2 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
3 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
4 * Akira Fujita <a-fujita@rs.jp.nec.com>
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2.1 of the GNU Lesser General Public License
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
17 #include <linux/quotaops.h>
18 #include "ext4_jbd2.h"
19 #include "ext4_extents.h"
23 * get_ext_path - Find an extent path for designated logical block number.
25 * @inode: an inode which is searched
26 * @lblock: logical block number to find an extent path
27 * @path: pointer to an extent path pointer (for output)
29 * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
33 get_ext_path(struct inode
*inode
, ext4_lblk_t lblock
,
34 struct ext4_ext_path
**path
)
38 *path
= ext4_ext_find_extent(inode
, lblock
, *path
);
42 } else if ((*path
)[ext_depth(inode
)].p_ext
== NULL
)
49 * copy_extent_status - Copy the extent's initialization status
51 * @src: an extent for getting initialize status
52 * @dest: an extent to be set the status
55 copy_extent_status(struct ext4_extent
*src
, struct ext4_extent
*dest
)
57 if (ext4_ext_is_uninitialized(src
))
58 ext4_ext_mark_uninitialized(dest
);
60 dest
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(dest
));
64 * mext_next_extent - Search for the next extent and set it to "extent"
66 * @inode: inode which is searched
67 * @path: this will obtain data for the next extent
68 * @extent: pointer to the next extent we have just gotten
70 * Search the next extent in the array of ext4_ext_path structure (@path)
71 * and set it to ext4_extent structure (@extent). In addition, the member of
72 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
73 * ext4_ext_path structure refers to the last extent, or a negative error
77 mext_next_extent(struct inode
*inode
, struct ext4_ext_path
*path
,
78 struct ext4_extent
**extent
)
80 int ppos
, leaf_ppos
= path
->p_depth
;
83 if (EXT_LAST_EXTENT(path
[ppos
].p_hdr
) > path
[ppos
].p_ext
) {
85 *extent
= ++path
[ppos
].p_ext
;
90 if (EXT_LAST_INDEX(path
[ppos
].p_hdr
) >
96 path
[ppos
].p_block
= idx_pblock(path
[ppos
].p_idx
);
97 if (path
[ppos
+1].p_bh
)
98 brelse(path
[ppos
+1].p_bh
);
100 sb_bread(inode
->i_sb
, path
[ppos
].p_block
);
101 if (!path
[ppos
+1].p_bh
)
104 ext_block_hdr(path
[ppos
+1].p_bh
);
106 /* Halfway index block */
107 while (++cur_ppos
< leaf_ppos
) {
108 path
[cur_ppos
].p_idx
=
109 EXT_FIRST_INDEX(path
[cur_ppos
].p_hdr
);
110 path
[cur_ppos
].p_block
=
111 idx_pblock(path
[cur_ppos
].p_idx
);
112 if (path
[cur_ppos
+1].p_bh
)
113 brelse(path
[cur_ppos
+1].p_bh
);
114 path
[cur_ppos
+1].p_bh
= sb_bread(inode
->i_sb
,
115 path
[cur_ppos
].p_block
);
116 if (!path
[cur_ppos
+1].p_bh
)
118 path
[cur_ppos
+1].p_hdr
=
119 ext_block_hdr(path
[cur_ppos
+1].p_bh
);
123 path
[leaf_ppos
].p_ext
= *extent
=
124 EXT_FIRST_EXTENT(path
[leaf_ppos
].p_hdr
);
128 /* We found the last extent */
133 * mext_check_null_inode - NULL check for two inodes
135 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
138 mext_check_null_inode(struct inode
*inode1
, struct inode
*inode2
,
139 const char *function
)
143 if (inode1
== NULL
) {
144 ext4_error(inode2
->i_sb
, function
,
145 "Both inodes should not be NULL: "
146 "inode1 NULL inode2 %lu", inode2
->i_ino
);
148 } else if (inode2
== NULL
) {
149 ext4_error(inode1
->i_sb
, function
,
150 "Both inodes should not be NULL: "
151 "inode1 %lu inode2 NULL", inode1
->i_ino
);
158 * mext_double_down_read - Acquire two inodes' read semaphore
160 * @orig_inode: original inode structure
161 * @donor_inode: donor inode structure
162 * Acquire read semaphore of the two inodes (orig and donor) by i_ino order.
165 mext_double_down_read(struct inode
*orig_inode
, struct inode
*donor_inode
)
167 struct inode
*first
= orig_inode
, *second
= donor_inode
;
170 * Use the inode number to provide the stable locking order instead
171 * of its address, because the C language doesn't guarantee you can
172 * compare pointers that don't come from the same array.
174 if (donor_inode
->i_ino
< orig_inode
->i_ino
) {
179 down_read(&EXT4_I(first
)->i_data_sem
);
180 down_read(&EXT4_I(second
)->i_data_sem
);
184 * mext_double_down_write - Acquire two inodes' write semaphore
186 * @orig_inode: original inode structure
187 * @donor_inode: donor inode structure
188 * Acquire write semaphore of the two inodes (orig and donor) by i_ino order.
191 mext_double_down_write(struct inode
*orig_inode
, struct inode
*donor_inode
)
193 struct inode
*first
= orig_inode
, *second
= donor_inode
;
196 * Use the inode number to provide the stable locking order instead
197 * of its address, because the C language doesn't guarantee you can
198 * compare pointers that don't come from the same array.
200 if (donor_inode
->i_ino
< orig_inode
->i_ino
) {
205 down_write(&EXT4_I(first
)->i_data_sem
);
206 down_write(&EXT4_I(second
)->i_data_sem
);
210 * mext_double_up_read - Release two inodes' read semaphore
212 * @orig_inode: original inode structure to be released its lock first
213 * @donor_inode: donor inode structure to be released its lock second
214 * Release read semaphore of two inodes (orig and donor).
