| blob | f42967b738eb677eb55ce9a9c37b5f92ae688463 |
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * file.c - NTFS kernel file operations. Part of the Linux-NTFS project.
4 *
5 * Copyright (c) 2001-2015 Anton Altaparmakov and Tuxera Inc.
6 */
8 #include <linux/backing-dev.h>
9 #include <linux/buffer_head.h>
10 #include <linux/gfp.h>
11 #include <linux/pagemap.h>
12 #include <linux/pagevec.h>
13 #include <linux/sched/signal.h>
14 #include <linux/swap.h>
15 #include <linux/uio.h>
16 #include <linux/writeback.h>
18 #include <asm/page.h>
19 #include <linux/uaccess.h>
30 /**
31 * ntfs_file_open - called when an inode is about to be opened
32 * @vi: inode to be opened
33 * @filp: file structure describing the inode
34 *
35 * Limit file size to the page cache limit on architectures where unsigned long
36 * is 32-bits. This is the most we can do for now without overflowing the page
37 * cache page index. Doing it this way means we don't run into problems because
38 * of existing too large files. It would be better to allow the user to read
39 * the beginning of the file but I doubt very much anyone is going to hit this
40 * check on a 32-bit architecture, so there is no point in adding the extra
41 * complexity required to support this.
42 *
43 * On 64-bit architectures, the check is hopefully optimized away by the
44 * compiler.
45 *
46 * After the check passes, just call generic_file_open() to do its work.
47 */
49 {
53 }
55 }
57 #ifdef NTFS_RW
59 /**
60 * ntfs_attr_extend_initialized - extend the initialized size of an attribute
61 * @ni: ntfs inode of the attribute to extend
62 * @new_init_size: requested new initialized size in bytes
63 *
64 * Extend the initialized size of an attribute described by the ntfs inode @ni
65 * to @new_init_size bytes. This involves zeroing any non-sparse space between
66 * the old initialized size and @new_init_size both in the page cache and on
67 * disk (if relevant complete pages are already uptodate in the page cache then
68 * these are simply marked dirty).
69 *
70 * As a side-effect, the file size (vfs inode->i_size) may be incremented as,
71 * in the resident attribute case, it is tied to the initialized size and, in
72 * the non-resident attribute case, it may not fall below the initialized size.
73 *
74 * Note that if the attribute is resident, we do not need to touch the page
75 * cache at all. This is because if the page cache page is not uptodate we
76 * bring it uptodate later, when doing the write to the mft record since we
77 * then already have the page mapped. And if the page is uptodate, the
78 * non-initialized region will already have been zeroed when the page was
79 * brought uptodate and the region may in fact already have been overwritten
80 * with new data via mmap() based writes, so we cannot just zero it. And since
81 * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped
82 * is unspecified, we choose not to do zeroing and thus we do not need to touch
83 * the page at all. For a more detailed explanation see ntfs_truncate() in
84 * fs/ntfs/inode.c.
85 *
86 * Return 0 on success and -errno on error. In the case that an error is
87 * encountered it is possible that the initialized size will already have been
88 * incremented some way towards @new_init_size but it is guaranteed that if
89 * this is the case, the necessary zeroing will also have happened and that all
90 * metadata is self-consistent.
91 *
92 * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be
93 * held by the caller.
94 */
96 {
97 s64 old_init_size;
98 loff_t old_i_size;
110 u32 attr_len;
118 "old_initialized_size 0x%llx, "
125 else
127 /* Use goto to reduce indentation and we need the label below anyway. */
136 }
141 }
148 }
152 /* The total length of the attribute value. */
155 /*
156 * Do the zeroing in the mft record and update the attribute size in
157 * the mft record.
158 */
162 /* Finally, update the sizes in the vfs and ntfs inodes. */
168 do_non_resident_extend:
169 /*
170 * If the new initialized size @new_init_size exceeds the current file
171 * size (vfs inode->i_size), we need to extend the file size to the
172 * new initialized size.
