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 struct ext4_extent_header *eh;
81 int ppos, leaf_ppos = path->p_depth;
84 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
86 *extent = ++path[ppos].p_ext;
87 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
92 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
98 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
99 if (path[ppos+1].p_bh)
100 brelse(path[ppos+1].p_bh);
102 sb_bread(inode->i_sb, path[ppos].p_block);
103 if (!path[ppos+1].p_bh)
106 ext_block_hdr(path[ppos+1].p_bh);
108 /* Halfway index block */
109 while (++cur_ppos < leaf_ppos) {
110 path[cur_ppos].p_idx =
111 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
112 path[cur_ppos].p_block =
113 idx_pblock(path[cur_ppos].p_idx);
114 if (path[cur_ppos+1].p_bh)
115 brelse(path[cur_ppos+1].p_bh);
116 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
117 path[cur_ppos].p_block);
118 if (!path[cur_ppos+1].p_bh)
120 path[cur_ppos+1].p_hdr =
121 ext_block_hdr(path[cur_ppos+1].p_bh);
124 path[leaf_ppos].p_ext = *extent = NULL;
126 eh = path[leaf_ppos].p_hdr;
127 if (le16_to_cpu(eh->eh_entries) == 0)
128 /* empty leaf is found */
132 path[leaf_ppos].p_ext = *extent =
133 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
134 path[leaf_ppos].p_block =
135 ext_pblock(path[leaf_ppos].p_ext);
139 /* We found the last extent */
144 * mext_check_null_inode - NULL check for two inodes
146 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
149 mext_check_null_inode(struct inode *inode1, struct inode *inode2,
150 const char *function)
154 if (inode1 == NULL) {
155 ext4_error(inode2->i_sb, function,
156 "Both inodes should not be NULL: "
157 "inode1 NULL inode2 %lu", inode2->i_ino);
159 } else if (inode2 == NULL) {
160 ext4_error(inode1->i_sb, function,
161 "Both inodes should not be NULL: "
162 "inode1 %lu inode2 NULL", inode1->i_ino);
169 * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
171 * @orig_inode: original inode structure
172 * @donor_inode: donor inode structure
173 * Acquire write lock of i_data_sem of the two inodes (orig and donor) by
177 double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
179 struct inode *first = orig_inode, *second = donor_inode;
182 * Use the inode number to provide the stable locking order instead
183 * of its address, because the C language doesn't guarantee you can
184 * compare pointers that don't come from the same array.
186 if (donor_inode->i_ino < orig_inode->i_ino) {
191 down_write(&EXT4_I(first)->i_data_sem);
192 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
196 * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
198 * @orig_inode: original inode structure to be released its lock first
199 * @donor_inode: donor inode structure to be released its lock second
200 * Release write lock of i_data_sem of two inodes (orig and donor).
203 double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
205 up_write(&EXT4_I(orig_inode)->i_data_sem);
206 up_write(&EXT4_I(donor_inode)->i_data_sem);
210 * mext_insert_across_blocks - Insert extents across leaf block
212 * @handle: journal handle
213 * @orig_inode: original inode
214 * @o_start: first original extent to be changed
215 * @o_end: last original extent to be changed
216 * @start_ext: first new extent to be inserted
217 * @new_ext: middle of new extent to be inserted
218 * @end_ext: last new extent to be inserted
220 * Allocate a new leaf block and insert extents into it. Return 0 on success,
221 * or a negative error value on failure.