217 mext_double_up_read(struct inode
*orig_inode
, struct inode
*donor_inode
)
219 up_read(&EXT4_I(orig_inode
)->i_data_sem
);
220 up_read(&EXT4_I(donor_inode
)->i_data_sem
);
224 * mext_double_up_write - Release two inodes' write semaphore
226 * @orig_inode: original inode structure to be released its lock first
227 * @donor_inode: donor inode structure to be released its lock second
228 * Release write semaphore of two inodes (orig and donor).
231 mext_double_up_write(struct inode
*orig_inode
, struct inode
*donor_inode
)
233 up_write(&EXT4_I(orig_inode
)->i_data_sem
);
234 up_write(&EXT4_I(donor_inode
)->i_data_sem
);
238 * mext_insert_across_blocks - Insert extents across leaf block
240 * @handle: journal handle
241 * @orig_inode: original inode
242 * @o_start: first original extent to be changed
243 * @o_end: last original extent to be changed
244 * @start_ext: first new extent to be inserted
245 * @new_ext: middle of new extent to be inserted
246 * @end_ext: last new extent to be inserted
248 * Allocate a new leaf block and insert extents into it. Return 0 on success,
249 * or a negative error value on failure.
252 mext_insert_across_blocks(handle_t
*handle
, struct inode
*orig_inode
,
253 struct ext4_extent
*o_start
, struct ext4_extent
*o_end
,
254 struct ext4_extent
*start_ext
, struct ext4_extent
*new_ext
,
255 struct ext4_extent
*end_ext
)
257 struct ext4_ext_path
*orig_path
= NULL
;
258 ext4_lblk_t eblock
= 0;
263 if (start_ext
->ee_len
&& new_ext
->ee_len
&& end_ext
->ee_len
) {
264 if (o_start
== o_end
) {
266 /* start_ext new_ext end_ext
267 * donor |---------|-----------|--------|
268 * orig |------------------------------|
273 /* start_ext new_ext end_ext
274 * donor |---------|----------|---------|
275 * orig |---------------|--------------|
277 o_end
->ee_block
= end_ext
->ee_block
;
278 o_end
->ee_len
= end_ext
->ee_len
;
279 ext4_ext_store_pblock(o_end
, ext_pblock(end_ext
));
282 o_start
->ee_len
= start_ext
->ee_len
;
285 } else if (start_ext
->ee_len
&& new_ext
->ee_len
&&
286 !end_ext
->ee_len
&& o_start
== o_end
) {
289 * donor |--------------|---------------|
290 * orig |------------------------------|
292 o_start
->ee_len
= start_ext
->ee_len
;
295 } else if (!start_ext
->ee_len
&& new_ext
->ee_len
&&
296 end_ext
->ee_len
&& o_start
== o_end
) {
299 * donor |--------------|---------------|
300 * orig |------------------------------|
302 o_end
->ee_block
= end_ext
->ee_block
;
303 o_end
->ee_len
= end_ext
->ee_len
;
304 ext4_ext_store_pblock(o_end
, ext_pblock(end_ext
));
307 * Set 0 to the extent block if new_ext was
310 if (new_ext
->ee_block
)
311 eblock
= le32_to_cpu(new_ext
->ee_block
);
315 ext4_debug("ext4 move extent: Unexpected insert case\n");
320 err
= get_ext_path(orig_inode
, eblock
, &orig_path
);
324 if (ext4_ext_insert_extent(handle
, orig_inode
,
325 orig_path
, new_ext
, 0))
330 err
= get_ext_path(orig_inode
,
331 le32_to_cpu(end_ext
->ee_block
) - 1, &orig_path
);
335 if (ext4_ext_insert_extent(handle
, orig_inode
,
336 orig_path
, end_ext
, 0))
341 ext4_ext_drop_refs(orig_path
);
350 * mext_insert_inside_block - Insert new extent to the extent block
352 * @o_start: first original extent to be moved
353 * @o_end: last original extent to be moved
354 * @start_ext: first new extent to be inserted
355 * @new_ext: middle of new extent to be inserted
356 * @end_ext: last new extent to be inserted
357 * @eh: extent header of target leaf block
358 * @range_to_move: used to decide how to insert extent
360 * Insert extents into the leaf block. The extent (@o_start) is overwritten
361 * by inserted extents.
364 mext_insert_inside_block(struct ext4_extent
*o_start
,
365 struct ext4_extent
*o_end
,
366 struct ext4_extent
*start_ext
,
367 struct ext4_extent
*new_ext
,
368 struct ext4_extent
*end_ext
,
369 struct ext4_extent_header
*eh
,
375 /* Move the existing extents */
376 if (range_to_move
&& o_end
< EXT_LAST_EXTENT(eh
)) {
377 len
= (unsigned long)(EXT_LAST_EXTENT(eh
) + 1) -
378 (unsigned long)(o_end
+ 1);
379 memmove(o_end
+ 1 + range_to_move
, o_end
+ 1, len
);
382 /* Insert start entry */
383 if (start_ext
->ee_len
)
384 o_start
[i
++].ee_len
= start_ext
->ee_len
;
386 /* Insert new entry */
387 if (new_ext
->ee_len
) {
388 o_start
[i
] = *new_ext
;
389 ext4_ext_store_pblock(&o_start
[i
++], ext_pblock(new_ext
));
392 /* Insert end entry */
394 o_start
[i
] = *end_ext
;
396 /* Increment the total entries counter on the extent block */
397 le16_add_cpu(&eh
->eh_entries
, range_to_move
);
401 * mext_insert_extents - Insert new extent
403 * @handle: journal handle
404 * @orig_inode: original inode
405 * @orig_path: path indicates first extent to be changed
406 * @o_start: first original extent to be changed
407 * @o_end: last original extent to be changed
408 * @start_ext: first new extent to be inserted
409 * @new_ext: middle of new extent to be inserted
410 * @end_ext: last new extent to be inserted
412 * Call the function to insert extents. If we cannot add more extents into
413 * the leaf block, we call mext_insert_across_blocks() to create a
414 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
415 * on success, or a negative error value on failure.
