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 <linux/slab.h>
19 #include "ext4_jbd2.h"
20 #include "ext4_extents.h"
24 * get_ext_path - Find an extent path for designated logical block number.
26 * @inode: an inode which is searched
27 * @lblock: logical block number to find an extent path
28 * @path: pointer to an extent path pointer (for output)
30 * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
34 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
35 struct ext4_ext_path **path)
39 *path = ext4_ext_find_extent(inode, lblock, *path);
43 } else if ((*path)[ext_depth(inode)].p_ext == NULL)
50 * copy_extent_status - Copy the extent's initialization status
52 * @src: an extent for getting initialize status
53 * @dest: an extent to be set the status
56 copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
58 if (ext4_ext_is_uninitialized(src))
59 ext4_ext_mark_uninitialized(dest);
61 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
65 * mext_next_extent - Search for the next extent and set it to "extent"
67 * @inode: inode which is searched
68 * @path: this will obtain data for the next extent
69 * @extent: pointer to the next extent we have just gotten
71 * Search the next extent in the array of ext4_ext_path structure (@path)
72 * and set it to ext4_extent structure (@extent). In addition, the member of
73 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
74 * ext4_ext_path structure refers to the last extent, or a negative error
78 mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
79 struct ext4_extent **extent)
81 struct ext4_extent_header *eh;
82 int ppos, leaf_ppos = path->p_depth;
85 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
87 *extent = ++path[ppos].p_ext;
88 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
93 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
99 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
100 if (path[ppos+1].p_bh)
101 brelse(path[ppos+1].p_bh);
103 sb_bread(inode->i_sb, path[ppos].p_block);
104 if (!path[ppos+1].p_bh)
107 ext_block_hdr(path[ppos+1].p_bh);
109 /* Halfway index block */
110 while (++cur_ppos < leaf_ppos) {
111 path[cur_ppos].p_idx =
112 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
113 path[cur_ppos].p_block =
114 idx_pblock(path[cur_ppos].p_idx);
115 if (path[cur_ppos+1].p_bh)
116 brelse(path[cur_ppos+1].p_bh);
117 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
118 path[cur_ppos].p_block);
119 if (!path[cur_ppos+1].p_bh)
121 path[cur_ppos+1].p_hdr =
122 ext_block_hdr(path[cur_ppos+1].p_bh);
125 path[leaf_ppos].p_ext = *extent = NULL;
127 eh = path[leaf_ppos].p_hdr;
128 if (le16_to_cpu(eh->eh_entries) == 0)
129 /* empty leaf is found */
133 path[leaf_ppos].p_ext = *extent =
134 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
135 path[leaf_ppos].p_block =
136 ext_pblock(path[leaf_ppos].p_ext);
140 /* We found the last extent */
145 * mext_check_null_inode - NULL check for two inodes
147 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
150 mext_check_null_inode(struct inode *inode1, struct inode *inode2,
151 const char *function, unsigned int line)
155 if (inode1 == NULL) {
156 __ext4_error(inode2->i_sb, function, line,
157 "Both inodes should not be NULL: "
158 "inode1 NULL inode2 %lu", inode2->i_ino);
160 } else if (inode2 == NULL) {
161 __ext4_error(inode1->i_sb, function, line,
162 "Both inodes should not be NULL: "
163 "inode1 %lu inode2 NULL", inode1->i_ino);
170 * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
172 * @orig_inode: original inode structure
173 * @donor_inode: donor inode structure
174 * Acquire write lock of i_data_sem of the two inodes (orig and donor) by
178 double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
180 struct inode *first = orig_inode, *second = donor_inode;
183 * Use the inode number to provide the stable locking order instead
184 * of its address, because the C language doesn't guarantee you can
185 * compare pointers that don't come from the same array.
187 if (donor_inode->i_ino < orig_inode->i_ino) {
192 down_write(&EXT4_I(first)->i_data_sem);
193 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
197 * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
199 * @orig_inode: original inode structure to be released its lock first
200 * @donor_inode: donor inode structure to be released its lock second
201 * Release write lock of i_data_sem of two inodes (orig and donor).
