| blob | 9d383e5eff0ea5519bffb34528b5aeccbe6f3831 |
1 /*
2 * file.c - NTFS kernel file operations. Part of the Linux-NTFS project.
3 *
4 * Copyright (c) 2001-2015 Anton Altaparmakov and Tuxera Inc.
5 *
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.
10 *
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.
15 *
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
20 */
22 #include <linux/backing-dev.h>
23 #include <linux/buffer_head.h>
24 #include <linux/gfp.h>
25 #include <linux/pagemap.h>
26 #include <linux/pagevec.h>
27 #include <linux/sched.h>
28 #include <linux/swap.h>
29 #include <linux/uio.h>
30 #include <linux/writeback.h>
32 #include <asm/page.h>
33 #include <asm/uaccess.h>
44 /**
45 * ntfs_file_open - called when an inode is about to be opened
46 * @vi: inode to be opened
47 * @filp: file structure describing the inode
48 *
49 * Limit file size to the page cache limit on architectures where unsigned long
50 * is 32-bits. This is the most we can do for now without overflowing the page
51 * cache page index. Doing it this way means we don't run into problems because
52 * of existing too large files. It would be better to allow the user to read
53 * the beginning of the file but I doubt very much anyone is going to hit this
54 * check on a 32-bit architecture, so there is no point in adding the extra
55 * complexity required to support this.
56 *
57 * On 64-bit architectures, the check is hopefully optimized away by the
58 * compiler.
59 *
60 * After the check passes, just call generic_file_open() to do its work.
61 */
63 {
67 }
69 }
71 #ifdef NTFS_RW
73 /**
74 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
75 * @ni: ntfs inode of the attribute to extend
76 * @new_init_size: requested new initialized size in bytes
77 *
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 already uptodate in the page cache then
82 * these are simply marked dirty).
83 *
84 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
85 * in the resident attribute case, it is tied to the initialized size and, in
86 * the non-resident attribute case, it may not fall below the initialized size.
87 *
88 * Note that if the attribute is resident, we do not need to touch the page
89 * cache at all. This is because if the page cache page is not uptodate we
90 * bring it uptodate later, when doing the write to the mft record since we
91 * then already have the page mapped. And if the page is uptodate, the
92 * non-initialized region will already have been zeroed when the page was
93 * brought uptodate and the region may in fact already have been overwritten
94 * with new data via mmap() based writes, so we cannot just zero it. And since
95 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
96 * is unspecified, we choose not to do zeroing and thus we do not need to touch
97 * the page at all. For a more detailed explanation see ntfs_truncate() in
98 * fs/ntfs/inode.c.
99 *
100 * Return 0 on success and -errno on error. In the case that an error is
101 * encountered it is possible that the initialized size will already have been
102 * incremented some way towards @new_init_size but it is guaranteed that if
103 * this is the case, the necessary zeroing will also have happened and that all
104 * metadata is self-consistent.
105 *
106 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
107 * held by the caller.
108 */
110 {
111 s64 old_init_size;
112 loff_t old_i_size;
124 u32 attr_len;
132 "old_initialized_size 0x%llx, "
139 else
141 /* Use goto to reduce indentation and we need the label below anyway. */
150 }
155 }
162 }
166 /* The total length of the attribute value. */
169 /*
170 * Do the zeroing in the mft record and update the attribute size in
171 * the mft record.
172 */
176 /* Finally, update the sizes in the vfs and ntfs inodes. */
182 do_non_resident_extend:
183 /*
184 * If the new initialized size @new_init_size exceeds the current file
185 * size (vfs inode->i_size), we need to extend the file size to the
186 * new initialized size.
187 */
194 }
199 }
206 }
215 /* Update the file size in the vfs inode. */
221 }
226 /*
227 * Read the page. If the page is not present, this will zero
228 * the uninitialized regions for us.
229 */
234 }
239 }
240 /*
241 * Update the initialized size in the ntfs inode. This is
242 * enough to make ntfs_writepage() work.
