data.non_resident.data_size =
1611 a->data.non_resident.initialized_size;
1613 write_unlock_irqrestore(&ni->size_lock, flags);
1614 /* Mark the mft record dirty, so it gets written back. */
1615 flush_dcache_mft_record_page(ctx->ntfs_ino);
1616 mark_mft_record_dirty(ctx->ntfs_ino);
1617 ntfs_attr_put_search_ctx(ctx);
1618 unmap_mft_record(base_ni);
1619 ntfs_debug("Done.");
1623 ntfs_attr_put_search_ctx(ctx);
1625 unmap_mft_record(base_ni);
1626 ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error "
1628 if (err != -ENOMEM) {
1629 NVolSetErrors(ni->vol);
1630 make_bad_inode(VFS_I(base_ni));
1637 * ntfs_commit_pages_after_write - commit the received data
1638 * @pages: array of destination pages
1639 * @nr_pages: number of pages in @pages
1640 * @pos: byte position in file at which the write begins
1641 * @bytes: number of bytes to be written
1643 * This is called from ntfs_file_buffered_write() with i_sem held on the inode
1644 * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are
1645 * locked but not kmap()ped. The source data has already been copied into the
1646 * @page. ntfs_prepare_pages_for_non_resident_write() has been called before
1647 * the data was copied (for non-resident attributes only) and it returned
1650 * Need to set uptodate and mark dirty all buffers within the boundary of the
1651 * write. If all buffers in a page are uptodate we set the page uptodate, too.
1653 * Setting the buffers dirty ensures that they get written out later when
1654 * ntfs_writepage() is invoked by the VM.
1656 * Finally, we need to update i_size and initialized_size as appropriate both
1657 * in the inode and the mft record.
1659 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1660 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1661 * page are uptodate, and updates i_size if the end of io is beyond i_size. In
1662 * that case, it also marks the inode dirty.
1664 * If things have gone as outlined in
1665 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1666 * content modifications here for non-resident attributes. For resident
1667 * attributes we need to do the uptodate bringing here which we combine with
1668 * the copying into the mft record which means we save one atomic kmap.
1670 * Return 0 on success or -errno on error.
1672 static int ntfs_commit_pages_after_write(struct page **pages,
1673 const unsigned nr_pages, s64 pos, size_t bytes)
1675 s64 end, initialized_size;
1678 ntfs_inode *ni, *base_ni;
1680 ntfs_attr_search_ctx *ctx;
1683 char *kattr, *kaddr;
1684 unsigned long flags;
1692 vi = page->mapping->host;
1694 ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page "
1695 "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%x.",
1696 vi->i_ino, ni->type, page->index, nr_pages,
1697 (long long)pos, bytes);
1698 if (NInoNonResident(ni))
1699 return ntfs_commit_pages_after_non_resident_write(pages,
1700 nr_pages, pos, bytes);
1701 BUG_ON(nr_pages > 1);
1703 * Attribute is resident, implying it is not compressed, encrypted, or
1709 base_ni = ni->ext.base_ntfs_ino;
1710 BUG_ON(NInoNonResident(ni));
1711 /* Map, pin, and lock the mft record. */
1712 m = map_mft_record(base_ni);
1719 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1720 if (unlikely(!ctx)) {
1724 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1725 CASE_SENSITIVE, 0, NULL, 0, ctx);
1726 if (unlikely(err)) {
1732 BUG_ON(a->non_resident);
1733 /* The total length of the attribute value. */
1734 attr_len = le32_to_cpu(a->data.resident.value_length);
1735 i_size = i_size_read(vi);
1736 BUG_ON(attr_len != i_size);
1737 BUG_ON(pos > attr_len);
1739 BUG_ON(end > le32_to_cpu(a->length) -
1740 le16_to_cpu(a->data.resident.value_offset));
1741 kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset);
1742 kaddr = kmap_atomic(page, KM_USER0);
1743 /* Copy the received data from the page to the mft record. */
1744 memcpy(kattr + pos, kaddr + pos, bytes);
1745 /* Update the attribute length if necessary. */
1746 if (end > attr_len) {
1748 a->data.resident.value_length = cpu_to_le32(attr_len);
1751 * If the page is not uptodate, bring the out of bounds area(s)
1752 * uptodate by copying data from the mft record to the page.