204 double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
206 up_write(&EXT4_I(orig_inode)->i_data_sem);
207 up_write(&EXT4_I(donor_inode)->i_data_sem);
211 * mext_insert_across_blocks - Insert extents across leaf block
213 * @handle: journal handle
214 * @orig_inode: original inode
215 * @o_start: first original extent to be changed
216 * @o_end: last original extent to be changed
217 * @start_ext: first new extent to be inserted
218 * @new_ext: middle of new extent to be inserted
219 * @end_ext: last new extent to be inserted
221 * Allocate a new leaf block and insert extents into it. Return 0 on success,
222 * or a negative error value on failure.
225 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
226 struct ext4_extent *o_start, struct ext4_extent *o_end,
227 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
228 struct ext4_extent *end_ext)
230 struct ext4_ext_path *orig_path = NULL;
231 ext4_lblk_t eblock = 0;
236 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
237 if (o_start == o_end) {
239 /* start_ext new_ext end_ext
240 * donor |---------|-----------|--------|
241 * orig |------------------------------|
246 /* start_ext new_ext end_ext
247 * donor |---------|----------|---------|
248 * orig |---------------|--------------|
250 o_end->ee_block = end_ext->ee_block;
251 o_end->ee_len = end_ext->ee_len;
252 ext4_ext_store_pblock(o_end, ext_pblock(end_ext));
255 o_start->ee_len = start_ext->ee_len;
256 eblock = le32_to_cpu(start_ext->ee_block);
259 } else if (start_ext->ee_len && new_ext->ee_len &&
260 !end_ext->ee_len && o_start == o_end) {
263 * donor |--------------|---------------|
264 * orig |------------------------------|
266 o_start->ee_len = start_ext->ee_len;
267 eblock = le32_to_cpu(start_ext->ee_block);
270 } else if (!start_ext->ee_len && new_ext->ee_len &&
271 end_ext->ee_len && o_start == o_end) {
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 * Set 0 to the extent block if new_ext was
285 if (new_ext->ee_block)
286 eblock = le32_to_cpu(new_ext->ee_block);
290 ext4_debug("ext4 move extent: Unexpected insert case\n");
295 err = get_ext_path(orig_inode, eblock, &orig_path);
299 if (ext4_ext_insert_extent(handle, orig_inode,
300 orig_path, new_ext, 0))
305 err = get_ext_path(orig_inode,
306 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
310 if (ext4_ext_insert_extent(handle, orig_inode,
311 orig_path, end_ext, 0))
316 ext4_ext_drop_refs(orig_path);
325 * mext_insert_inside_block - Insert new extent to the extent block
327 * @o_start: first original extent to be moved
328 * @o_end: last original extent to be moved
329 * @start_ext: first new extent to be inserted
330 * @new_ext: middle of new extent to be inserted
331 * @end_ext: last new extent to be inserted
332 * @eh: extent header of target leaf block
333 * @range_to_move: used to decide how to insert extent
335 * Insert extents into the leaf block. The extent (@o_start) is overwritten
336 * by inserted extents.
339 mext_insert_inside_block(struct ext4_extent *o_start,
340 struct ext4_extent *o_end,
341 struct ext4_extent *start_ext,
342 struct ext4_extent *new_ext,
343 struct ext4_extent *end_ext,
344 struct ext4_extent_header *eh,
350 /* Move the existing extents */
351 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
352 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
353 (unsigned long)(o_end + 1);
354 memmove(o_end + 1 + range_to_move, o_end + 1, len);
357 /* Insert start entry */
358 if (start_ext->ee_len)
359 o_start[i++].ee_len = start_ext->ee_len;
361 /* Insert new entry */
362 if (new_ext->ee_len) {
363 o_start[i] = *new_ext;
364 ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext));
367 /* Insert end entry */
369 o_start[i] = *end_ext;
371 /* Increment the total entries counter on the extent block */
372 le16_add_cpu(&eh->eh_entries, range_to_move);
376 * mext_insert_extents - Insert new extent
378 * @handle: journal handle
379 * @orig_inode: original inode
380 * @orig_path: path indicates first extent to be changed
381 * @o_start: first original extent to be changed
382 * @o_end: last original extent to be changed
383 * @start_ext: first new extent to be inserted
384 * @new_ext: middle of new extent to be inserted
385 * @end_ext: last new extent to be inserted
387 * Call the function to insert extents. If we cannot add more extents into
388 * the leaf block, we call mext_insert_across_blocks() to create a
389 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
390 * on success, or a negative error value on failure.