243 */
249 /* Set the page dirty so it gets written out. */
252 /*
253 * Play nice with the vm and the rest of the system. This is
254 * very much needed as we can potentially be modifying the
255 * initialised size from a very small value to a really huge
256 * value, e.g.
257 * f = open(somefile, O_TRUNC);
258 * truncate(f, 10GiB);
259 * seek(f, 10GiB);
260 * write(f, 1);
261 * And this would mean we would be marking dirty hundreds of
262 * thousands of pages or as in the above example more than
263 * two and a half million pages!
264 *
265 * TODO: For sparse pages could optimize this workload by using
266 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This
267 * would be set in readpage for sparse pages and here we would
268 * not need to mark dirty any pages which have this bit set.
269 * The only caveat is that we have to clear the bit everywhere
270 * where we allocate any clusters that lie in the page or that
271 * contain the page.
272 *
273 * TODO: An even greater optimization would be for us to only
274 * call readpage() on pages which are not in sparse regions as
275 * determined from the runlist. This would greatly reduce the
276 * number of pages we read and make dirty in the case of sparse
277 * files.
278 */
285 /* Now bring in sync the initialized_size in the mft record. */
291 }
296 }
303 }
308 done:
318 init_err_out:
322 err_out:
329 }
333 {
334 loff_t pos;
336 ssize_t err;
351 /*
352 * All checks have passed. Before we start doing any writing we want
353 * to abort any totally illegal writes.
354 */
357 /* If file is encrypted, deny access, just like NT4. */
359 /* Only $DATA attributes can be encrypted. */
360 /*
361 * Reminder for later: Encrypted files are _always_
362 * non-resident so that the content can always be encrypted.
363 */
367 }
369 /* Only unnamed $DATA attribute can be compressed. */
371 /*
372 * Reminder for later: If resident, the data is not actually
373 * compressed. Only on the switch to non-resident does
374 * compression kick in. This is in contrast to encrypted files
375 * (see above).
376 */
381 }
388 /*
389 * Our ->update_time method always succeeds thus file_update_time()
390 * cannot fail either so there is no need to check the return code.
391 */
394 /* The first byte after the last cluster being written to. */
397 /*
398 * If the write goes beyond the allocated size, extend the allocation
399 * to cover the whole of the write, rounded up to the nearest cluster.
400 */
405 /*
406 * Extend the allocation without changing the data size.
407 *
408 * Note we ensure the allocation is big enough to at least
409 * write some data but we do not require the allocation to be
410 * complete, i.e. it may be partial.
411 */
415 /* If the extension was partial truncate the write. */
418 "attribute type 0x%x, because "
419 "the allocation was only "
424 }
430 /* Perform a partial write if possible or fail. */
433 "attribute type 0x%x, because "
434 "extending the allocation "
443 "write to inode "
444 "0x%lx, attribute "
445 "type 0x%x, because "
446 "extending the "
447 "allocation failed "
452 else
454 "inode 0x%lx, "
455 "attribute type 0x%x, "
456 "because there is not "
461 }
462 }
463 }
464 /*
465 * If the write starts beyond the initialized size, extend it up to the
466 * beginning of the write and initialize all non-sparse space between
467 * the old initialized size and the new one. This automatically also
468 * increments the vfs inode->i_size to keep it above or equal to the
469 * initialized_size.
470 */
475 /*
476 * Wait for ongoing direct i/o to complete before proceeding.
477 * New direct i/o cannot start as we hold i_mutex.
478 */
483 "0x%lx, attribute type 0x%x, because "
484 "extending the initialized size "
488 }
489 out:
491 }
493 /**
494 * __ntfs_grab_cache_pages - obtain a number of locked pages
495 * @mapping: address space mapping from which to obtain page cache pages
496 * @index: starting index in @mapping at which to begin obtaining pages
497 * @nr_pages: number of page cache pages to obtain
498 * @pages: array of pages in which to return the obtained page cache pages
499 * @cached_page: allocated but as yet unused page
500 *
501 * Obtain @nr_pages locked page cache pages from the mapping @mapping and
502 * starting at index @index.