224 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
225 struct ext4_extent *o_start, struct ext4_extent *o_end,
226 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
227 struct ext4_extent *end_ext)
229 struct ext4_ext_path *orig_path = NULL;
230 ext4_lblk_t eblock = 0;
235 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
236 if (o_start == o_end) {
238 /* start_ext new_ext end_ext
239 * donor |---------|-----------|--------|
240 * orig |------------------------------|
245 /* start_ext new_ext end_ext
246 * donor |---------|----------|---------|
247 * orig |---------------|--------------|
249 o_end->ee_block = end_ext->ee_block;
250 o_end->ee_len = end_ext->ee_len;
251 ext4_ext_store_pblock(o_end, ext_pblock(end_ext));
254 o_start->ee_len = start_ext->ee_len;
257 } else if (start_ext->ee_len && new_ext->ee_len &&
258 !end_ext->ee_len && o_start == o_end) {
261 * donor |--------------|---------------|
262 * orig |------------------------------|
264 o_start->ee_len = start_ext->ee_len;
267 } else if (!start_ext->ee_len && new_ext->ee_len &&
268 end_ext->ee_len && o_start == o_end) {
271 * donor |--------------|---------------|
272 * orig |------------------------------|
274 o_end->ee_block = end_ext->ee_block;
275 o_end->ee_len = end_ext->ee_len;
276 ext4_ext_store_pblock(o_end, ext_pblock(end_ext));
279 * Set 0 to the extent block if new_ext was
282 if (new_ext->ee_block)
283 eblock = le32_to_cpu(new_ext->ee_block);
287 ext4_debug("ext4 move extent: Unexpected insert case\n");
292 err = get_ext_path(orig_inode, eblock, &orig_path);
296 if (ext4_ext_insert_extent(handle, orig_inode,
297 orig_path, new_ext, 0))
302 err = get_ext_path(orig_inode,
303 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
307 if (ext4_ext_insert_extent(handle, orig_inode,
308 orig_path, end_ext, 0))
313 ext4_ext_drop_refs(orig_path);
322 * mext_insert_inside_block - Insert new extent to the extent block
324 * @o_start: first original extent to be moved
325 * @o_end: last original extent to be moved
326 * @start_ext: first new extent to be inserted
327 * @new_ext: middle of new extent to be inserted
328 * @end_ext: last new extent to be inserted
329 * @eh: extent header of target leaf block
330 * @range_to_move: used to decide how to insert extent
332 * Insert extents into the leaf block. The extent (@o_start) is overwritten
333 * by inserted extents.
336 mext_insert_inside_block(struct ext4_extent *o_start,
337 struct ext4_extent *o_end,
338 struct ext4_extent *start_ext,
339 struct ext4_extent *new_ext,
340 struct ext4_extent *end_ext,
341 struct ext4_extent_header *eh,
347 /* Move the existing extents */
348 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
349 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
350 (unsigned long)(o_end + 1);
351 memmove(o_end + 1 + range_to_move, o_end + 1, len);
354 /* Insert start entry */
355 if (start_ext->ee_len)
356 o_start[i++].ee_len = start_ext->ee_len;
358 /* Insert new entry */
359 if (new_ext->ee_len) {
360 o_start[i] = *new_ext;
361 ext4_ext_store_pblock(&o_start[i++], ext_pblock(new_ext));
364 /* Insert end entry */
366 o_start[i] = *end_ext;
368 /* Increment the total entries counter on the extent block */
369 le16_add_cpu(&eh->eh_entries, range_to_move);
373 * mext_insert_extents - Insert new extent
375 * @handle: journal handle
376 * @orig_inode: original inode
377 * @orig_path: path indicates first extent to be changed
378 * @o_start: first original extent to be changed
379 * @o_end: last original extent to be changed
380 * @start_ext: first new extent to be inserted
381 * @new_ext: middle of new extent to be inserted
382 * @end_ext: last new extent to be inserted
384 * Call the function to insert extents. If we cannot add more extents into
385 * the leaf block, we call mext_insert_across_blocks() to create a
386 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
387 * on success, or a negative error value on failure.
390 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
391 struct ext4_ext_path *orig_path,
392 struct ext4_extent *o_start,
393 struct ext4_extent *o_end,
394 struct ext4_extent *start_ext,
395 struct ext4_extent *new_ext,
396 struct ext4_extent *end_ext)
398 struct ext4_extent_header *eh;
399 unsigned long need_slots, slots_range;
400 int range_to_move, depth, ret;
403 * The extents need to be inserted
404 * start_extent + new_extent + end_extent.