418 mext_insert_extents(handle_t
*handle
, struct inode
*orig_inode
,
419 struct ext4_ext_path
*orig_path
,
420 struct ext4_extent
*o_start
,
421 struct ext4_extent
*o_end
,
422 struct ext4_extent
*start_ext
,
423 struct ext4_extent
*new_ext
,
424 struct ext4_extent
*end_ext
)
426 struct ext4_extent_header
*eh
;
427 unsigned long need_slots
, slots_range
;
428 int range_to_move
, depth
, ret
;
431 * The extents need to be inserted
432 * start_extent + new_extent + end_extent.
434 need_slots
= (start_ext
->ee_len
? 1 : 0) + (end_ext
->ee_len
? 1 : 0) +
435 (new_ext
->ee_len
? 1 : 0);
437 /* The number of slots between start and end */
438 slots_range
= ((unsigned long)(o_end
+ 1) - (unsigned long)o_start
+ 1)
439 / sizeof(struct ext4_extent
);
441 /* Range to move the end of extent */
442 range_to_move
= need_slots
- slots_range
;
443 depth
= orig_path
->p_depth
;
445 eh
= orig_path
->p_hdr
;
448 /* Register to journal */
449 ret
= ext4_journal_get_write_access(handle
, orig_path
->p_bh
);
455 if (range_to_move
> 0 &&
456 (range_to_move
> le16_to_cpu(eh
->eh_max
)
457 - le16_to_cpu(eh
->eh_entries
))) {
459 ret
= mext_insert_across_blocks(handle
, orig_inode
, o_start
,
460 o_end
, start_ext
, new_ext
, end_ext
);
464 mext_insert_inside_block(o_start
, o_end
, start_ext
, new_ext
,
465 end_ext
, eh
, range_to_move
);
468 ret
= ext4_handle_dirty_metadata(handle
, orig_inode
,
473 ret
= ext4_mark_inode_dirty(handle
, orig_inode
);
482 * mext_leaf_block - Move one leaf extent block into the inode.
484 * @handle: journal handle
485 * @orig_inode: original inode
486 * @orig_path: path indicates first extent to be changed
487 * @dext: donor extent
488 * @from: start offset on the target file
490 * In order to insert extents into the leaf block, we must divide the extent
491 * in the leaf block into three extents. The one is located to be inserted
492 * extents, and the others are located around it.
494 * Therefore, this function creates structures to save extents of the leaf
495 * block, and inserts extents by calling mext_insert_extents() with
496 * created extents. Return 0 on success, or a negative error value on failure.
499 mext_leaf_block(handle_t
*handle
, struct inode
*orig_inode
,
500 struct ext4_ext_path
*orig_path
, struct ext4_extent
*dext
,
503 struct ext4_extent
*oext
, *o_start
, *o_end
, *prev_ext
;
504 struct ext4_extent new_ext
, start_ext
, end_ext
;
505 ext4_lblk_t new_ext_end
;
506 ext4_fsblk_t new_phys_end
;
507 int oext_alen
, new_ext_alen
, end_ext_alen
;
508 int depth
= ext_depth(orig_inode
);
511 o_start
= o_end
= oext
= orig_path
[depth
].p_ext
;
512 oext_alen
= ext4_ext_get_actual_len(oext
);
513 start_ext
.ee_len
= end_ext
.ee_len
= 0;
515 new_ext
.ee_block
= cpu_to_le32(*from
);
516 ext4_ext_store_pblock(&new_ext
, ext_pblock(dext
));
517 new_ext
.ee_len
= dext
->ee_len
;
518 new_ext_alen
= ext4_ext_get_actual_len(&new_ext
);
519 new_ext_end
= le32_to_cpu(new_ext
.ee_block
) + new_ext_alen
- 1;
520 new_phys_end
= ext_pblock(&new_ext
) + new_ext_alen
- 1;
523 * Case: original extent is first
528 if (le32_to_cpu(oext
->ee_block
) < le32_to_cpu(new_ext
.ee_block
) &&
529 le32_to_cpu(new_ext
.ee_block
) <
530 le32_to_cpu(oext
->ee_block
) + oext_alen
) {
531 start_ext
.ee_len
= cpu_to_le16(le32_to_cpu(new_ext
.ee_block
) -
532 le32_to_cpu(oext
->ee_block
));
533 copy_extent_status(oext
, &start_ext
);
534 } else if (oext
> EXT_FIRST_EXTENT(orig_path
[depth
].p_hdr
)) {
537 * We can merge new_ext into previous extent,
538 * if these are contiguous and same extent type.
540 if (ext4_can_extents_be_merged(orig_inode
, prev_ext
,
543 start_ext
.ee_len
= cpu_to_le16(
544 ext4_ext_get_actual_len(prev_ext
) +
546 copy_extent_status(prev_ext
, &start_ext
);
552 * Case: new_ext_end must be less than oext
556 if (le32_to_cpu(oext
->ee_block
) + oext_alen
- 1 < new_ext_end
) {
557 ext4_error(orig_inode
->i_sb
, __func__
,
558 "new_ext_end(%u) should be less than or equal to "
559 "oext->ee_block(%u) + oext_alen(%d) - 1",
560 new_ext_end
, le32_to_cpu(oext
->ee_block
),
567 * Case: new_ext is smaller than original extent
568 * oext |---------------|
569 * new_ext |-----------|
572 if (le32_to_cpu(oext
->ee_block
) <= new_ext_end
&&
573 new_ext_end
< le32_to_cpu(oext
->ee_block
) + oext_alen
- 1) {
575 cpu_to_le16(le32_to_cpu(oext
->ee_block
) +
576 oext_alen
- 1 - new_ext_end
);
577 copy_extent_status(oext
, &end_ext
);
578 end_ext_alen
= ext4_ext_get_actual_len(&end_ext
);
579 ext4_ext_store_pblock(&end_ext
,
580 (ext_pblock(o_end
) + oext_alen
- end_ext_alen
));
582 cpu_to_le32(le32_to_cpu(o_end
->ee_block
) +
583 oext_alen
- end_ext_alen
);
586 ret
= mext_insert_extents(handle
, orig_inode
, orig_path
, o_start
,
587 o_end
, &start_ext
, &new_ext
, &end_ext
);
593 * mext_calc_swap_extents - Calculate extents for extent swapping.