393 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
394 struct ext4_ext_path *orig_path,
395 struct ext4_extent *o_start,
396 struct ext4_extent *o_end,
397 struct ext4_extent *start_ext,
398 struct ext4_extent *new_ext,
399 struct ext4_extent *end_ext)
401 struct ext4_extent_header *eh;
402 unsigned long need_slots, slots_range;
403 int range_to_move, depth, ret;
406 * The extents need to be inserted
407 * start_extent + new_extent + end_extent.
409 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
410 (new_ext->ee_len ? 1 : 0);
412 /* The number of slots between start and end */
413 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
414 / sizeof(struct ext4_extent);
416 /* Range to move the end of extent */
417 range_to_move = need_slots - slots_range;
418 depth = orig_path->p_depth;
420 eh = orig_path->p_hdr;
423 /* Register to journal */
424 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
430 if (range_to_move > 0 &&
431 (range_to_move > le16_to_cpu(eh->eh_max)
432 - le16_to_cpu(eh->eh_entries))) {
434 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
435 o_end, start_ext, new_ext, end_ext);
439 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
440 end_ext, eh, range_to_move);
443 ret = ext4_handle_dirty_metadata(handle, orig_inode,
448 ret = ext4_mark_inode_dirty(handle, orig_inode);
457 * mext_leaf_block - Move one leaf extent block into the inode.
459 * @handle: journal handle
460 * @orig_inode: original inode
461 * @orig_path: path indicates first extent to be changed
462 * @dext: donor extent
463 * @from: start offset on the target file
465 * In order to insert extents into the leaf block, we must divide the extent
466 * in the leaf block into three extents. The one is located to be inserted
467 * extents, and the others are located around it.
469 * Therefore, this function creates structures to save extents of the leaf
470 * block, and inserts extents by calling mext_insert_extents() with
471 * created extents. Return 0 on success, or a negative error value on failure.
474 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
475 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
478 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
479 struct ext4_extent new_ext, start_ext, end_ext;
480 ext4_lblk_t new_ext_end;
481 int oext_alen, new_ext_alen, end_ext_alen;
482 int depth = ext_depth(orig_inode);
485 start_ext.ee_block = end_ext.ee_block = 0;
486 o_start = o_end = oext = orig_path[depth].p_ext;
487 oext_alen = ext4_ext_get_actual_len(oext);
488 start_ext.ee_len = end_ext.ee_len = 0;
490 new_ext.ee_block = cpu_to_le32(*from);
491 ext4_ext_store_pblock(&new_ext, ext_pblock(dext));
492 new_ext.ee_len = dext->ee_len;
493 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
494 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
497 * Case: original extent is first
502 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
503 le32_to_cpu(new_ext.ee_block) <
504 le32_to_cpu(oext->ee_block) + oext_alen) {
505 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
506 le32_to_cpu(oext->ee_block));
507 start_ext.ee_block = oext->ee_block;
508 copy_extent_status(oext, &start_ext);
509 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
512 * We can merge new_ext into previous extent,
513 * if these are contiguous and same extent type.
515 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
518 start_ext.ee_len = cpu_to_le16(
519 ext4_ext_get_actual_len(prev_ext) +
521 start_ext.ee_block = oext->ee_block;
522 copy_extent_status(prev_ext, &start_ext);
528 * Case: new_ext_end must be less than oext
532 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
533 EXT4_ERROR_INODE(orig_inode,
534 "new_ext_end(%u) should be less than or equal to "
535 "oext->ee_block(%u) + oext_alen(%d) - 1",
536 new_ext_end, le32_to_cpu(oext->ee_block),
543 * Case: new_ext is smaller than original extent
544 * oext |---------------|
545 * new_ext |-----------|
548 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
549 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
551 cpu_to_le16(le32_to_cpu(oext->ee_block) +
552 oext_alen - 1 - new_ext_end);
553 copy_extent_status(oext, &end_ext);
554 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
555 ext4_ext_store_pblock(&end_ext,
556 (ext_pblock(o_end) + oext_alen - end_ext_alen));
558 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
559 oext_alen - end_ext_alen);
562 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
563 o_end, &start_ext, &new_ext, &end_ext);
569 * mext_calc_swap_extents - Calculate extents for extent swapping.
571 * @tmp_dext: the extent that will belong to the original inode
572 * @tmp_oext: the extent that will belong to the donor inode
573 * @orig_off: block offset of original inode
574 * @donor_off: block offset of donor inode
575 * @max_count: the maximum length of extents
577 * Return 0 on success, or a negative error value on failure.