503 *
504 * If a page is newly created, add it to lru list
505 *
506 * Note, the page locks are obtained in ascending page index order.
507 */
511 {
518 FGP_ACCESSED);
525 }
526 }
528 index,
534 }
537 }
538 index++;
539 nr++;
541 out:
543 err_out:
547 }
549 }
552 {
557 }
559 /**
560 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
561 * @pages: array of destination pages
562 * @nr_pages: number of pages in @pages
563 * @pos: byte position in file at which the write begins
564 * @bytes: number of bytes to be written
565 *
566 * This is called for non-resident attributes from ntfs_file_buffered_write()
567 * with i_mutex held on the inode (@pages[0]->mapping->host). There are
568 * @nr_pages pages in @pages which are locked but not kmap()ped. The source
569 * data has not yet been copied into the @pages.
570 *
571 * Need to fill any holes with actual clusters, allocate buffers if necessary,
572 * ensure all the buffers are mapped, and bring uptodate any buffers that are
573 * only partially being written to.
574 *
575 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
576 * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside
577 * the same cluster and that they are the entirety of that cluster, and that
578 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
579 *
580 * i_size is not to be modified yet.
581 *
582 * Return 0 on success or -errno on error.
583 */
586 {
588 LCN lcn;
590 sector_t lcn_block;
629 /*
630 * create_empty_buffers() will create uptodate/dirty buffers if
631 * the page is uptodate/dirty.
632 */
637 }
649 /*
650 * Loop over each page and for each page over each buffer. Use goto to
651 * reduce indentation.
652 */
654 do_next_page:
659 VCN cdelta;
660 s64 bh_end;
663 /* Clear buffer_new on all buffers to reinitialise state. */
670 /*
671 * The buffer is already mapped. If it is uptodate,
672 * ignore it.
673 */
676 /*
677 * The buffer is not uptodate. If the page is uptodate
678 * set the buffer uptodate and otherwise ignore it.
679 */
683 }
684 /*
685 * Neither the page nor the buffer are uptodate. If
686 * the buffer is only partially being written to, we
687 * need to read it in before the write, i.e. now.
688 */
691 /*
692 * If the buffer is fully or partially within
693 * the initialized size, do an actual read.
694 * Otherwise, simply zero the buffer.
695 */
704 blocksize);
706 }
707 }
709 }
710 /* Unmapped buffer. Need to map it. */
712 /*
713 * If the current buffer is in the same clusters as the map
714 * cache, there is no need to check the runlist again. The
715 * map cache is made up of @vcn, which is the first cached file
716 * cluster, @vcn_len which is the number of cached file
717 * clusters, @lcn is the device cluster corresponding to @vcn,
718 * and @lcn_block is the block number corresponding to @lcn.
719 */
722 map_buffer_cached:
726 blocksize_bits)) +
729 /*
730 * If the page is uptodate so is the buffer. If the
731 * buffer is fully outside the write, we ignore it if
732 * it was already allocated and we mark it dirty so it
733 * gets written out if we allocated it. On the other
734 * hand, if we allocated the buffer but we are not
735 * marking it dirty we set buffer_new so we can do
736 * error recovery.
737 */
742 /* We allocated the buffer. */
747 else
749 }
751 }
752 /* Page is _not_ uptodate. */
754 /*
755 * Buffer was already allocated. If it is not
756 * uptodate and is only partially being written
757 * to, we need to read it in before the write,
758 * i.e. now.
759 */
764 /*
765 * If the buffer is fully or partially
766 * within the initialized size, do an
767 * actual read. Otherwise, simply zero
768 * the buffer.