595 * @tmp_dext: the extent that will belong to the original inode
596 * @tmp_oext: the extent that will belong to the donor inode
597 * @orig_off: block offset of original inode
598 * @donor_off: block offset of donor inode
599 * @max_count: the maximun length of extents
601 * Return 0 on success, or a negative error value on failure.
604 mext_calc_swap_extents(struct ext4_extent
*tmp_dext
,
605 struct ext4_extent
*tmp_oext
,
606 ext4_lblk_t orig_off
, ext4_lblk_t donor_off
,
607 ext4_lblk_t max_count
)
609 ext4_lblk_t diff
, orig_diff
;
610 struct ext4_extent dext_old
, oext_old
;
612 BUG_ON(orig_off
!= donor_off
);
614 /* original and donor extents have to cover the same block offset */
615 if (orig_off
< le32_to_cpu(tmp_oext
->ee_block
) ||
616 le32_to_cpu(tmp_oext
->ee_block
) +
617 ext4_ext_get_actual_len(tmp_oext
) - 1 < orig_off
)
620 if (orig_off
< le32_to_cpu(tmp_dext
->ee_block
) ||
621 le32_to_cpu(tmp_dext
->ee_block
) +
622 ext4_ext_get_actual_len(tmp_dext
) - 1 < orig_off
)
625 dext_old
= *tmp_dext
;
626 oext_old
= *tmp_oext
;
628 /* When tmp_dext is too large, pick up the target range. */
629 diff
= donor_off
- le32_to_cpu(tmp_dext
->ee_block
);
631 ext4_ext_store_pblock(tmp_dext
, ext_pblock(tmp_dext
) + diff
);
633 cpu_to_le32(le32_to_cpu(tmp_dext
->ee_block
) + diff
);
634 tmp_dext
->ee_len
= cpu_to_le16(le16_to_cpu(tmp_dext
->ee_len
) - diff
);
636 if (max_count
< ext4_ext_get_actual_len(tmp_dext
))
637 tmp_dext
->ee_len
= cpu_to_le16(max_count
);
639 orig_diff
= orig_off
- le32_to_cpu(tmp_oext
->ee_block
);
640 ext4_ext_store_pblock(tmp_oext
, ext_pblock(tmp_oext
) + orig_diff
);
642 /* Adjust extent length if donor extent is larger than orig */
643 if (ext4_ext_get_actual_len(tmp_dext
) >
644 ext4_ext_get_actual_len(tmp_oext
) - orig_diff
)
645 tmp_dext
->ee_len
= cpu_to_le16(le16_to_cpu(tmp_oext
->ee_len
) -
648 tmp_oext
->ee_len
= cpu_to_le16(ext4_ext_get_actual_len(tmp_dext
));
650 copy_extent_status(&oext_old
, tmp_dext
);
651 copy_extent_status(&dext_old
, tmp_oext
);
657 * mext_replace_branches - Replace original extents with new extents
659 * @handle: journal handle
660 * @orig_inode: original inode
661 * @donor_inode: donor inode
662 * @from: block offset of orig_inode
663 * @count: block count to be replaced
665 * Replace original inode extents and donor inode extents page by page.
666 * We implement this replacement in the following three steps:
667 * 1. Save the block information of original and donor inodes into
669 * 2. Change the block information of original inode to point at the
670 * donor inode blocks.
671 * 3. Change the block information of donor inode to point at the saved
672 * original inode blocks in the dummy extents.
674 * Return 0 on success, or a negative error value on failure.
677 mext_replace_branches(handle_t
*handle
, struct inode
*orig_inode
,
678 struct inode
*donor_inode
, ext4_lblk_t from
,
681 struct ext4_ext_path
*orig_path
= NULL
;
682 struct ext4_ext_path
*donor_path
= NULL
;
683 struct ext4_extent
*oext
, *dext
;
684 struct ext4_extent tmp_dext
, tmp_oext
;
685 ext4_lblk_t orig_off
= from
, donor_off
= from
;
688 int replaced_count
= 0;
691 mext_double_down_write(orig_inode
, donor_inode
);
693 /* Get the original extent for the block "orig_off" */
694 err
= get_ext_path(orig_inode
, orig_off
, &orig_path
);
698 /* Get the donor extent for the head */
699 err
= get_ext_path(donor_inode
, donor_off
, &donor_path
);
702 depth
= ext_depth(orig_inode
);
703 oext
= orig_path
[depth
].p_ext
;
706 depth
= ext_depth(donor_inode
);
707 dext
= donor_path
[depth
].p_ext
;
710 err
= mext_calc_swap_extents(&tmp_dext
, &tmp_oext
, orig_off
,
715 /* Loop for the donor extents */
717 /* The extent for donor must be found. */
719 ext4_error(donor_inode
->i_sb
, __func__
,
720 "The extent for donor must be found");
723 } else if (donor_off
!= le32_to_cpu(tmp_dext
.ee_block
)) {
724 ext4_error(donor_inode
->i_sb
, __func__
,
725 "Donor offset(%u) and the first block of donor "
726 "extent(%u) should be equal",
728 le32_to_cpu(tmp_dext
.ee_block
));
733 /* Set donor extent to orig extent */
734 err
= mext_leaf_block(handle
, orig_inode
,
735 orig_path
, &tmp_dext
, &orig_off
);
739 /* Set orig extent to donor extent */
740 err
= mext_leaf_block(handle
, donor_inode
,
741 donor_path
, &tmp_oext
, &donor_off