173 */
180 }
185 }
192 }
201 /* Update the file size in the vfs inode. */
207 }
212 /*
213 * Read the page. If the page is not present, this will zero
214 * the uninitialized regions for us.
215 */
220 }
225 }
226 /*
227 * Update the initialized size in the ntfs inode. This is
228 * enough to make ntfs_writepage() work.
229 */
235 /* Set the page dirty so it gets written out. */
238 /*
239 * Play nice with the vm and the rest of the system. This is
240 * very much needed as we can potentially be modifying the
241 * initialised size from a very small value to a really huge
242 * value, e.g.
243 * f = open(somefile, O_TRUNC);
244 * truncate(f, 10GiB);
245 * seek(f, 10GiB);
246 * write(f, 1);
247 * And this would mean we would be marking dirty hundreds of
248 * thousands of pages or as in the above example more than
249 * two and a half million pages!
250 *
251 * TODO: For sparse pages could optimize this workload by using
252 * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This
253 * would be set in readpage for sparse pages and here we would
254 * not need to mark dirty any pages which have this bit set.
255 * The only caveat is that we have to clear the bit everywhere
256 * where we allocate any clusters that lie in the page or that
257 * contain the page.
258 *
259 * TODO: An even greater optimization would be for us to only
260 * call readpage() on pages which are not in sparse regions as
261 * determined from the runlist. This would greatly reduce the
262 * number of pages we read and make dirty in the case of sparse
263 * files.
264 */
271 /* Now bring in sync the initialized_size in the mft record. */
277 }
282 }
289 }
294 done:
304 init_err_out:
308 err_out:
315 }
319 {
320 loff_t pos;
322 ssize_t err;
337 /*
338 * All checks have passed. Before we start doing any writing we want
339 * to abort any totally illegal writes.
340 */
343 /* If file is encrypted, deny access, just like NT4. */
345 /* Only $DATA attributes can be encrypted. */
346 /*
347 * Reminder for later: Encrypted files are _always_
348 * non-resident so that the content can always be encrypted.
349 */
353 }
355 /* Only unnamed $DATA attribute can be compressed. */
357 /*
358 * Reminder for later: If resident, the data is not actually
359 * compressed. Only on the switch to non-resident does
360 * compression kick in. This is in contrast to encrypted files
361 * (see above).
362 */
367 }
374 /*
375 * Our ->update_time method always succeeds thus file_update_time()
376 * cannot fail either so there is no need to check the return code.
377 */
380 /* The first byte after the last cluster being written to. */
383 /*
384 * If the write goes beyond the allocated size, extend the allocation
385 * to cover the whole of the write, rounded up to the nearest cluster.
386 */
391 /*
392 * Extend the allocation without changing the data size.
393 *
394 * Note we ensure the allocation is big enough to at least
395 * write some data but we do not require the allocation to be
396 * complete, i.e. it may be partial.
397 */
401 /* If the extension was partial truncate the write. */
404 "attribute type 0x%x, because "
405 "the allocation was only "
410 }
416 /* Perform a partial write if possible or fail. */
419 "attribute type 0x%x, because "
420 "extending the allocation "
429 "write to inode "
430 "0x%lx, attribute "
431 "type 0x%x, because "
432 "extending the "
433 "allocation failed "
438 else
440 "inode 0x%lx, "
441 "attribute type 0x%x, "
442 "because there is not "
447 }
448 }
449 }
450 /*
451 * If the write starts beyond the initialized size, extend it up to the
452 * beginning of the write and initialize all non-sparse space between
453 * the old initialized size and the new one. This automatically also
454 * increments the vfs inode->i_size to keep it above or equal to the
455 * initialized_size.
456 */
461 /*
462 * Wait for ongoing direct i/o to complete before proceeding.
463 * New direct i/o cannot start as we hold i_mutex.
464 */
469 "0x%lx, attribute type 0x%x, because "
470 "extending the initialized size "
474 }
475 out:
477 }
479 /**
480 * __ntfs_grab_cache_pages - obtain a number of locked pages
481 * @mapping: address space mapping from which to obtain page cache pages
482 * @index: starting index in @mapping at which to begin obtaining pages
483 * @nr_pages: number of page cache pages to obtain
484 * @pages: array of pages in which to return the obtained page cache pages
485 * @cached_page: allocated but as yet unused page
486 *
487 * Obtain @nr_pages locked page cache pages from the mapping @mapping and
488 * starting at index @index.