406 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
407 (new_ext->ee_len ? 1 : 0);
409 /* The number of slots between start and end */
410 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
411 / sizeof(struct ext4_extent);
413 /* Range to move the end of extent */
414 range_to_move = need_slots - slots_range;
415 depth = orig_path->p_depth;
417 eh = orig_path->p_hdr;
420 /* Register to journal */
421 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
427 if (range_to_move > 0 &&
428 (range_to_move > le16_to_cpu(eh->eh_max)
429 - le16_to_cpu(eh->eh_entries))) {
431 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
432 o_end, start_ext, new_ext, end_ext);
436 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
437 end_ext, eh, range_to_move);
440 ret = ext4_handle_dirty_metadata(handle, orig_inode,
445 ret = ext4_mark_inode_dirty(handle, orig_inode);
454 * mext_leaf_block - Move one leaf extent block into the inode.
456 * @handle: journal handle
457 * @orig_inode: original inode
458 * @orig_path: path indicates first extent to be changed
459 * @dext: donor extent
460 * @from: start offset on the target file
462 * In order to insert extents into the leaf block, we must divide the extent
463 * in the leaf block into three extents. The one is located to be inserted
464 * extents, and the others are located around it.
466 * Therefore, this function creates structures to save extents of the leaf
467 * block, and inserts extents by calling mext_insert_extents() with
468 * created extents. Return 0 on success, or a negative error value on failure.
471 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
472 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
475 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
476 struct ext4_extent new_ext, start_ext, end_ext;
477 ext4_lblk_t new_ext_end;
478 ext4_fsblk_t new_phys_end;
479 int oext_alen, new_ext_alen, end_ext_alen;
480 int depth = ext_depth(orig_inode);
483 o_start = o_end = oext = orig_path[depth].p_ext;
484 oext_alen = ext4_ext_get_actual_len(oext);
485 start_ext.ee_len = end_ext.ee_len = 0;
487 new_ext.ee_block = cpu_to_le32(*from);
488 ext4_ext_store_pblock(&new_ext, ext_pblock(dext));
489 new_ext.ee_len = dext->ee_len;
490 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
491 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
492 new_phys_end = ext_pblock(&new_ext) + new_ext_alen - 1;
495 * Case: original extent is first
500 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
501 le32_to_cpu(new_ext.ee_block) <
502 le32_to_cpu(oext->ee_block) + oext_alen) {
503 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
504 le32_to_cpu(oext->ee_block));
505 copy_extent_status(oext, &start_ext);
506 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
509 * We can merge new_ext into previous extent,
510 * if these are contiguous and same extent type.
512 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
515 start_ext.ee_len = cpu_to_le16(
516 ext4_ext_get_actual_len(prev_ext) +
518 copy_extent_status(prev_ext, &start_ext);
524 * Case: new_ext_end must be less than oext
528 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
529 ext4_error(orig_inode->i_sb, __func__,
530 "new_ext_end(%u) should be less than or equal to "
531 "oext->ee_block(%u) + oext_alen(%d) - 1",
532 new_ext_end, le32_to_cpu(oext->ee_block),
539 * Case: new_ext is smaller than original extent
540 * oext |---------------|
541 * new_ext |-----------|
544 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
545 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
547 cpu_to_le16(le32_to_cpu(oext->ee_block) +
548 oext_alen - 1 - new_ext_end);
549 copy_extent_status(oext, &end_ext);
550 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
551 ext4_ext_store_pblock(&end_ext,
552 (ext_pblock(o_end) + oext_alen - end_ext_alen));
554 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
555 oext_alen - end_ext_alen);
558 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
559 o_end, &start_ext, &new_ext, &end_ext);
565 * mext_calc_swap_extents - Calculate extents for extent swapping.
567 * @tmp_dext: the extent that will belong to the original inode
568 * @tmp_oext: the extent that will belong to the donor inode
569 * @orig_off: block offset of original inode
570 * @donor_off: block offset of donor inode
571 * @max_count: the maximum length of extents
573 * Return 0 on success, or a negative error value on failure.