);
745 dext_alen
= ext4_ext_get_actual_len(&tmp_dext
);
746 replaced_count
+= dext_alen
;
747 donor_off
+= dext_alen
;
748 orig_off
+= dext_alen
;
750 /* Already moved the expected blocks */
751 if (replaced_count
>= count
)
755 ext4_ext_drop_refs(orig_path
);
756 err
= get_ext_path(orig_inode
, orig_off
, &orig_path
);
759 depth
= ext_depth(orig_inode
);
760 oext
= orig_path
[depth
].p_ext
;
761 if (le32_to_cpu(oext
->ee_block
) +
762 ext4_ext_get_actual_len(oext
) <= orig_off
) {
769 ext4_ext_drop_refs(donor_path
);
770 err
= get_ext_path(donor_inode
, donor_off
, &donor_path
);
773 depth
= ext_depth(donor_inode
);
774 dext
= donor_path
[depth
].p_ext
;
775 if (le32_to_cpu(dext
->ee_block
) +
776 ext4_ext_get_actual_len(dext
) <= donor_off
) {
782 err
= mext_calc_swap_extents(&tmp_dext
, &tmp_oext
, orig_off
,
783 donor_off
, count
- replaced_count
);
790 ext4_ext_drop_refs(orig_path
);
794 ext4_ext_drop_refs(donor_path
);
798 mext_double_up_write(orig_inode
, donor_inode
);
803 * move_extent_per_page - Move extent data per page
805 * @o_filp: file structure of original file
806 * @donor_inode: donor inode
807 * @orig_page_offset: page index on original file
808 * @data_offset_in_page: block index where data swapping starts
809 * @block_len_in_page: the number of blocks to be swapped
810 * @uninit: orig extent is uninitialized or not
812 * Save the data in original inode blocks and replace original inode extents
813 * with donor inode extents by calling mext_replace_branches().
814 * Finally, write out the saved data in new original inode blocks. Return 0
815 * on success, or a negative error value on failure.
818 move_extent_per_page(struct file
*o_filp
, struct inode
*donor_inode
,
819 pgoff_t orig_page_offset
, int data_offset_in_page
,
820 int block_len_in_page
, int uninit
)
822 struct inode
*orig_inode
= o_filp
->f_dentry
->d_inode
;
823 struct address_space
*mapping
= orig_inode
->i_mapping
;
824 struct buffer_head
*bh
;
825 struct page
*page
= NULL
;
826 const struct address_space_operations
*a_ops
= mapping
->a_ops
;
828 ext4_lblk_t orig_blk_offset
;
829 long long offs
= orig_page_offset
<< PAGE_CACHE_SHIFT
;
830 unsigned long blocksize
= orig_inode
->i_sb
->s_blocksize
;
831 unsigned int w_flags
= 0;
832 unsigned int tmp_data_len
, data_len
;
835 int blocks_per_page
= PAGE_CACHE_SIZE
>> orig_inode
->i_blkbits
;
838 * It needs twice the amount of ordinary journal buffers because
839 * inode and donor_inode may change each different metadata blocks.
841 jblocks
= ext4_writepage_trans_blocks(orig_inode
) * 2;
842 handle
= ext4_journal_start(orig_inode
, jblocks
);
843 if (IS_ERR(handle
)) {
844 ret
= PTR_ERR(handle
);
848 if (segment_eq(get_fs(), KERNEL_DS
))
849 w_flags
|= AOP_FLAG_UNINTERRUPTIBLE
;
851 orig_blk_offset
= orig_page_offset
* blocks_per_page
+
855 * If orig extent is uninitialized one,
856 * it's not necessary force the page into memory
857 * and then force it to be written out again.
858 * Just swap data blocks between orig and donor.
861 ret
= mext_replace_branches(handle
, orig_inode
,
862 donor_inode
, orig_blk_offset
,
865 /* Clear the inode cache not to refer to the old data */
866 ext4_ext_invalidate_cache(orig_inode
);
867 ext4_ext_invalidate_cache(donor_inode
);
871 offs
= (long long)orig_blk_offset
<< orig_inode
->i_blkbits
;
873 /* Calculate data_len */
874 if ((orig_blk_offset
+ block_len_in_page
- 1) ==
875 ((orig_inode
->i_size
- 1) >> orig_inode
->i_blkbits
)) {
876 /* Replace the last block */
877 tmp_data_len
= orig_inode
->i_size
& (blocksize
- 1);
879 * If data_len equal zero, it shows data_len is multiples of
880 * blocksize. So we set appropriate value.
882 if (tmp_data_len
== 0)
883 tmp_data_len
= blocksize
;
885 data_len
= tmp_data_len
+
886 ((block_len_in_page
- 1) << orig_inode
->i_blkbits
);
888 data_len
= block_len_in_page
<< orig_inode
->i_blkbits
;
891 ret
= a_ops
->write_begin(o_filp
, mapping
, offs
, data_len
, w_flags
,
893 if (unlikely(ret
< 0))
896 if (!PageUptodate(page
)) {
897 mapping
->a_ops
->readpage(o_filp
, page
);
902 * try_to_release_page() doesn't call releasepage in writeback mode.
903 * We should care about the order of writing to the same file
904 * by multiple move extent processes.
905 * It needs to call wait_on_page_writeback() to wait for the
906 * writeback of the page.