489 *
490 * If a page is newly created, add it to lru list
491 *
492 * Note, the page locks are obtained in ascending page index order.
493 */
497 {
504 FGP_ACCESSED);
511 }
512 }
514 index,
520 }
523 }
524 index++;
525 nr++;
527 out:
529 err_out:
533 }
535 }
538 {
543 }
545 /**
546 * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data
547 * @pages: array of destination pages
548 * @nr_pages: number of pages in @pages
549 * @pos: byte position in file at which the write begins
550 * @bytes: number of bytes to be written
551 *
552 * This is called for non-resident attributes from ntfs_file_buffered_write()
553 * with i_mutex held on the inode (@pages[0]->mapping->host). There are
554 * @nr_pages pages in @pages which are locked but not kmap()ped. The source
555 * data has not yet been copied into the @pages.
556 *
557 * Need to fill any holes with actual clusters, allocate buffers if necessary,
558 * ensure all the buffers are mapped, and bring uptodate any buffers that are
559 * only partially being written to.
560 *
561 * If @nr_pages is greater than one, we are guaranteed that the cluster size is
562 * greater than PAGE_SIZE, that all pages in @pages are entirely inside
563 * the same cluster and that they are the entirety of that cluster, and that
564 * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole.
565 *
566 * i_size is not to be modified yet.
567 *
568 * Return 0 on success or -errno on error.
569 */
572 {
574 LCN lcn;
576 sector_t lcn_block;
615 /*
616 * create_empty_buffers() will create uptodate/dirty buffers if
617 * the page is uptodate/dirty.
618 */
623 }
635 /*
636 * Loop over each page and for each page over each buffer. Use goto to
637 * reduce indentation.
638 */
640 do_next_page:
645 VCN cdelta;
646 s64 bh_end;
649 /* Clear buffer_new on all buffers to reinitialise state. */
656 /*
657 * The buffer is already mapped. If it is uptodate,
658 * ignore it.
659 */
662 /*
663 * The buffer is not uptodate. If the page is uptodate
664 * set the buffer uptodate and otherwise ignore it.
665 */
669 }
670 /*
671 * Neither the page nor the buffer are uptodate. If
672 * the buffer is only partially being written to, we
673 * need to read it in before the write, i.e. now.
674 */
677 /*
678 * If the buffer is fully or partially within
679 * the initialized size, do an actual read.
680 * Otherwise, simply zero the buffer.
681 */
690 blocksize);
692 }
693 }
695 }
696 /* Unmapped buffer. Need to map it. */
698 /*
699 * If the current buffer is in the same clusters as the map
700 * cache, there is no need to check the runlist again. The
701 * map cache is made up of @vcn, which is the first cached file
702 * cluster, @vcn_len which is the number of cached file
703 * clusters, @lcn is the device cluster corresponding to @vcn,
704 * and @lcn_block is the block number corresponding to @lcn.
705 */
708 map_buffer_cached:
712 blocksize_bits)) +
715 /*
716 * If the page is uptodate so is the buffer. If the
717 * buffer is fully outside the write, we ignore it if
718 * it was already allocated and we mark it dirty so it
719 * gets written out if we allocated it. On the other
720 * hand, if we allocated the buffer but we are not
721 * marking it dirty we set buffer_new so we can do
722 * error recovery.
723 */
728 /* We allocated the buffer. */
732 else
734 }
736 }
737 /* Page is _not_ uptodate. */
739 /*
740 * Buffer was already allocated. If it is not
741 * uptodate and is only partially being written
742 * to, we need to read it in before the write,
743 * i.e. now.
744 */
749 /*
750 * If the buffer is fully or partially
751 * within the initialized size, do an
752 * actual read. Otherwise, simply zero
753 * the buffer.