576 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
577 struct ext4_extent *tmp_oext,
578 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
579 ext4_lblk_t max_count)
581 ext4_lblk_t diff, orig_diff;
582 struct ext4_extent dext_old, oext_old;
584 BUG_ON(orig_off != donor_off);
586 /* original and donor extents have to cover the same block offset */
587 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
588 le32_to_cpu(tmp_oext->ee_block) +
589 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
592 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
593 le32_to_cpu(tmp_dext->ee_block) +
594 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
597 dext_old = *tmp_dext;
598 oext_old = *tmp_oext;
600 /* When tmp_dext is too large, pick up the target range. */
601 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
603 ext4_ext_store_pblock(tmp_dext, ext_pblock(tmp_dext) + diff);
605 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
606 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
608 if (max_count < ext4_ext_get_actual_len(tmp_dext))
609 tmp_dext->ee_len = cpu_to_le16(max_count);
611 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
612 ext4_ext_store_pblock(tmp_oext, ext_pblock(tmp_oext) + orig_diff);
614 /* Adjust extent length if donor extent is larger than orig */
615 if (ext4_ext_get_actual_len(tmp_dext) >
616 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
617 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
620 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
622 copy_extent_status(&oext_old, tmp_dext);
623 copy_extent_status(&dext_old, tmp_oext);
629 * mext_replace_branches - Replace original extents with new extents
631 * @handle: journal handle
632 * @orig_inode: original inode
633 * @donor_inode: donor inode
634 * @from: block offset of orig_inode
635 * @count: block count to be replaced
636 * @err: pointer to save return value
638 * Replace original inode extents and donor inode extents page by page.
639 * We implement this replacement in the following three steps:
640 * 1. Save the block information of original and donor inodes into
642 * 2. Change the block information of original inode to point at the
643 * donor inode blocks.
644 * 3. Change the block information of donor inode to point at the saved
645 * original inode blocks in the dummy extents.
647 * Return replaced block count.
650 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
651 struct inode *donor_inode, ext4_lblk_t from,
652 ext4_lblk_t count, int *err)
654 struct ext4_ext_path *orig_path = NULL;
655 struct ext4_ext_path *donor_path = NULL;
656 struct ext4_extent *oext, *dext;
657 struct ext4_extent tmp_dext, tmp_oext;
658 ext4_lblk_t orig_off = from, donor_off = from;
660 int replaced_count = 0;
663 /* Get the original extent for the block "orig_off" */
664 *err = get_ext_path(orig_inode, orig_off, &orig_path);
668 /* Get the donor extent for the head */
669 *err = get_ext_path(donor_inode, donor_off, &donor_path);
672 depth = ext_depth(orig_inode);
673 oext = orig_path[depth].p_ext;
676 depth = ext_depth(donor_inode);
677 dext = donor_path[depth].p_ext;
680 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
685 /* Loop for the donor extents */
687 /* The extent for donor must be found. */
689 ext4_error(donor_inode->i_sb, __func__,
690 "The extent for donor must be found");
693 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
694 ext4_error(donor_inode->i_sb, __func__,
695 "Donor offset(%u) and the first block of donor "
696 "extent(%u) should be equal",
698 le32_to_cpu(tmp_dext.ee_block));
703 /* Set donor extent to orig extent */
704 *err = mext_leaf_block(handle, orig_inode,
705 orig_path, &tmp_dext, &orig_off);
709 /* Set orig extent to donor extent */
710 *err = mext_leaf_block(handle, donor_inode,
711 donor_path, &tmp_oext, &donor_off);
715 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
716 replaced_count += dext_alen;
717 donor_off += dext_alen;
718 orig_off += dext_alen;
720 /* Already moved the expected blocks */
721 if (replaced_count >= count)
725 ext4_ext_drop_refs(orig_path);
726 *err = get_ext_path(orig_inode, orig_off, &orig_path);
729 depth = ext_depth(orig_inode);
730 oext = orig_path[depth].p_ext;
734 ext4_ext_drop_refs(donor_path);
735 *err = get_ext_path(donor_inode, donor_off, &donor_path);
738 depth = ext_depth(donor_inode);
739 dext = donor_path[depth].p_ext;
742 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
743 donor_off, count - replaced_count);
750 ext4_ext_drop_refs(orig_path);
754 ext4_ext_drop_refs(donor_path);
758 return replaced_count;
762 * move_extent_per_page - Move extent data per page
764 * @o_filp: file structure of original file
765 * @donor_inode: donor inode
766 * @orig_page_offset: page index on original file
767 * @data_offset_in_page: block index where data swapping starts
768 * @block_len_in_page: the number of blocks to be swapped
769 * @uninit: orig extent is uninitialized or not
770 * @err: pointer to save return value
772 * Save the data in original inode blocks and replace original inode extents
773 * with donor inode extents by calling mext_replace_branches().