908 if (PageWriteback(page
))
909 wait_on_page_writeback(page
);
911 /* Release old bh and drop refs */
912 try_to_release_page(page
, 0);
914 ret
= mext_replace_branches(handle
, orig_inode
, donor_inode
,
915 orig_blk_offset
, block_len_in_page
);
919 /* Clear the inode cache not to refer to the old data */
920 ext4_ext_invalidate_cache(orig_inode
);
921 ext4_ext_invalidate_cache(donor_inode
);
923 if (!page_has_buffers(page
))
924 create_empty_buffers(page
, 1 << orig_inode
->i_blkbits
, 0);
926 bh
= page_buffers(page
);
927 for (i
= 0; i
< data_offset_in_page
; i
++)
928 bh
= bh
->b_this_page
;
930 for (i
= 0; i
< block_len_in_page
; i
++) {
931 ret
= ext4_get_block(orig_inode
,
932 (sector_t
)(orig_blk_offset
+ i
), bh
, 0);
936 if (bh
->b_this_page
!= NULL
)
937 bh
= bh
->b_this_page
;
940 ret
= a_ops
->write_end(o_filp
, mapping
, offs
, data_len
, data_len
,
945 if (unlikely(page
)) {
946 if (PageLocked(page
))
948 page_cache_release(page
);
949 ext4_journal_stop(handle
);
952 ext4_journal_stop(handle
);
954 return ret
< 0 ? ret
: 0;
958 * mext_check_argumants - Check whether move extent can be done
960 * @orig_inode: original inode
961 * @donor_inode: donor inode
962 * @orig_start: logical start offset in block for orig
963 * @donor_start: logical start offset in block for donor
964 * @len: the number of blocks to be moved
965 * @moved_len: moved block length
967 * Check the arguments of ext4_move_extents() whether the files can be
968 * exchanged with each other.
969 * Return 0 on success, or a negative error value on failure.
972 mext_check_arguments(struct inode
*orig_inode
,
973 struct inode
*donor_inode
, __u64 orig_start
,
974 __u64 donor_start
, __u64
*len
, __u64 moved_len
)
976 ext4_lblk_t orig_blocks
, donor_blocks
;
977 unsigned int blkbits
= orig_inode
->i_blkbits
;
978 unsigned int blocksize
= 1 << blkbits
;
980 /* Regular file check */
981 if (!S_ISREG(orig_inode
->i_mode
) || !S_ISREG(donor_inode
->i_mode
)) {
982 ext4_debug("ext4 move extent: The argument files should be "
983 "regular file [ino:orig %lu, donor %lu]\n",
984 orig_inode
->i_ino
, donor_inode
->i_ino
);
988 /* Ext4 move extent does not support swapfile */
989 if (IS_SWAPFILE(orig_inode
) || IS_SWAPFILE(donor_inode
)) {
990 ext4_debug("ext4 move extent: The argument files should "
991 "not be swapfile [ino:orig %lu, donor %lu]\n",
992 orig_inode
->i_ino
, donor_inode
->i_ino
);
996 /* Files should be in the same ext4 FS */
997 if (orig_inode
->i_sb
!= donor_inode
->i_sb
) {
998 ext4_debug("ext4 move extent: The argument files "
999 "should be in same FS [ino:orig %lu, donor %lu]\n",
1000 orig_inode
->i_ino
, donor_inode
->i_ino
);
1004 /* Ext4 move extent supports only extent based file */
1005 if (!(EXT4_I(orig_inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1006 ext4_debug("ext4 move extent: orig file is not extents "
1007 "based file [ino:orig %lu]\n", orig_inode
->i_ino
);
1009 } else if (!(EXT4_I(donor_inode
)->i_flags
& EXT4_EXTENTS_FL
)) {
1010 ext4_debug("ext4 move extent: donor file is not extents "
1011 "based file [ino:donor %lu]\n", donor_inode
->i_ino
);
1015 if ((!orig_inode
->i_size
) || (!donor_inode
->i_size
)) {
1016 ext4_debug("ext4 move extent: File size is 0 byte\n");
1020 /* Start offset should be same */
1021 if (orig_start
!= donor_start
) {
1022 ext4_debug("ext4 move extent: orig and donor's start "
1023 "offset are not same [ino:orig %lu, donor %lu]\n",
1024 orig_inode
->i_ino
, donor_inode
->i_ino
);
1029 ext4_debug("ext4 move extent: moved_len should be 0 "
1030 "[ino:orig %lu, donor %lu]\n", orig_inode
->i_ino
,
1031 donor_inode
->i_ino
);
1035 if ((orig_start
> EXT_MAX_BLOCK
) ||
1036 (donor_start
> EXT_MAX_BLOCK
) ||
1037 (*len
> EXT_MAX_BLOCK
) ||
1038 (orig_start
+ *len
> EXT_MAX_BLOCK
)) {
1039 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1040 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK
,
1041 orig_inode
->i_ino
, donor_inode
->i_ino
);
1045 if (orig_inode
->i_size
> donor_inode
->i_size
) {
1046 donor_blocks
= (donor_inode
->i_size
+ blocksize
- 1) >> blkbits
;
1047 /* TODO: eliminate this artificial restriction */
1048 if (orig_start
>= donor_blocks
) {
1049 ext4_debug("ext4 move extent: orig start offset "
1050 "[%llu] should be less than donor file blocks "
1051 "[%u] [ino:orig %lu, donor %lu]\n",
1052 orig_start
, donor_blocks
,
1053 orig_inode
->i_ino
, donor_inode
->i_ino
);
1057 /* TODO: eliminate this artificial restriction */
1058 if (orig_start
+ *len
> donor_blocks
) {
1059 ext4_debug("ext4 move extent: End offset [%llu] should "
1060 "be less than donor file blocks [%u]."