1754 if (!PageUptodate(page)) {
1756 memcpy(kaddr, kattr, pos);
1758 memcpy(kaddr + end, kattr + end, attr_len - end);
1759 /* Zero the region outside the end of the attribute value. */
1760 memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len);
1761 flush_dcache_page(page);
1762 SetPageUptodate(page);
1764 kunmap_atomic(kaddr, KM_USER0);
1765 /* Update initialized_size/i_size if necessary. */
1766 read_lock_irqsave(&ni->size_lock, flags);
1767 initialized_size = ni->initialized_size;
1768 BUG_ON(end > ni->allocated_size);
1769 read_unlock_irqrestore(&ni->size_lock, flags);
1770 BUG_ON(initialized_size != i_size);
1771 if (end > initialized_size) {
1772 unsigned long flags;
1774 write_lock_irqsave(&ni->size_lock, flags);
1775 ni->initialized_size = end;
1776 i_size_write(vi, end);
1777 write_unlock_irqrestore(&ni->size_lock, flags);
1779 /* Mark the mft record dirty, so it gets written back. */
1780 flush_dcache_mft_record_page(ctx->ntfs_ino);
1781 mark_mft_record_dirty(ctx->ntfs_ino);
1782 ntfs_attr_put_search_ctx(ctx);
1783 unmap_mft_record(base_ni);
1784 ntfs_debug("Done.");
1787 if (err == -ENOMEM) {
1788 ntfs_warning(vi->i_sb, "Error allocating memory required to "
1789 "commit the write.");
1790 if (PageUptodate(page)) {
1791 ntfs_warning(vi->i_sb, "Page is uptodate, setting "
1792 "dirty so the write will be retried "
1793 "later on by the VM.");
1795 * Put the page on mapping->dirty_pages, but leave its
1796 * buffers' dirty state as-is.
1798 __set_page_dirty_nobuffers(page);
1801 ntfs_error(vi->i_sb, "Page is not uptodate. Written "
1802 "data has been lost.");
1804 ntfs_error(vi->i_sb, "Resident attribute commit write failed "
1805 "with error %i.", err);
1806 NVolSetErrors(ni->vol);
1807 make_bad_inode(VFS_I(base_ni));
1811 ntfs_attr_put_search_ctx(ctx);
1813 unmap_mft_record(base_ni);
1818 * ntfs_file_buffered_write -
1820 * Locking: The vfs is holding ->i_sem on the inode.
1822 static ssize_t ntfs_file_buffered_write(struct kiocb *iocb,
1823 const struct iovec *iov, unsigned long nr_segs,
1824 loff_t pos, loff_t *ppos, size_t count)
1826 struct file *file = iocb->ki_filp;
1827 struct address_space *mapping = file->f_mapping;
1828 struct inode *vi = mapping->host;
1829 ntfs_inode *ni = NTFS_I(vi);
1830 ntfs_volume *vol = ni->vol;
1831 struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER];
1832 struct page *cached_page = NULL;
1833 char __user *buf = NULL;
1837 unsigned long flags;
1838 size_t bytes, iov_ofs;
1839 ssize_t status, written;
1842 struct pagevec lru_pvec;
1844 ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
1845 "pos 0x%llx, count 0x%lx.",
1846 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
1847 (unsigned long long)pos, (unsigned long)count);
1848 if (unlikely(!count))
1850 BUG_ON(NInoMstProtected(ni));
1852 * If the attribute is not an index root and it is encrypted or
1853 * compressed, we cannot write to it yet. Note we need to check for
1854 * AT_INDEX_ALLOCATION since this is the type of both directory and
1857 if (ni->type != AT_INDEX_ALLOCATION) {
1858 /* If file is encrypted, deny access, just like NT4. */
1859 if (NInoEncrypted(ni)) {
1860 ntfs_debug("Denying write access to encrypted file.");
1863 if (NInoCompressed(ni)) {
1864 ntfs_error(vi->i_sb, "Writing to compressed files is "
1865 "not implemented yet. Sorry.");
1870 * If a previous ntfs_truncate() failed, repeat it and abort if it
1873 if (unlikely(NInoTruncateFailed(ni))) {
1874 down_write(&vi->i_alloc_sem);
1875 err = ntfs_truncate(vi);
1876 up_write(&vi->i_alloc_sem);
1877 if (err || NInoTruncateFailed(ni)) {
1880 ntfs_error(vol->sb, "Cannot perform write to inode "
1881 "0x%lx, attribute type 0x%x, because "
1882 "ntfs_truncate() failed (error code "
1884 (unsigned)le32_to_cpu(ni->type), err);
1888 /* The first byte after the write. */
1891 * If the write goes beyond the allocated size, extend the allocation
1892 * to cover the whole of the write, rounded up to the nearest cluster.