580 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
581 struct ext4_extent *tmp_oext,
582 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
583 ext4_lblk_t max_count)
585 ext4_lblk_t diff, orig_diff;
586 struct ext4_extent dext_old, oext_old;
588 BUG_ON(orig_off != donor_off);
590 /* original and donor extents have to cover the same block offset */
591 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
592 le32_to_cpu(tmp_oext->ee_block) +
593 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
596 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
597 le32_to_cpu(tmp_dext->ee_block) +
598 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
601 dext_old = *tmp_dext;
602 oext_old = *tmp_oext;
604 /* When tmp_dext is too large, pick up the target range. */
605 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
607 ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff);
609 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
610 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
612 if (max_count < ext4_ext_get_actual_len(tmp_dext))
613 tmp_dext->ee_len = cpu_to_le16(max_count);
615 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
616 ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff);
618 /* Adjust extent length if donor extent is larger than orig */
619 if (ext4_ext_get_actual_len(tmp_dext) >
620 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
621 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
624 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
626 copy_extent_status(&oext_old, tmp_dext);
627 copy_extent_status(&dext_old, tmp_oext);
633 * mext_replace_branches - Replace original extents with new extents
635 * @handle: journal handle
636 * @orig_inode: original inode
637 * @donor_inode: donor inode
638 * @from: block offset of orig_inode
639 * @count: block count to be replaced
640 * @err: pointer to save return value
642 * Replace original inode extents and donor inode extents page by page.
643 * We implement this replacement in the following three steps:
644 * 1. Save the block information of original and donor inodes into
646 * 2. Change the block information of original inode to point at the
647 * donor inode blocks.
648 * 3. Change the block information of donor inode to point at the saved
649 * original inode blocks in the dummy extents.
651 * Return replaced block count.
654 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
655 struct inode *donor_inode, ext4_lblk_t from,
656 ext4_lblk_t count, int *err)
658 struct ext4_ext_path *orig_path = NULL;
659 struct ext4_ext_path *donor_path = NULL;
660 struct ext4_extent *oext, *dext;
661 struct ext4_extent tmp_dext, tmp_oext;
662 ext4_lblk_t orig_off = from, donor_off = from;
664 int replaced_count = 0;
667 /* Protect extent trees against block allocations via delalloc */
668 double_down_write_data_sem(orig_inode, donor_inode);
670 /* Get the original extent for the block "orig_off" */
671 *err = get_ext_path(orig_inode, orig_off, &orig_path);
675 /* Get the donor extent for the head */
676 *err = get_ext_path(donor_inode, donor_off, &donor_path);
679 depth = ext_depth(orig_inode);
680 oext = orig_path[depth].p_ext;
683 depth = ext_depth(donor_inode);
684 dext = donor_path[depth].p_ext;
687 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
692 /* Loop for the donor extents */
694 /* The extent for donor must be found. */
696 EXT4_ERROR_INODE(donor_inode,
697 "The extent for donor must be found");
700 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
701 EXT4_ERROR_INODE(donor_inode,
702 "Donor offset(%u) and the first block of donor "
703 "extent(%u) should be equal",
705 le32_to_cpu(tmp_dext.ee_block));
710 /* Set donor extent to orig extent */
711 *err = mext_leaf_block(handle, orig_inode,
712 orig_path, &tmp_dext, &orig_off);
716 /* Set orig extent to donor extent */
717 *err = mext_leaf_block(handle, donor_inode,
718 donor_path, &tmp_oext, &donor_off);
722 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
723 replaced_count += dext_alen;
724 donor_off += dext_alen;
725 orig_off += dext_alen;
727 /* Already moved the expected blocks */
728 if (replaced_count >= count)
732 ext4_ext_drop_refs(orig_path);
733 *err = get_ext_path(orig_inode, orig_off, &orig_path);
736 depth = ext_depth(orig_inode);
737 oext = orig_path[depth].p_ext;
741 ext4_ext_drop_refs(donor_path);
742 *err = get_ext_path(donor_inode, donor_off, &donor_path);
745 depth = ext_depth(donor_inode);
746 dext = donor_path[depth].p_ext;
749 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
750 donor_off, count - replaced_count);
757 ext4_ext_drop_refs(orig_path);
761 ext4_ext_drop_refs(donor_path);
765 ext4_ext_invalidate_cache(orig_inode);
766 ext4_ext_invalidate_cache(donor_inode);
768 double_up_write_data_sem(orig_inode, donor_inode);
770 return replaced_count;
774 * move_extent_per_page - Move extent data per page
776 * @o_filp: file structure of original file
777 * @donor_inode: donor inode
778 * @orig_page_offset: page index on original file
779 * @data_offset_in_page: block index where data swapping starts
780 * @block_len_in_page: the number of blocks to be swapped
781 * @uninit: orig extent is uninitialized or not
782 * @err: pointer to save return value
784 * Save the data in original inode blocks and replace original inode extents
785 * with donor inode extents by calling mext_replace_branches().