754 */
756 flags);
759 flags);
765 blocksize);
767 }
768 }
770 }
771 /* We allocated the buffer. */
773 /*
774 * If the buffer is fully outside the write, zero it,
775 * set it uptodate, and mark it dirty so it gets
776 * written out. If it is partially being written to,
777 * zero region surrounding the write but leave it to
778 * commit write to do anything else. Finally, if the
779 * buffer is fully being overwritten, do nothing.
780 */
784 blocksize);
786 }
789 }
800 }
804 }
807 }
809 }
810 /*
811 * Slow path: this is the first buffer in the cluster. If it
812 * is outside allocated size and is not uptodate, zero it and
813 * set it uptodate.
814 */
825 }
827 }
831 retry_remap:
833 }
835 /* Seek to element containing target cluster. */
837 rl++;
840 /*
841 * Successful remap, setup the map cache and
842 * use that to deal with the buffer.
843 */
848 blocksize_bits);
850 /*
851 * If the number of remaining clusters touched
852 * by the write is smaller or equal to the
853 * number of cached clusters, unlock the
854 * runlist as the map cache will be used from
855 * now on.
856 */
864 }
866 }
869 /*
870 * If it is not a hole and not out of bounds, the runlist is
871 * probably unmapped so try to map it now.
872 */
875 /* Attempt to map runlist. */
877 /*
878 * We need the runlist locked for
879 * writing, so if it is locked for
880 * reading relock it now and retry in
881 * case it changed whilst we dropped
882 * the lock.
883 */
888 }
890 NULL);
894 }
895 /*
896 * If @vcn is out of bounds, pretend @lcn is
897 * LCN_ENOENT. As long as the buffer is out
898 * of bounds this will work fine.
899 */
904 }
907 /* Failed to map the buffer, even after retrying. */
910 "attribute type 0x%x, vcn 0x%llx, "
911 "vcn offset 0x%x, because its "
912 "location on disk could not be "
919 err);
921 }
922 rl_not_mapped_enoent:
923 /*
924 * The buffer is in a hole or out of bounds. We need to fill
925 * the hole, unless the buffer is in a cluster which is not
926 * touched by the write, in which case we just leave the buffer
927 * unmapped. This can only happen when the cluster size is
928 * less than the page cache size.
929 */
935 /*
936 * If the buffer is uptodate we skip it. If it
937 * is not but the page is uptodate, we can set
938 * the buffer uptodate. If the page is not
939 * uptodate, we can clear the buffer and set it
940 * uptodate. Whether this is worthwhile is
941 * debatable and this could be removed.
942 */
948 blocksize);
950 }
952 }
953 }
954 /*
955 * Out of bounds buffer is invalid if it was not really out of
956 * bounds.
957 */
959 /*
960 * We need the runlist locked for writing, so if it is locked
961 * for reading relock it now and retry in case it changed
962 * whilst we dropped the lock.
963 */
970 }
971 /* Find the previous last allocated cluster. */
979 }
980 }
986 err);
988 }
997 "allocated cluster in error "
998 "code path. Run chkdsk to "
1001 }
1004 }
1009 /* Map and lock the mft record and get the attribute record. */
1012 else
1018 }
1024 }
1032 }
1035 /*
1036 * Find the runlist element with which the attribute extent
1037 * starts. Note, we cannot use the _attr_ version because we
1038 * have mapped the mft record. That is ok because we know the
1039 * runlist fragment must be mapped already to have ever gotten
1040 * here, so we can just use the _rl_ version.
1041 */
1048 /*
1049 * If @highest_vcn is zero, calculate the real highest_vcn
1050 * (which can really be zero).
1051 */
1056 /*
1057 * Determine the size of the mapping pairs array for the new
1058 * extent, i.e. the old extent with the hole filled.
1059 */
1061 highest_vcn);
1068 }
1069 /*
1070 * Resize the attribute record to fit the new mapping pairs
1071 * array.