1894 read_lock_irqsave(&ni->size_lock, flags);
1895 ll = ni->allocated_size;
1896 read_unlock_irqrestore(&ni->size_lock, flags);
1898 /* Extend the allocation without changing the data size. */
1899 ll = ntfs_attr_extend_allocation(ni, end, -1, pos);
1900 if (likely(ll >= 0)) {
1902 /* If the extension was partial truncate the write. */
1904 ntfs_debug("Truncating write to inode 0x%lx, "
1905 "attribute type 0x%x, because "
1906 "the allocation was only "
1907 "partially extended.",
1908 vi->i_ino, (unsigned)
1909 le32_to_cpu(ni->type));
1915 read_lock_irqsave(&ni->size_lock, flags);
1916 ll = ni->allocated_size;
1917 read_unlock_irqrestore(&ni->size_lock, flags);
1918 /* Perform a partial write if possible or fail. */
1920 ntfs_debug("Truncating write to inode 0x%lx, "
1921 "attribute type 0x%x, because "
1922 "extending the allocation "
1923 "failed (error code %i).",
1924 vi->i_ino, (unsigned)
1925 le32_to_cpu(ni->type), err);
1929 ntfs_error(vol->sb, "Cannot perform write to "
1930 "inode 0x%lx, attribute type "
1931 "0x%x, because extending the "
1932 "allocation failed (error "
1933 "code %i).", vi->i_ino,
1935 le32_to_cpu(ni->type), err);
1940 pagevec_init(&lru_pvec, 0);
1943 * If the write starts beyond the initialized size, extend it up to the
1944 * beginning of the write and initialize all non-sparse space between
1945 * the old initialized size and the new one. This automatically also
1946 * increments the vfs inode->i_size to keep it above or equal to the
1949 read_lock_irqsave(&ni->size_lock, flags);
1950 ll = ni->initialized_size;
1951 read_unlock_irqrestore(&ni->size_lock, flags);
1953 err = ntfs_attr_extend_initialized(ni, pos, &cached_page,
1956 ntfs_error(vol->sb, "Cannot perform write to inode "
1957 "0x%lx, attribute type 0x%x, because "
1958 "extending the initialized size "
1959 "failed (error code %i).", vi->i_ino,
1960 (unsigned)le32_to_cpu(ni->type), err);
1966 * Determine the number of pages per cluster for non-resident
1970 if (vol->cluster_size > PAGE_CACHE_SIZE && NInoNonResident(ni))
1971 nr_pages = vol->cluster_size >> PAGE_CACHE_SHIFT;
1972 /* Finally, perform the actual write. */
1974 if (likely(nr_segs == 1))
1975 buf = iov->iov_base;
1977 iov_ofs = 0; /* Offset in the current iovec. */
1980 pgoff_t idx, start_idx;
1981 unsigned ofs, do_pages, u;
1984 start_idx = idx = pos >> PAGE_CACHE_SHIFT;
1985 ofs = pos & ~PAGE_CACHE_MASK;
1986 bytes = PAGE_CACHE_SIZE - ofs;
1989 vcn = pos >> vol->cluster_size_bits;
1990 if (vcn != last_vcn) {
1993 * Get the lcn of the vcn the write is in. If
1994 * it is a hole, need to lock down all pages in
1997 down_read(&ni->runlist.lock);
1998 lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >>
1999 vol->cluster_size_bits, FALSE);
2000 up_read(&ni->runlist.lock);
2001 if (unlikely(lcn < LCN_HOLE)) {
2003 if (lcn == LCN_ENOMEM)
2006 ntfs_error(vol->sb, "Cannot "
2009 "attribute type 0x%x, "
2010 "because the attribute "
2012 vi->i_ino, (unsigned)
2013 le32_to_cpu(ni->type));
2016 if (lcn == LCN_HOLE) {
2017 start_idx = (pos & ~(s64)
2018 vol->cluster_size_mask)
2019 >> PAGE_CACHE_SHIFT;
2020 bytes = vol->cluster_size - (pos &
2021 vol->cluster_size_mask);
2022 do_pages = nr_pages;
2029 * Bring in the user page(s) that we will copy from _first_.
2030 * Otherwise there is a nasty deadlock on copying from the same
2031 * page(s) as we are writing to, without it/them being marked
2032 * up-to-date. Note, at present there is nothing to stop the
2033 * pages being swapped out between us bringing them into memory
2034 * and doing the actual copying.