786 * Finally, write out the saved data in new original inode blocks. Return
787 * replaced block count.
790 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
791 pgoff_t orig_page_offset, int data_offset_in_page,
792 int block_len_in_page, int uninit, int *err)
794 struct inode *orig_inode = o_filp->f_dentry->d_inode;
795 struct address_space *mapping = orig_inode->i_mapping;
796 struct buffer_head *bh;
797 struct page *page = NULL;
798 const struct address_space_operations *a_ops = mapping->a_ops;
800 ext4_lblk_t orig_blk_offset;
801 long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
802 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
803 unsigned int w_flags = 0;
804 unsigned int tmp_data_size, data_size, replaced_size;
808 int replaced_count = 0;
809 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
812 * It needs twice the amount of ordinary journal buffers because
813 * inode and donor_inode may change each different metadata blocks.
815 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
816 handle = ext4_journal_start(orig_inode, jblocks);
817 if (IS_ERR(handle)) {
818 *err = PTR_ERR(handle);
822 if (segment_eq(get_fs(), KERNEL_DS))
823 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
825 orig_blk_offset = orig_page_offset * blocks_per_page +
829 * If orig extent is uninitialized one,
830 * it's not necessary force the page into memory
831 * and then force it to be written out again.
832 * Just swap data blocks between orig and donor.
835 replaced_count = mext_replace_branches(handle, orig_inode,
836 donor_inode, orig_blk_offset,
837 block_len_in_page, err);
841 offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
843 /* Calculate data_size */
844 if ((orig_blk_offset + block_len_in_page - 1) ==
845 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
846 /* Replace the last block */
847 tmp_data_size = orig_inode->i_size & (blocksize - 1);
849 * If data_size equal zero, it shows data_size is multiples of
850 * blocksize. So we set appropriate value.
852 if (tmp_data_size == 0)
853 tmp_data_size = blocksize;
855 data_size = tmp_data_size +
856 ((block_len_in_page - 1) << orig_inode->i_blkbits);
858 data_size = block_len_in_page << orig_inode->i_blkbits;
860 replaced_size = data_size;
862 *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags,
864 if (unlikely(*err < 0))
867 if (!PageUptodate(page)) {
868 mapping->a_ops->readpage(o_filp, page);
873 * try_to_release_page() doesn't call releasepage in writeback mode.
874 * We should care about the order of writing to the same file
875 * by multiple move extent processes.
876 * It needs to call wait_on_page_writeback() to wait for the
877 * writeback of the page.
879 if (PageWriteback(page))
880 wait_on_page_writeback(page);
882 /* Release old bh and drop refs */
883 try_to_release_page(page, 0);
885 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
886 orig_blk_offset, block_len_in_page,
889 if (replaced_count) {
890 block_len_in_page = replaced_count;
892 block_len_in_page << orig_inode->i_blkbits;
897 if (!page_has_buffers(page))
898 create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0);
900 bh = page_buffers(page);
901 for (i = 0; i < data_offset_in_page; i++)
902 bh = bh->b_this_page;
904 for (i = 0; i < block_len_in_page; i++) {
905 *err = ext4_get_block(orig_inode,
906 (sector_t)(orig_blk_offset + i), bh, 0);
910 if (bh->b_this_page != NULL)
911 bh = bh->b_this_page;
914 *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size,
919 if (unlikely(page)) {
920 if (PageLocked(page))
922 page_cache_release(page);
923 ext4_journal_stop(handle);
926 ext4_journal_stop(handle);
931 return replaced_count;
935 * mext_check_arguments - Check whether move extent can be done
937 * @orig_inode: original inode
938 * @donor_inode: donor inode
939 * @orig_start: logical start offset in block for orig
940 * @donor_start: logical start offset in block for donor
941 * @len: the number of blocks to be moved
943 * Check the arguments of ext4_move_extents() whether the files can be
944 * exchanged with each other.