769 */
771 flags);
774 flags);
780 blocksize);
782 }
783 }
785 }
786 /* We allocated the buffer. */
788 /*
789 * If the buffer is fully outside the write, zero it,
790 * set it uptodate, and mark it dirty so it gets
791 * written out. If it is partially being written to,
792 * zero region surrounding the write but leave it to
793 * commit write to do anything else. Finally, if the
794 * buffer is fully being overwritten, do nothing.
795 */
799 blocksize);
801 }
804 }
815 }
819 }
822 }
824 }
825 /*
826 * Slow path: this is the first buffer in the cluster. If it
827 * is outside allocated size and is not uptodate, zero it and
828 * set it uptodate.
829 */
840 }
842 }
846 retry_remap:
848 }
850 /* Seek to element containing target cluster. */
852 rl++;
855 /*
856 * Successful remap, setup the map cache and
857 * use that to deal with the buffer.
858 */
863 blocksize_bits);
865 /*
866 * If the number of remaining clusters touched
867 * by the write is smaller or equal to the
868 * number of cached clusters, unlock the
869 * runlist as the map cache will be used from
870 * now on.
871 */
879 }
881 }
884 /*
885 * If it is not a hole and not out of bounds, the runlist is
886 * probably unmapped so try to map it now.
887 */
890 /* Attempt to map runlist. */
892 /*
893 * We need the runlist locked for
894 * writing, so if it is locked for
895 * reading relock it now and retry in
896 * case it changed whilst we dropped
897 * the lock.
898 */
903 }
905 NULL);
909 }
910 /*
911 * If @vcn is out of bounds, pretend @lcn is
912 * LCN_ENOENT. As long as the buffer is out
913 * of bounds this will work fine.
914 */
919 }
922 /* Failed to map the buffer, even after retrying. */
925 "attribute type 0x%x, vcn 0x%llx, "
926 "vcn offset 0x%x, because its "
927 "location on disk could not be "
934 err);
936 }
937 rl_not_mapped_enoent:
938 /*
939 * The buffer is in a hole or out of bounds. We need to fill
940 * the hole, unless the buffer is in a cluster which is not
941 * touched by the write, in which case we just leave the buffer
942 * unmapped. This can only happen when the cluster size is
943 * less than the page cache size.
944 */
950 /*
951 * If the buffer is uptodate we skip it. If it
952 * is not but the page is uptodate, we can set
953 * the buffer uptodate. If the page is not
954 * uptodate, we can clear the buffer and set it
955 * uptodate. Whether this is worthwhile is
956 * debatable and this could be removed.
957 */
963 blocksize);
965 }
967 }
968 }
969 /*
970 * Out of bounds buffer is invalid if it was not really out of
971 * bounds.
972 */
974 /*
975 * We need the runlist locked for writing, so if it is locked
976 * for reading relock it now and retry in case it changed
977 * whilst we dropped the lock.
978 */
985 }
986 /* Find the previous last allocated cluster. */
994 }
995 }
1001 err);
1003 }
1012 "allocated cluster in error "
1013 "code path. Run chkdsk to "
1016 }
1019 }
1024 /* Map and lock the mft record and get the attribute record. */
1027 else
1033 }
1039 }
1047 }
1050 /*
1051 * Find the runlist element with which the attribute extent
1052 * starts. Note, we cannot use the _attr_ version because we
1053 * have mapped the mft record. That is ok because we know the
1054 * runlist fragment must be mapped already to have ever gotten
1055 * here, so we can just use the _rl_ version.
1056 */
1063 /*
1064 * If @highest_vcn is zero, calculate the real highest_vcn
1065 * (which can really be zero).
1066 */
1071 /*
1072 * Determine the size of the mapping pairs array for the new
1073 * extent, i.e. the old extent with the hole filled.
1074 */
1076 highest_vcn);
1083 }
1084 /*
1085 * Resize the attribute record to fit the new mapping pairs
1086 * array.