1072 */
1078 // TODO: Deal with this by using the current attribute
1079 // and fill it with as much of the mapping pairs
1080 // array as possible. Then loop over each attribute
1081 // extent rewriting the mapping pairs arrays as we go
1082 // along and if when we reach the end we have not
1083 // enough space, try to resize the last attribute
1084 // extent and if even that fails, add a new attribute
1085 // extent.
1086 // We could also try to resize at each step in the hope
1087 // that we will not need to rewrite every single extent.
1088 // Note, we may need to decompress some extents to fill
1089 // the runlist as we are walking the extents...
1091 "record for the extended attribute "
1092 "record. This case is not "
1096 }
1098 /*
1099 * Generate the mapping pairs array directly into the attribute
1100 * record.
1101 */
1107 "attribute type 0x%x, because building "
1108 "the mapping pairs failed with error "
1113 }
1114 /* Update the highest_vcn but only if it was not set. */
1118 /*
1119 * If the attribute is sparse/compressed, update the compressed
1120 * size in the ntfs_inode structure and the attribute record.
1121 */
1123 /*
1124 * If we are not in the first attribute extent, switch
1125 * to it, but first ensure the changes will make it to
1126 * disk later.
1127 */
1138 }
1139 /* @m is not used any more so do not set it. */
1141 }
1147 }
1148 /* Ensure the changes make it to disk. */
1153 /* Successfully filled the hole. */
1157 /* Setup the map cache and use that to deal with the buffer. */
1163 /*
1164 * If the number of remaining clusters in the @pages is smaller
1165 * or equal to the number of cached clusters, unlock the
1166 * runlist as the map cache will be used from now on.
1167 */
1172 }
1175 /* If there are no errors, do the next page. */
1178 /* If there are no errors, release the runlist lock if we took it. */
1186 }
1187 /* If we issued read requests, let them complete. */
1198 /*
1199 * If the buffer overflows the initialized size, need
1200 * to zero the overflowing region.
1201 */
1208 blocksize);
1209 }
1212 }
1214 /* Clear buffer_new on all buffers. */
1225 }
1227 /* Get back to the attribute extent we modified. */
1232 "attribute extent of attribute in "
1233 "error code path. Run chkdsk to "
1240 /*
1241 * The only thing that is now wrong is the compressed
1242 * size of the base attribute extent which chkdsk
1243 * should be able to fix.
1244 */
1250 }
1251 }
1252 /*
1253 * If the runlist has been modified, need to restore it by punching a
1254 * hole into it and we then need to deallocate the on-disk cluster as
1255 * well. Note, we only modify the runlist if we are able to generate a
1256 * new mapping pairs array, i.e. only when the mapped attribute extent
1257 * is not switched.
1258 */
1261 /* Make the file cluster we allocated sparse in the runlist. */
1264 "attribute runlist in error code "
1265 "path. Run chkdsk to recover the "
1270 /*
1271 * Deallocate the on-disk cluster we allocated but only
1272 * if we succeeded in punching its vcn out of the
1273 * runlist.
1274 */
1278 "allocated cluster in error "
1279 "code path. Run chkdsk to "
1282 }
1284 }
1285 }
1286 /*
1287 * Resize the attribute record to its old size and rebuild the mapping
1288 * pairs array. Note, we only can do this if the runlist has been
1289 * restored to its old state which also implies that the mapped
1290 * attribute extent is not switched.
1291 */
1295 "record in error code path. Run "
1306 "mapping pairs array in error "
1307 "code path. Run chkdsk to "
1310 }
1313 }
1314 }
1315 /* Release the mft record and the attribute. */
1319 }
1320 /* Release the runlist lock. */
1325 /*
1326 * Zero out any newly allocated blocks to avoid exposing stale data.
1327 * If BH_New is set, we know that the block was newly allocated above
1328 * and that it has not been fully zeroed and marked dirty yet.
1329 */
1349 blocksize);
1351 }
1352 }
1358 }
1362 {
1364 /*
1365 * Warning: Do not do the decrement at the same time as the call to
1366 * flush_dcache_page() because it is a NULL macro on i386 and hence the
1367 * decrement never happens so the loop never terminates.