945 * Return 0 on success, or a negative error value on failure.
948 mext_check_arguments(struct inode *orig_inode,
949 struct inode *donor_inode, __u64 orig_start,
950 __u64 donor_start, __u64 *len)
952 ext4_lblk_t orig_blocks, donor_blocks;
953 unsigned int blkbits = orig_inode->i_blkbits;
954 unsigned int blocksize = 1 << blkbits;
956 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
957 ext4_debug("ext4 move extent: suid or sgid is set"
958 " to donor file [ino:orig %lu, donor %lu]\n",
959 orig_inode->i_ino, donor_inode->i_ino);
963 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
966 /* Ext4 move extent does not support swapfile */
967 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
968 ext4_debug("ext4 move extent: The argument files should "
969 "not be swapfile [ino:orig %lu, donor %lu]\n",
970 orig_inode->i_ino, donor_inode->i_ino);
974 /* Files should be in the same ext4 FS */
975 if (orig_inode->i_sb != donor_inode->i_sb) {
976 ext4_debug("ext4 move extent: The argument files "
977 "should be in same FS [ino:orig %lu, donor %lu]\n",
978 orig_inode->i_ino, donor_inode->i_ino);
982 /* Ext4 move extent supports only extent based file */
983 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
984 ext4_debug("ext4 move extent: orig file is not extents "
985 "based file [ino:orig %lu]\n", orig_inode->i_ino);
987 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
988 ext4_debug("ext4 move extent: donor file is not extents "
989 "based file [ino:donor %lu]\n", donor_inode->i_ino);
993 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
994 ext4_debug("ext4 move extent: File size is 0 byte\n");
998 /* Start offset should be same */
999 if (orig_start != donor_start) {
1000 ext4_debug("ext4 move extent: orig and donor's start "
1001 "offset are not same [ino:orig %lu, donor %lu]\n",
1002 orig_inode->i_ino, donor_inode->i_ino);
1006 if ((orig_start > EXT_MAX_BLOCK) ||
1007 (donor_start > EXT_MAX_BLOCK) ||
1008 (*len > EXT_MAX_BLOCK) ||
1009 (orig_start + *len > EXT_MAX_BLOCK)) {
1010 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1011 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK,
1012 orig_inode->i_ino, donor_inode->i_ino);
1016 if (orig_inode->i_size > donor_inode->i_size) {
1017 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1018 /* TODO: eliminate this artificial restriction */
1019 if (orig_start >= donor_blocks) {
1020 ext4_debug("ext4 move extent: orig start offset "
1021 "[%llu] should be less than donor file blocks "
1022 "[%u] [ino:orig %lu, donor %lu]\n",
1023 orig_start, donor_blocks,
1024 orig_inode->i_ino, donor_inode->i_ino);
1028 /* TODO: eliminate this artificial restriction */
1029 if (orig_start + *len > donor_blocks) {
1030 ext4_debug("ext4 move extent: End offset [%llu] should "
1031 "be less than donor file blocks [%u]."
1032 "So adjust length from %llu to %llu "
1033 "[ino:orig %lu, donor %lu]\n",
1034 orig_start + *len, donor_blocks,
1035 *len, donor_blocks - orig_start,
1036 orig_inode->i_ino, donor_inode->i_ino);
1037 *len = donor_blocks - orig_start;
1040 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1041 if (orig_start >= orig_blocks) {
1042 ext4_debug("ext4 move extent: start offset [%llu] "
1043 "should be less than original file blocks "
1044 "[%u] [ino:orig %lu, donor %lu]\n",
1045 orig_start, orig_blocks,
1046 orig_inode->i_ino, donor_inode->i_ino);
1050 if (orig_start + *len > orig_blocks) {
1051 ext4_debug("ext4 move extent: Adjust length "
1052 "from %llu to %llu. Because it should be "
1053 "less than original file blocks "
1054 "[ino:orig %lu, donor %lu]\n",
1055 *len, orig_blocks - orig_start,
1056 orig_inode->i_ino, donor_inode->i_ino);
1057 *len = orig_blocks - orig_start;
1062 ext4_debug("ext4 move extent: len should not be 0 "
1063 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1064 donor_inode->i_ino);
1072 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1074 * @inode1: the inode structure
1075 * @inode2: the inode structure
1077 * Lock two inodes' i_mutex by i_ino order.