2036 if (likely(nr_segs == 1))
2037 ntfs_fault_in_pages_readable(buf, bytes);
2039 ntfs_fault_in_pages_readable_iovec(iov, iov_ofs, bytes);
2040 /* Get and lock @do_pages starting at index @start_idx. */
2041 status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages,
2042 pages, &cached_page, &lru_pvec);
2043 if (unlikely(status))
2046 * For non-resident attributes, we need to fill any holes with
2047 * actual clusters and ensure all bufferes are mapped. We also
2048 * need to bring uptodate any buffers that are only partially
2051 if (NInoNonResident(ni)) {
2052 status = ntfs_prepare_pages_for_non_resident_write(
2053 pages, do_pages, pos, bytes);
2054 if (unlikely(status)) {
2058 unlock_page(pages[--do_pages]);
2059 page_cache_release(pages[do_pages]);
2062 * The write preparation may have instantiated
2063 * allocated space outside i_size. Trim this
2064 * off again. We can ignore any errors in this
2065 * case as we will just be waisting a bit of
2066 * allocated space, which is not a disaster.
2068 i_size = i_size_read(vi);
2069 if (pos + bytes > i_size)
2070 vmtruncate(vi, i_size);
2074 u = (pos >> PAGE_CACHE_SHIFT) - pages[0]->index;
2075 if (likely(nr_segs == 1)) {
2076 copied = ntfs_copy_from_user(pages + u, do_pages - u,
2080 copied = ntfs_copy_from_user_iovec(pages + u,
2081 do_pages - u, ofs, &iov, &iov_ofs,
2083 ntfs_flush_dcache_pages(pages + u, do_pages - u);
2084 status = ntfs_commit_pages_after_write(pages, do_pages, pos,
2086 if (likely(!status)) {
2090 if (unlikely(copied != bytes))
2094 unlock_page(pages[--do_pages]);
2095 mark_page_accessed(pages[do_pages]);
2096 page_cache_release(pages[do_pages]);
2098 if (unlikely(status))
2100 balance_dirty_pages_ratelimited(mapping);
2106 page_cache_release(cached_page);
2107 /* For now, when the user asks for O_SYNC, we actually give O_DSYNC. */
2108 if (likely(!status)) {
2109 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(vi))) {
2110 if (!mapping->a_ops->writepage || !is_sync_kiocb(iocb))
2111 status = generic_osync_inode(vi, mapping,
2112 OSYNC_METADATA|OSYNC_DATA);
2115 pagevec_lru_add(&lru_pvec);
2116 ntfs_debug("Done. Returning %s (written 0x%lx, status %li).",
2117 written ? "written" : "status", (unsigned long)written,
2119 return written ? written : status;
2123 * ntfs_file_aio_write_nolock -
2125 static ssize_t ntfs_file_aio_write_nolock(struct kiocb *iocb,
2126 const struct iovec *iov, unsigned long nr_segs, loff_t *ppos)
2128 struct file *file = iocb->ki_filp;
2129 struct address_space *mapping = file->f_mapping;
2130 struct inode *inode = mapping->host;
2133 size_t count; /* after file limit checks */
2134 ssize_t written, err;
2137 for (seg = 0; seg < nr_segs; seg++) {
2138 const struct iovec *iv = &iov[seg];
2140 * If any segment has a negative length, or the cumulative
2141 * length ever wraps negative then return -EINVAL.
2143 count += iv->iov_len;
2144 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
2146 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
2151 count -= iv->iov_len; /* This segment is no good */
2155 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2156 /* We can write back this queue in page reclaim. */
2157 current->backing_dev_info = mapping->backing_dev_info;
2159 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2164 err = remove_suid(file->f_dentry);
2167 inode_update_time(inode, 1);
2168 written = ntfs_file_buffered_write(iocb, iov, nr_segs, pos, ppos,
2171 current->backing_dev_info = NULL;
2172 return written ? written : err;
2176 * ntfs_file_aio_write -
2178 static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const char __user *buf,
2179 size_t count, loff_t pos)
2181 struct file *file = iocb->ki_filp;
2182 struct address_space *mapping = file->f_mapping;
2183 struct inode *inode = mapping->host;
2185 struct iovec local_iov = { .iov_base = (void __user *)buf,
2188 BUG_ON(iocb->ki_pos != pos);
2190 down(&inode->i_sem);
2191 ret = ntfs_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
2193 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2194 int err = sync_page_range(inode, mapping, pos, ret);
2202 * ntfs_file_writev -
2204 * Basically the same as generic_file_writev() except that it ends up calling
2205 * ntfs_file_aio_write_nolock() instead of __generic_file_aio_write_nolock().