1368 */
1373 }
1375 /**
1376 * ntfs_commit_pages_after_non_resident_write - commit the received data
1377 * @pages: array of destination pages
1378 * @nr_pages: number of pages in @pages
1379 * @pos: byte position in file at which the write begins
1380 * @bytes: number of bytes to be written
1381 *
1382 * See description of ntfs_commit_pages_after_write(), below.
1383 */
1387 {
1405 s64 bh_pos;
1414 s64 bh_end;
1423 }
1425 /*
1426 * If all buffers are now uptodate but the page is not, set the
1427 * page uptodate.
1428 */
1432 /*
1433 * Finally, if we do not need to update initialized_size or i_size we
1434 * are finished.
1435 */
1442 }
1443 /*
1444 * Update initialized_size/i_size as appropriate, both in the inode and
1445 * the mft record.
1446 */
1449 else
1451 /* Map, pin, and lock the mft record. */
1458 }
1464 }
1471 }
1482 }
1484 /* Mark the mft record dirty, so it gets written back. */
1491 err_out:
1501 }
1503 /**
1504 * ntfs_commit_pages_after_write - commit the received data
1505 * @pages: array of destination pages
1506 * @nr_pages: number of pages in @pages
1507 * @pos: byte position in file at which the write begins
1508 * @bytes: number of bytes to be written
1509 *
1510 * This is called from ntfs_file_buffered_write() with i_mutex held on the inode
1511 * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are
1512 * locked but not kmap()ped. The source data has already been copied into the
1513 * @page. ntfs_prepare_pages_for_non_resident_write() has been called before
1514 * the data was copied (for non-resident attributes only) and it returned
1515 * success.
1516 *
1517 * Need to set uptodate and mark dirty all buffers within the boundary of the
1518 * write. If all buffers in a page are uptodate we set the page uptodate, too.
1519 *
1520 * Setting the buffers dirty ensures that they get written out later when
1521 * ntfs_writepage() is invoked by the VM.
1522 *
1523 * Finally, we need to update i_size and initialized_size as appropriate both
1524 * in the inode and the mft record.
1525 *
1526 * This is modelled after fs/buffer.c::generic_commit_write(), which marks
1527 * buffers uptodate and dirty, sets the page uptodate if all buffers in the
1528 * page are uptodate, and updates i_size if the end of io is beyond i_size. In
1529 * that case, it also marks the inode dirty.
1530 *
1531 * If things have gone as outlined in
1532 * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page
1533 * content modifications here for non-resident attributes. For resident
1534 * attributes we need to do the uptodate bringing here which we combine with
1535 * the copying into the mft record which means we save one atomic kmap.
1536 *
1537 * Return 0 on success or -errno on error.
1538 */
1541 {
1543 loff_t i_size;
1552 u32 attr_len;
1569 /*
1570 * Attribute is resident, implying it is not compressed, encrypted, or
1571 * sparse.
1572 */
1575 else
1578 /* Map, pin, and lock the mft record. */
1585 }
1590 }
1597 }
1600 /* The total length of the attribute value. */
1610 /* Copy the received data from the page to the mft record. */
1612 /* Update the attribute length if necessary. */
1616 }
1617 /*
1618 * If the page is not uptodate, bring the out of bounds area(s)
1619 * uptodate by copying data from the mft record to the page.
1620 */
1626 /* Zero the region outside the end of the attribute value. */
1630 }
1632 /* Update initialized_size/i_size if necessary. */
1643 }
1644 /* Mark the mft record dirty, so it gets written back. */
1651 err_out:
1657 "dirty so the write will be retried "
1659 /*
1660 * Put the page on mapping->dirty_pages, but leave its
1661 * buffers' dirty state as-is.
1662 */
1672 }
1678 }
1680 /*
1681 * Copy as much as we can into the pages and return the number of bytes which
1682 * were successfully copied. If a fault is encountered then clear the pages
1683 * out to (ofs + bytes) and return the number of bytes which were copied.