1061 "So adjust length from %llu to %llu "
1062 "[ino:orig %lu, donor %lu]\n",
1063 orig_start
+ *len
, donor_blocks
,
1064 *len
, donor_blocks
- orig_start
,
1065 orig_inode
->i_ino
, donor_inode
->i_ino
);
1066 *len
= donor_blocks
- orig_start
;
1069 orig_blocks
= (orig_inode
->i_size
+ blocksize
- 1) >> blkbits
;
1070 if (orig_start
>= orig_blocks
) {
1071 ext4_debug("ext4 move extent: start offset [%llu] "
1072 "should be less than original file blocks "
1073 "[%u] [ino:orig %lu, donor %lu]\n",
1074 orig_start
, orig_blocks
,
1075 orig_inode
->i_ino
, donor_inode
->i_ino
);
1079 if (orig_start
+ *len
> orig_blocks
) {
1080 ext4_debug("ext4 move extent: Adjust length "
1081 "from %llu to %llu. Because it should be "
1082 "less than original file blocks "
1083 "[ino:orig %lu, donor %lu]\n",
1084 *len
, orig_blocks
- orig_start
,
1085 orig_inode
->i_ino
, donor_inode
->i_ino
);
1086 *len
= orig_blocks
- orig_start
;
1091 ext4_debug("ext4 move extent: len shoudld not be 0 "
1092 "[ino:orig %lu, donor %lu]\n", orig_inode
->i_ino
,
1093 donor_inode
->i_ino
);
1101 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1103 * @inode1: the inode structure
1104 * @inode2: the inode structure
1106 * Lock two inodes' i_mutex by i_ino order.
1107 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1110 mext_inode_double_lock(struct inode
*inode1
, struct inode
*inode2
)
1114 BUG_ON(inode1
== NULL
&& inode2
== NULL
);
1116 ret
= mext_check_null_inode(inode1
, inode2
, __func__
);
1120 if (inode1
== inode2
) {
1121 mutex_lock(&inode1
->i_mutex
);
1125 if (inode1
->i_ino
< inode2
->i_ino
) {
1126 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_PARENT
);
1127 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_CHILD
);
1129 mutex_lock_nested(&inode2
->i_mutex
, I_MUTEX_PARENT
);
1130 mutex_lock_nested(&inode1
->i_mutex
, I_MUTEX_CHILD
);
1138 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1140 * @inode1: the inode that is released first
1141 * @inode2: the inode that is released second
1143 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1147 mext_inode_double_unlock(struct inode
*inode1
, struct inode
*inode2
)
1151 BUG_ON(inode1
== NULL
&& inode2
== NULL
);
1153 ret
= mext_check_null_inode(inode1
, inode2
, __func__
);
1158 mutex_unlock(&inode1
->i_mutex
);
1160 if (inode2
&& inode2
!= inode1
)
1161 mutex_unlock(&inode2
->i_mutex
);
1168 * ext4_move_extents - Exchange the specified range of a file
1170 * @o_filp: file structure of the original file
1171 * @d_filp: file structure of the donor file
1172 * @orig_start: start offset in block for orig
1173 * @donor_start: start offset in block for donor
1174 * @len: the number of blocks to be moved
1175 * @moved_len: moved block length
1177 * This function returns 0 and moved block length is set in moved_len
1178 * if succeed, otherwise returns error value.
1180 * Note: ext4_move_extents() proceeds the following order.
1181 * 1:ext4_move_extents() calculates the last block number of moving extent
1182 * function by the start block number (orig_start) and the number of blocks
1183 * to be moved (len) specified as arguments.
1184 * If the {orig, donor}_start points a hole, the extent's start offset
1185 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1186 * after hole behind.
1187 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1188 * or the ext_cur exceeds the block_end which is last logical block number.
1189 * 3:To get the length of continues area, call mext_next_extent()
1190 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1191 * until find un-continuous extent, the start logical block number exceeds
1192 * the block_end or the extent points to the last extent.
1193 * 4:Exchange the original inode data with donor inode data
1194 * from orig_page_offset to seq_end_page.
1195 * The start indexes of data are specified as arguments.
1196 * That of the original inode is orig_page_offset,
1197 * and the donor inode is also orig_page_offset
1198 * (To easily handle blocksize != pagesize case, the offset for the
1199 * donor inode is block unit).
1200 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1201 * then returns to step 2.
1202 * 6:Release holecheck_path, orig_path and set the len to moved_len
1203 * which shows the number of moved blocks.
1204 * The moved_len is useful for the command to calculate the file offset
1205 * for starting next move extent ioctl.
1206 * 7:Return 0 on success, or a negative error value on failure.
1209 ext4_move_extents(struct file
*o_filp
, struct file
*d_filp
,
1210 __u64 orig_start
, __u64 donor_start
, __u64 len
,
1213 struct inode
*orig_inode
= o_filp
->f_dentry
->d_inode
;
1214 struct inode
*donor_inode
= d_filp
->f_dentry
->d_inode
;
1215 struct ext4_ext_path
*orig_path
= NULL
, *holecheck_path
= NULL
;
1216 struct ext4_extent
*ext_prev
, *ext_cur
, *ext_dummy
;
1217 ext4_lblk_t block_start
= orig_start
;
1218 ext4_lblk_t block_end
, seq_start
, add_blocks
, file_end
, seq_blocks
= 0;
1219 ext4_lblk_t rest_blocks
;
1220 pgoff_t orig_page_offset
= 0, seq_end_page
;
1221 int ret1
, ret2
, depth
, last_extent
= 0;
1222 int blocks_per_page
= PAGE_CACHE_SIZE
>> orig_inode
->i_blkbits
;
1223 int data_offset_in_page
;
1224 int block_len_in_page