2207 static ssize_t ntfs_file_writev(struct file *file, const struct iovec *iov,
2208 unsigned long nr_segs, loff_t *ppos)
2210 struct address_space *mapping = file->f_mapping;
2211 struct inode *inode = mapping->host;
2215 down(&inode->i_sem);
2216 init_sync_kiocb(&kiocb, file);
2217 ret = ntfs_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos);
2218 if (ret == -EIOCBQUEUED)
2219 ret = wait_on_sync_kiocb(&kiocb);
2221 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2222 int err = sync_page_range(inode, mapping, *ppos - ret, ret);
2230 * ntfs_file_write - simple wrapper for ntfs_file_writev()
2232 static ssize_t ntfs_file_write(struct file *file, const char __user *buf,
2233 size_t count, loff_t *ppos)
2235 struct iovec local_iov = { .iov_base = (void __user *)buf,
2238 return ntfs_file_writev(file, &local_iov, 1, ppos);
2242 * ntfs_file_fsync - sync a file to disk
2243 * @filp: file to be synced
2244 * @dentry: dentry describing the file to sync
2245 * @datasync: if non-zero only flush user data and not metadata
2247 * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync
2248 * system calls. This function is inspired by fs/buffer.c::file_fsync().
2250 * If @datasync is false, write the mft record and all associated extent mft
2251 * records as well as the $DATA attribute and then sync the block device.
2253 * If @datasync is true and the attribute is non-resident, we skip the writing
2254 * of the mft record and all associated extent mft records (this might still
2255 * happen due to the write_inode_now() call).
2257 * Also, if @datasync is true, we do not wait on the inode to be written out
2258 * but we always wait on the page cache pages to be written out.
2260 * Note: In the past @filp could be NULL so we ignore it as we don't need it
2263 * Locking: Caller must hold i_sem on the inode.
2265 * TODO: We should probably also write all attribute/index inodes associated
2266 * with this inode but since we have no simple way of getting to them we ignore
2267 * this problem for now.
2269 static int ntfs_file_fsync(struct file *filp, struct dentry *dentry,
2272 struct inode *vi = dentry->d_inode;
2275 ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2276 BUG_ON(S_ISDIR(vi->i_mode));
2277 if (!datasync || !NInoNonResident(NTFS_I(vi)))
2278 ret = ntfs_write_inode(vi, 1);
2279 write_inode_now(vi, !datasync);
2281 * NOTE: If we were to use mapping->private_list (see ext2 and
2282 * fs/buffer.c) for dirty blocks then we could optimize the below to be
2283 * sync_mapping_buffers(vi->i_mapping).
2285 err = sync_blockdev(vi->i_sb->s_bdev);
2286 if (unlikely(err && !ret))
2289 ntfs_debug("Done.");
2291 ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error "
2292 "%u.", datasync ? "data" : "", vi->i_ino, -ret);
2296 #endif /* NTFS_RW */
2298 struct file_operations ntfs_file_ops = {
2299 .llseek = generic_file_llseek, /* Seek inside file. */
2300 .read = generic_file_read, /* Read from file. */
2301 .aio_read = generic_file_aio_read, /* Async read from file. */
2302 .readv = generic_file_readv, /* Read from file. */
2304 .write = ntfs_file_write, /* Write to file. */
2305 .aio_write = ntfs_file_aio_write, /* Async write to file. */
2306 .writev = ntfs_file_writev, /* Write to file. */
2307 /*.release = ,*/ /* Last file is closed. See
2309 ext2_release_file() for
2310 how to use this to discard
2311 preallocated space for
2312 write opened files. */
2313 .fsync = ntfs_file_fsync, /* Sync a file to disk. */
2314 /*.aio_fsync = ,*/ /* Sync all outstanding async
2317 #endif /* NTFS_RW */
2318 /*.ioctl = ,*/ /* Perform function on the
2319 mounted filesystem. */
2320 .mmap = generic_file_mmap, /* Mmap file. */
2321 .open = ntfs_file_open, /* Open file. */
2322 .sendfile = generic_file_sendfile, /* Zero-copy data send with
2323 the data source being on
2324 the ntfs partition. We do
2325 not need to care about the
2326 data destination. */
2327 /*.sendpage = ,*/ /* Zero-copy data send with
2328 the data destination being
2329 on the ntfs partition. We
2330 do not need to care about
2334 struct inode_operations ntfs_file_inode_ops = {
2336 .truncate = ntfs_truncate_vfs,
2337 .setattr = ntfs_setattr,
2338 #endif /* NTFS_RW */
2341 struct file_operations ntfs_empty_file_ops = {};
2343 struct inode_operations ntfs_empty_inode_ops = {};