1087 */
1093 // TODO: Deal with this by using the current attribute
1094 // and fill it with as much of the mapping pairs
1095 // array as possible. Then loop over each attribute
1096 // extent rewriting the mapping pairs arrays as we go
1097 // along and if when we reach the end we have not
1098 // enough space, try to resize the last attribute
1099 // extent and if even that fails, add a new attribute
1100 // extent.
1101 // We could also try to resize at each step in the hope
1102 // that we will not need to rewrite every single extent.
1103 // Note, we may need to decompress some extents to fill
1104 // the runlist as we are walking the extents...
1106 "record for the extended attribute "
1107 "record. This case is not "
1111 }
1113 /*
1114 * Generate the mapping pairs array directly into the attribute
1115 * record.
1116 */
1122 "attribute type 0x%x, because building "
1123 "the mapping pairs failed with error "
1128 }
1129 /* Update the highest_vcn but only if it was not set. */
1133 /*
1134 * If the attribute is sparse/compressed, update the compressed
1135 * size in the ntfs_inode structure and the attribute record.
1136 */
1138 /*
1139 * If we are not in the first attribute extent, switch
1140 * to it, but first ensure the changes will make it to
1141 * disk later.
1142 */
1153 }
1154 /* @m is not used any more so do not set it. */
1156 }
1162 }
1163 /* Ensure the changes make it to disk. */
1168 /* Successfully filled the hole. */
1172 /* Setup the map cache and use that to deal with the buffer. */
1178 /*
1179 * If the number of remaining clusters in the @pages is smaller
1180 * or equal to the number of cached clusters, unlock the
1181 * runlist as the map cache will be used from now on.
1182 */
1187 }
1190 /* If there are no errors, do the next page. */
1193 /* If there are no errors, release the runlist lock if we took it. */
1201 }
1202 /* If we issued read requests, let them complete. */
1213 /*
1214 * If the buffer overflows the initialized size, need
1215 * to zero the overflowing region.
1216 */
1223 blocksize);
1224 }
1227 }
1229 /* Clear buffer_new on all buffers. */
1240 }
1242 /* Get back to the attribute extent we modified. */
1247 "attribute extent of attribute in "
1248 "error code path. Run chkdsk to "
1255 /*
1256 * The only thing that is now wrong is the compressed
1257 * size of the base attribute extent which chkdsk
1258 * should be able to fix.
1259 */
1265 }
1266 }
1267 /*
1268 * If the runlist has been modified, need to restore it by punching a
1269 * hole into it and we then need to deallocate the on-disk cluster as
1270 * well. Note, we only modify the runlist if we are able to generate a
1271 * new mapping pairs array, i.e. only when the mapped attribute extent
1272 * is not switched.
1273 */
1276 /* Make the file cluster we allocated sparse in the runlist. */
1279 "attribute runlist in error code "
1280 "path. Run chkdsk to recover the "
1285 /*
1286 * Deallocate the on-disk cluster we allocated but only
1287 * if we succeeded in punching its vcn out of the
1288 * runlist.
1289 */
1293 "allocated cluster in error "
1294 "code path. Run chkdsk to "
1297 }
1299 }
1300 }
1301 /*
1302 * Resize the attribute record to its old size and rebuild the mapping
1303 * pairs array. Note, we only can do this if the runlist has been
1304 * restored to its old state which also implies that the mapped
1305 * attribute extent is not switched.
1306 */
1310 "record in error code path. Run "
1321 "mapping pairs array in error "
1322 "code path. Run chkdsk to "
1325 }
1328 }
1329 }
1330 /* Release the mft record and the attribute. */
1334 }
1335 /* Release the runlist lock. */
1340 /*
1341 * Zero out any newly allocated blocks to avoid exposing stale data.
1342 * If BH_New is set, we know that the block was newly allocated above
1343 * and that it has not been fully zeroed and marked dirty yet.
1344 */
1364 blocksize);
1366 }
1367 }
1373 }
1377 {
1379 /*
1380 * Warning: Do not do the decrement at the same time as the call to
1381 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1382 * decrement never happens so the loop never terminates.