;
1227 /* orig and donor should be different file */
1228 if (orig_inode
->i_ino
== donor_inode
->i_ino
) {
1229 ext4_debug("ext4 move extent: The argument files should not "
1230 "be same file [ino:orig %lu, donor %lu]\n",
1231 orig_inode
->i_ino
, donor_inode
->i_ino
);
1235 /* protect orig and donor against a truncate */
1236 ret1
= mext_inode_double_lock(orig_inode
, donor_inode
);
1240 mext_double_down_read(orig_inode
, donor_inode
);
1241 /* Check the filesystem environment whether move_extent can be done */
1242 ret1
= mext_check_arguments(orig_inode
, donor_inode
, orig_start
,
1243 donor_start
, &len
, *moved_len
);
1244 mext_double_up_read(orig_inode
, donor_inode
);
1248 file_end
= (i_size_read(orig_inode
) - 1) >> orig_inode
->i_blkbits
;
1249 block_end
= block_start
+ len
- 1;
1250 if (file_end
< block_end
)
1251 len
-= block_end
- file_end
;
1253 ret1
= get_ext_path(orig_inode
, block_start
, &orig_path
);
1257 /* Get path structure to check the hole */
1258 ret1
= get_ext_path(orig_inode
, block_start
, &holecheck_path
);
1262 depth
= ext_depth(orig_inode
);
1263 ext_cur
= holecheck_path
[depth
].p_ext
;
1266 * Get proper starting location of block replacement if block_start was
1269 if (le32_to_cpu(ext_cur
->ee_block
) +
1270 ext4_ext_get_actual_len(ext_cur
) - 1 < block_start
) {
1272 * The hole exists between extents or the tail of
1275 last_extent
= mext_next_extent(orig_inode
,
1276 holecheck_path
, &ext_cur
);
1277 if (last_extent
< 0) {
1281 last_extent
= mext_next_extent(orig_inode
, orig_path
,
1283 if (last_extent
< 0) {
1287 seq_start
= le32_to_cpu(ext_cur
->ee_block
);
1288 } else if (le32_to_cpu(ext_cur
->ee_block
) > block_start
)
1289 /* The hole exists at the beginning of original file. */
1290 seq_start
= le32_to_cpu(ext_cur
->ee_block
);
1292 seq_start
= block_start
;
1294 /* No blocks within the specified range. */
1295 if (le32_to_cpu(ext_cur
->ee_block
) > block_end
) {
1296 ext4_debug("ext4 move extent: The specified range of file "
1297 "may be the hole\n");
1302 /* Adjust start blocks */
1303 add_blocks
= min(le32_to_cpu(ext_cur
->ee_block
) +
1304 ext4_ext_get_actual_len(ext_cur
), block_end
+ 1) -
1305 max(le32_to_cpu(ext_cur
->ee_block
), block_start
);
1307 while (!last_extent
&& le32_to_cpu(ext_cur
->ee_block
) <= block_end
) {
1308 seq_blocks
+= add_blocks
;
1310 /* Adjust tail blocks */
1311 if (seq_start
+ seq_blocks
- 1 > block_end
)
1312 seq_blocks
= block_end
- seq_start
+ 1;
1315 last_extent
= mext_next_extent(orig_inode
, holecheck_path
,
1317 if (last_extent
< 0) {
1321 add_blocks
= ext4_ext_get_actual_len(ext_cur
);
1324 * Extend the length of contiguous block (seq_blocks)
1325 * if extents are contiguous.
1327 if (ext4_can_extents_be_merged(orig_inode
,
1328 ext_prev
, ext_cur
) &&
1329 block_end
>= le32_to_cpu(ext_cur
->ee_block
) &&
1333 /* Is original extent is uninitialized */
1334 uninit
= ext4_ext_is_uninitialized(ext_prev
);
1336 data_offset_in_page
= seq_start
% blocks_per_page
;
1339 * Calculate data blocks count that should be swapped
1340 * at the first page.
1342 if (data_offset_in_page
+ seq_blocks
> blocks_per_page
) {
1343 /* Swapped blocks are across pages */
1345 blocks_per_page
- data_offset_in_page
;
1347 /* Swapped blocks are in a page */
1348 block_len_in_page
= seq_blocks
;
1351 orig_page_offset
= seq_start
>>
1352 (PAGE_CACHE_SHIFT
- orig_inode
->i_blkbits
);
1353 seq_end_page
= (seq_start
+ seq_blocks
- 1) >>
1354 (PAGE_CACHE_SHIFT
- orig_inode
->i_blkbits
);
1355 seq_start
= le32_to_cpu(ext_cur
->ee_block
);
1356 rest_blocks
= seq_blocks
;
1358 /* Discard preallocations of two inodes */
1359 down_write(&EXT4_I(orig_inode
)->i_data_sem
);
1360 ext4_discard_preallocations(orig_inode
);
1361 up_write(&EXT4_I(orig_inode
)->i_data_sem
);
1363 down_write(&EXT4_I(donor_inode
)->i_data_sem
);
1364 ext4_discard_preallocations(donor_inode
);
1365 up_write(&EXT4_I(donor_inode
)->i_data_sem
);
1367 while (orig_page_offset
<= seq_end_page
) {
1369 /* Swap original branches with new branches */
1370 ret1
= move_extent_per_page(o_filp
, donor_inode
,
1372 data_offset_in_page
,
1373 block_len_in_page
, uninit
);
1377 /* Count how many blocks we have exchanged */
1378 *moved_len
+= block_len_in_page
;
1379 if (*moved_len
> len
) {
1380 ext4_error(orig_inode
->i_sb
, __func__
,
1381 "We replaced blocks too much! "
1382 "sum of replaced: %llu requested: %llu",
1388 data_offset_in_page
= 0;
1389 rest_blocks
-= block_len_in_page
;
1390 if (rest_blocks
> blocks_per_page
)
1391 block_len_in_page
= blocks_per_page
;
1393 block_len_in_page
= rest_blocks
;
1396 /* Decrease buffer counter */
1398 ext4_ext_drop_refs(holecheck_path
);
1399 ret1
= get_ext_path(orig_inode
, seq_start
, &holecheck_path
);
1402 depth
= holecheck_path
->p_depth
;
1404 /* Decrease buffer counter */
1406 ext4_ext_drop_refs(orig_path
);
1407 ret1
= get_ext_path(orig_inode
, seq_start
, &orig_path
);
1411 ext_cur
= holecheck_path
[depth
].p_ext
;
1412 add_blocks
= ext4_ext_get_actual_len(ext_cur
);
1418 ext4_ext_drop_refs(orig_path
);
1421 if (holecheck_path
) {
1422 ext4_ext_drop_refs(holecheck_path
);
1423 kfree(holecheck_path
);
1426 ret2
= mext_inode_double_unlock(orig_inode
, donor_inode
);