1684 */
1687 {
1698 len);
1708 out:
1710 err:
1711 /* Zero the rest of the target like __copy_from_user(). */
1722 }
1724 /**
1725 * ntfs_perform_write - perform buffered write to a file
1726 * @file: file to write to
1727 * @i: iov_iter with data to write
1728 * @pos: byte offset in file at which to begin writing to
1729 */
1731 loff_t pos)
1732 {
1739 VCN last_vcn;
1740 LCN lcn;
1750 /*
1751 * If a previous ntfs_truncate() failed, repeat it and abort if it
1752 * fails again.
1753 */
1763 "0x%lx, attribute type 0x%x, because "
1764 "ntfs_truncate() failed (error code "
1768 }
1769 }
1770 /*
1771 * Determine the number of pages per cluster for non-resident
1772 * attributes.
1773 */
1779 VCN vcn;
1792 /*
1793 * Get the lcn of the vcn the write is in. If
1794 * it is a hole, need to lock down all pages in
1795 * the cluster.
1796 */
1807 "perform write to "
1808 "inode 0x%lx, "
1809 "attribute type 0x%x, "
1810 "because the attribute "
1814 }
1816 }
1824 }
1825 }
1826 }
1829 again:
1830 /*
1831 * Bring in the user page(s) that we will copy from _first_.
1832 * Otherwise there is a nasty deadlock on copying from the same
1833 * page(s) as we are writing to, without it/them being marked
1834 * up-to-date. Note, at present there is nothing to stop the
1835 * pages being swapped out between us bringing them into memory
1836 * and doing the actual copying.
1837 */
1841 }
1842 /* Get and lock @do_pages starting at index @start_idx. */
1847 /*
1848 * For non-resident attributes, we need to fill any holes with
1849 * actual clusters and ensure all bufferes are mapped. We also
1850 * need to bring uptodate any buffers that are only partially
1851 * being written to.
1852 */
1862 }
1863 }
1874 }
1886 /*
1887 * We failed to copy anything. Fall back to single
1888 * segment length write.
1889 *
1890 * This is needed to avoid possible livelock in the
1891 * case that all segments in the iov cannot be copied
1892 * at once without a pagefault.
1893 */
1898 }
1906 }
1914 }
1916 /**
1917 * ntfs_file_write_iter - simple wrapper for ntfs_file_write_iter_nolock()
1918 * @iocb: IO state structure
1919 * @from: iov_iter with data to write
1920 *
1921 * Basically the same as generic_file_write_iter() except that it ends up
1922 * up calling ntfs_perform_write() instead of generic_perform_write() and that
1923 * O_DIRECT is not implemented.
1924 */
1926 {
1930 ssize_t err;
1933 /* We can write back this queue in page reclaim. */
1944 }
1946 /**
1947 * ntfs_file_fsync - sync a file to disk
1948 * @filp: file to be synced
1949 * @datasync: if non-zero only flush user data and not metadata
1950 *
1951 * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync
1952 * system calls. This function is inspired by fs/buffer.c::file_fsync().
1953 *
1954 * If @datasync is false, write the mft record and all associated extent mft
1955 * records as well as the $DATA attribute and then sync the block device.
1956 *
1957 * If @datasync is true and the attribute is non-resident, we skip the writing
1958 * of the mft record and all associated extent mft records (this might still
1959 * happen due to the write_inode_now() call).
1960 *
1961 * Also, if @datasync is true, we do not wait on the inode to be written out
1962 * but we always wait on the page cache pages to be written out.
1963 *
1964 * Locking: Caller must hold i_mutex on the inode.
1965 *
1966 * TODO: We should probably also write all attribute/index inodes associated
1967 * with this inode but since we have no simple way of getting to them we ignore
1968 * this problem for now.
1969 */
1972 {
1987 /*
1988 * NOTE: If we were to use mapping->private_list (see ext2 and
1989 * fs/buffer.c) for dirty blocks then we could optimize the below to be
1990 * sync_mapping_buffers(vi->i_mapping).
1991 */
1997 else
2002 }
2009 #ifdef NTFS_RW
2016 };
2019 #ifdef NTFS_RW
2022 };