1383 */
1388 }
1390 /**
1391 * ntfs_commit_pages_after_non_resident_write - commit the received data
1392 * @pages: array of destination pages
1393 * @nr_pages: number of pages in @pages
1394 * @pos: byte position in file at which the write begins
1395 * @bytes: number of bytes to be written
1396 *
1397 * See description of ntfs_commit_pages_after_write(), below.
1398 */
1402 {
1420 s64 bh_pos;
1429 s64 bh_end;
1438 }
1440 /*
1441 * If all buffers are now uptodate but the page is not, set the
1442 * page uptodate.
1443 */
1447 /*
1448 * Finally, if we do not need to update initialized_size or i_size we
1449 * are finished.
1450 */
1457 }
1458 /*
1459 * Update initialized_size/i_size as appropriate, both in the inode and
1460 * the mft record.
1461 */
1464 else
1466 /* Map, pin, and lock the mft record. */
1473 }
1479 }
1486 }
1497 }
1499 /* Mark the mft record dirty, so it gets written back. */
1506 err_out:
1516 }
1518 /**
1519 * ntfs_commit_pages_after_write - commit the received data
1520 * @pages: array of destination pages
1521 * @nr_pages: number of pages in @pages
1522 * @pos: byte position in file at which the write begins
1523 * @bytes: number of bytes to be written
1524 *
1525 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1526 * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are
1527 * locked but not kmap()ped. The source data has already been copied into the
1528 * @page. ntfs_prepare_pages_for_non_resident_write() has been called before
1529 * the data was copied (for non-resident attributes only) and it returned
1530 * success.
1531 *
1532 * Need to set uptodate and mark dirty all buffers within the boundary of the
1533 * write. If all buffers in a page are uptodate we set the page uptodate, too.
1534 *
1535 * Setting the buffers dirty ensures that they get written out later when
1536 * ntfs_writepage() is invoked by the VM.
1537 *
1538 * Finally, we need to update i_size and initialized_size as appropriate both
1539 * in the inode and the mft record.
1540 *
1541 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1542 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1543 * page are uptodate, and updates i_size if the end of io is beyond i_size. In
1544 * that case, it also marks the inode dirty.
1545 *
1546 * If things have gone as outlined in
1547 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1548 * content modifications here for non-resident attributes. For resident
1549 * attributes we need to do the uptodate bringing here which we combine with
1550 * the copying into the mft record which means we save one atomic kmap.
1551 *
1552 * Return 0 on success or -errno on error.
1553 */
1556 {
1558 loff_t i_size;
1567 u32 attr_len;
1584 /*
1585 * Attribute is resident, implying it is not compressed, encrypted, or
1586 * sparse.
1587 */
1590 else
1593 /* Map, pin, and lock the mft record. */
1600 }
1605 }
1612 }
1615 /* The total length of the attribute value. */
1625 /* Copy the received data from the page to the mft record. */
1627 /* Update the attribute length if necessary. */
1631 }
1632 /*
1633 * If the page is not uptodate, bring the out of bounds area(s)
1634 * uptodate by copying data from the mft record to the page.
1635 */
1641 /* Zero the region outside the end of the attribute value. */
1645 }
1647 /* Update initialized_size/i_size if necessary. */
1658 }
1659 /* Mark the mft record dirty, so it gets written back. */
1666 err_out:
1672 "dirty so the write will be retried "
1674 /*
1675 * Put the page on mapping->dirty_pages, but leave its
1676 * buffers' dirty state as-is.
1677 */
1687 }
1693 }
1695 /*
1696 * Copy as much as we can into the pages and return the number of bytes which
1697 * were successfully copied. If a fault is encountered then clear the pages
1698 * out to (ofs + bytes) and return the number of bytes which were copied.