1078 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1081 mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1085 BUG_ON(inode1 == NULL && inode2 == NULL);
1087 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1091 if (inode1 == inode2) {
1092 mutex_lock(&inode1->i_mutex);
1096 if (inode1->i_ino < inode2->i_ino) {
1097 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1098 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1100 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1101 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1109 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1111 * @inode1: the inode that is released first
1112 * @inode2: the inode that is released second
1114 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1118 mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1122 BUG_ON(inode1 == NULL && inode2 == NULL);
1124 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1129 mutex_unlock(&inode1->i_mutex);
1131 if (inode2 && inode2 != inode1)
1132 mutex_unlock(&inode2->i_mutex);
1139 * ext4_move_extents - Exchange the specified range of a file
1141 * @o_filp: file structure of the original file
1142 * @d_filp: file structure of the donor file
1143 * @orig_start: start offset in block for orig
1144 * @donor_start: start offset in block for donor
1145 * @len: the number of blocks to be moved
1146 * @moved_len: moved block length
1148 * This function returns 0 and moved block length is set in moved_len
1149 * if succeed, otherwise returns error value.
1151 * Note: ext4_move_extents() proceeds the following order.
1152 * 1:ext4_move_extents() calculates the last block number of moving extent
1153 * function by the start block number (orig_start) and the number of blocks
1154 * to be moved (len) specified as arguments.
1155 * If the {orig, donor}_start points a hole, the extent's start offset
1156 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1157 * after hole behind.
1158 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1159 * or the ext_cur exceeds the block_end which is last logical block number.
1160 * 3:To get the length of continues area, call mext_next_extent()
1161 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1162 * until find un-continuous extent, the start logical block number exceeds
1163 * the block_end or the extent points to the last extent.
1164 * 4:Exchange the original inode data with donor inode data
1165 * from orig_page_offset to seq_end_page.
1166 * The start indexes of data are specified as arguments.
1167 * That of the original inode is orig_page_offset,
1168 * and the donor inode is also orig_page_offset
1169 * (To easily handle blocksize != pagesize case, the offset for the
1170 * donor inode is block unit).
1171 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1172 * then returns to step 2.
1173 * 6:Release holecheck_path, orig_path and set the len to moved_len
1174 * which shows the number of moved blocks.
1175 * The moved_len is useful for the command to calculate the file offset
1176 * for starting next move extent ioctl.
1177 * 7:Return 0 on success, or a negative error value on failure.
1180 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1181 __u64 orig_start, __u64 donor_start, __u64 len,
1184 struct inode *orig_inode = o_filp->f_dentry->d_inode;
1185 struct inode *donor_inode = d_filp->f_dentry->d_inode;
1186 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1187 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1188 ext4_lblk_t block_start = orig_start;
1189 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1190 ext4_lblk_t rest_blocks;
1191 pgoff_t orig_page_offset = 0, seq_end_page;
1192 int ret1, ret2, depth, last_extent = 0;
1193 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1194 int data_offset_in_page;
1195 int block_len_in_page;
1198 /* orig and donor should be different file */
1199 if (orig_inode->i_ino == donor_inode->i_ino) {
1200 ext4_debug("ext4 move extent: The argument files should not "
1201 "be same file [ino:orig %lu, donor %lu]\n",
1202 orig_inode->i_ino, donor_inode->i_ino);
1206 /* Regular file check */
1207 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1208 ext4_debug("ext4 move extent: The argument files should be "
1209 "regular file [ino:orig %lu, donor %lu]\n",
1210 orig_inode->i_ino, donor_inode->i_ino);
1214 /* Protect orig and donor inodes against a truncate */
1215 ret1 = mext_inode_double_lock(orig_inode, donor_inode);
1219 /* Protect extent tree against block allocations via delalloc */
1220 double_down_write_data_sem(orig_inode, donor_inode);
1221 /* Check the filesystem environment whether move_extent can be done */