1699 */
1702 {
1713 len);
1723 out:
1725 err:
1726 /* Zero the rest of the target like __copy_from_user(). */
1737 }
1739 /**
1740 * ntfs_perform_write - perform buffered write to a file
1741 * @file: file to write to
1742 * @i: iov_iter with data to write
1743 * @pos: byte offset in file at which to begin writing to
1744 */
1746 loff_t pos)
1747 {
1754 VCN last_vcn;
1755 LCN lcn;
1765 /*
1766 * If a previous ntfs_truncate() failed, repeat it and abort if it
1767 * fails again.
1768 */
1778 "0x%lx, attribute type 0x%x, because "
1779 "ntfs_truncate() failed (error code "
1783 }
1784 }
1785 /*
1786 * Determine the number of pages per cluster for non-resident
1787 * attributes.
1788 */
1794 VCN vcn;
1807 /*
1808 * Get the lcn of the vcn the write is in. If
1809 * it is a hole, need to lock down all pages in
1810 * the cluster.
1811 */
1822 "perform write to "
1823 "inode 0x%lx, "
1824 "attribute type 0x%x, "
1825 "because the attribute "
1829 }
1831 }
1839 }
1840 }
1841 }
1844 again:
1845 /*
1846 * Bring in the user page(s) that we will copy from _first_.
1847 * Otherwise there is a nasty deadlock on copying from the same
1848 * page(s) as we are writing to, without it/them being marked
1849 * up-to-date. Note, at present there is nothing to stop the
1850 * pages being swapped out between us bringing them into memory
1851 * and doing the actual copying.
1852 */
1856 }
1857 /* Get and lock @do_pages starting at index @start_idx. */
1862 /*
1863 * For non-resident attributes, we need to fill any holes with
1864 * actual clusters and ensure all bufferes are mapped. We also
1865 * need to bring uptodate any buffers that are only partially
1866 * being written to.
1867 */
1877 }
1878 }
1889 }
1901 /*
1902 * We failed to copy anything. Fall back to single
1903 * segment length write.
1904 *
1905 * This is needed to avoid possible livelock in the
1906 * case that all segments in the iov cannot be copied
1907 * at once without a pagefault.
1908 */
1913 }
1921 }
1929 }
1931 /**
1932 * ntfs_file_write_iter - simple wrapper for ntfs_file_write_iter_nolock()
1933 * @iocb: IO state structure
1934 * @from: iov_iter with data to write
1935 *
1936 * Basically the same as generic_file_write_iter() except that it ends up
1937 * up calling ntfs_perform_write() instead of generic_perform_write() and that
1938 * O_DIRECT is not implemented.
1939 */
1941 {
1945 ssize_t err;
1948 /* We can write back this queue in page reclaim. */
1959 }
1962 }
1964 /**
1965 * ntfs_file_fsync - sync a file to disk
1966 * @filp: file to be synced
1967 * @datasync: if non-zero only flush user data and not metadata
1968 *
1969 * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync
1970 * system calls. This function is inspired by fs/buffer.c::file_fsync().
1971 *
1972 * If @datasync is false, write the mft record and all associated extent mft
1973 * records as well as the $DATA attribute and then sync the block device.
1974 *
1975 * If @datasync is true and the attribute is non-resident, we skip the writing
1976 * of the mft record and all associated extent mft records (this might still
1977 * happen due to the write_inode_now() call).
1978 *
1979 * Also, if @datasync is true, we do not wait on the inode to be written out
1980 * but we always wait on the page cache pages to be written out.
1981 *
1982 * Locking: Caller must hold i_mutex on the inode.
1983 *
1984 * TODO: We should probably also write all attribute/index inodes associated
1985 * with this inode but since we have no simple way of getting to them we ignore
1986 * this problem for now.
1987 */
1990 {
2005 /*
2006 * NOTE: If we were to use mapping->private_list (see ext2 and
2007 * fs/buffer.c) for dirty blocks then we could optimize the below to be
2008 * sync_mapping_buffers(vi->i_mapping).
2009 */
2015 else
2020 }
2027 #ifdef NTFS_RW
2034 };
2037 #ifdef NTFS_RW
2040 };