1222 ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start,
1227 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1228 block_end = block_start + len - 1;
1229 if (file_end < block_end)
1230 len -= block_end - file_end;
1232 ret1 = get_ext_path(orig_inode, block_start, &orig_path);
1236 /* Get path structure to check the hole */
1237 ret1 = get_ext_path(orig_inode, block_start, &holecheck_path);
1241 depth = ext_depth(orig_inode);
1242 ext_cur = holecheck_path[depth].p_ext;
1245 * Get proper starting location of block replacement if block_start was
1248 if (le32_to_cpu(ext_cur->ee_block) +
1249 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1251 * The hole exists between extents or the tail of
1254 last_extent = mext_next_extent(orig_inode,
1255 holecheck_path, &ext_cur);
1256 if (last_extent < 0) {
1260 last_extent = mext_next_extent(orig_inode, orig_path,
1262 if (last_extent < 0) {
1266 seq_start = le32_to_cpu(ext_cur->ee_block);
1267 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1268 /* The hole exists at the beginning of original file. */
1269 seq_start = le32_to_cpu(ext_cur->ee_block);
1271 seq_start = block_start;
1273 /* No blocks within the specified range. */
1274 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1275 ext4_debug("ext4 move extent: The specified range of file "
1276 "may be the hole\n");
1281 /* Adjust start blocks */
1282 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1283 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1284 max(le32_to_cpu(ext_cur->ee_block), block_start);
1286 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1287 seq_blocks += add_blocks;
1289 /* Adjust tail blocks */
1290 if (seq_start + seq_blocks - 1 > block_end)
1291 seq_blocks = block_end - seq_start + 1;
1294 last_extent = mext_next_extent(orig_inode, holecheck_path,
1296 if (last_extent < 0) {
1300 add_blocks = ext4_ext_get_actual_len(ext_cur);
1303 * Extend the length of contiguous block (seq_blocks)
1304 * if extents are contiguous.
1306 if (ext4_can_extents_be_merged(orig_inode,
1307 ext_prev, ext_cur) &&
1308 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1312 /* Is original extent is uninitialized */
1313 uninit = ext4_ext_is_uninitialized(ext_prev);
1315 data_offset_in_page = seq_start % blocks_per_page;
1318 * Calculate data blocks count that should be swapped
1319 * at the first page.
1321 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1322 /* Swapped blocks are across pages */
1324 blocks_per_page - data_offset_in_page;
1326 /* Swapped blocks are in a page */
1327 block_len_in_page = seq_blocks;
1330 orig_page_offset = seq_start >>
1331 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1332 seq_end_page = (seq_start + seq_blocks - 1) >>
1333 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1334 seq_start = le32_to_cpu(ext_cur->ee_block);
1335 rest_blocks = seq_blocks;
1338 * Up semaphore to avoid following problems:
1339 * a. transaction deadlock among ext4_journal_start,
1340 * ->write_begin via pagefault, and jbd2_journal_commit
1341 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1342 * in move_extent_per_page
1344 double_up_write_data_sem(orig_inode, donor_inode);
1346 while (orig_page_offset <= seq_end_page) {
1348 /* Swap original branches with new branches */
1349 block_len_in_page = move_extent_per_page(
1350 o_filp, donor_inode,
1352 data_offset_in_page,
1353 block_len_in_page, uninit,
1356 /* Count how many blocks we have exchanged */
1357 *moved_len += block_len_in_page;
1360 if (*moved_len > len) {
1361 EXT4_ERROR_INODE(orig_inode,
1362 "We replaced blocks too much! "
1363 "sum of replaced: %llu requested: %llu",
1370 data_offset_in_page = 0;
1371 rest_blocks -= block_len_in_page;
1372 if (rest_blocks > blocks_per_page)
1373 block_len_in_page = blocks_per_page;
1375 block_len_in_page = rest_blocks;
1378 double_down_write_data_sem(orig_inode, donor_inode);
1382 /* Decrease buffer counter */
1384 ext4_ext_drop_refs(holecheck_path);
1385 ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path);
1388 depth = holecheck_path->p_depth;
1390 /* Decrease buffer counter */
1392 ext4_ext_drop_refs(orig_path);
1393 ret1 = get_ext_path(orig_inode, seq_start, &orig_path);
1397 ext_cur = holecheck_path[depth].p_ext;
1398 add_blocks = ext4_ext_get_actual_len(ext_cur);
1404 ext4_discard_preallocations(orig_inode);
1405 ext4_discard_preallocations(donor_inode);
1409 ext4_ext_drop_refs(orig_path);
1412 if (holecheck_path) {
1413 ext4_ext_drop_refs(holecheck_path);
1414 kfree(holecheck_path);
1416 double_up_write_data_sem(orig_inode, donor_inode);
1417 ret2 = mext_inode_double_unlock(orig_inode, donor_inode);