774 * Finally, write out the saved data in new original inode blocks. Return
775 * replaced block count.
778 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
779 pgoff_t orig_page_offset, int data_offset_in_page,
780 int block_len_in_page, int uninit, int *err)
782 struct inode *orig_inode = o_filp->f_dentry->d_inode;
783 struct address_space *mapping = orig_inode->i_mapping;
784 struct buffer_head *bh;
785 struct page *page = NULL;
786 const struct address_space_operations *a_ops = mapping->a_ops;
788 ext4_lblk_t orig_blk_offset;
789 long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
790 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
791 unsigned int w_flags = 0;
792 unsigned int tmp_data_size, data_size, replaced_size;
796 int replaced_count = 0;
797 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
800 * It needs twice the amount of ordinary journal buffers because
801 * inode and donor_inode may change each different metadata blocks.
803 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
804 handle = ext4_journal_start(orig_inode, jblocks);
805 if (IS_ERR(handle)) {
806 *err = PTR_ERR(handle);
810 if (segment_eq(get_fs(), KERNEL_DS))
811 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
813 orig_blk_offset = orig_page_offset * blocks_per_page +
817 * If orig extent is uninitialized one,
818 * it's not necessary force the page into memory
819 * and then force it to be written out again.
820 * Just swap data blocks between orig and donor.
824 * Protect extent trees against block allocations
827 double_down_write_data_sem(orig_inode, donor_inode);
828 replaced_count = mext_replace_branches(handle, orig_inode,
829 donor_inode, orig_blk_offset,
830 block_len_in_page, err);
832 /* Clear the inode cache not to refer to the old data */
833 ext4_ext_invalidate_cache(orig_inode);
834 ext4_ext_invalidate_cache(donor_inode);
835 double_up_write_data_sem(orig_inode, donor_inode);
839 offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
841 /* Calculate data_size */
842 if ((orig_blk_offset + block_len_in_page - 1) ==
843 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
844 /* Replace the last block */
845 tmp_data_size = orig_inode->i_size & (blocksize - 1);
847 * If data_size equal zero, it shows data_size is multiples of
848 * blocksize. So we set appropriate value.
850 if (tmp_data_size == 0)
851 tmp_data_size = blocksize;
853 data_size = tmp_data_size +
854 ((block_len_in_page - 1) << orig_inode->i_blkbits);
856 data_size = block_len_in_page << orig_inode->i_blkbits;
858 replaced_size = data_size;
860 *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags,
862 if (unlikely(*err < 0))
865 if (!PageUptodate(page)) {
866 mapping->a_ops->readpage(o_filp, page);
871 * try_to_release_page() doesn't call releasepage in writeback mode.
872 * We should care about the order of writing to the same file
873 * by multiple move extent processes.
874 * It needs to call wait_on_page_writeback() to wait for the
875 * writeback of the page.
877 if (PageWriteback(page))
878 wait_on_page_writeback(page);
880 /* Release old bh and drop refs */
881 try_to_release_page(page, 0);
883 /* Protect extent trees against block allocations via delalloc */
884 double_down_write_data_sem(orig_inode, donor_inode);
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;
894 double_up_write_data_sem(orig_inode, donor_inode);
899 /* Clear the inode cache not to refer to the old data */
900 ext4_ext_invalidate_cache(orig_inode);
901 ext4_ext_invalidate_cache(donor_inode);
903 double_up_write_data_sem(orig_inode, donor_inode);
905 if (!page_has_buffers(page))
906 create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0);
908 bh = page_buffers(page);
909 for (i = 0; i < data_offset_in_page; i++)
910 bh = bh->b_this_page;
912 for (i = 0; i < block_len_in_page; i++) {
913 *err = ext4_get_block(orig_inode,
914 (sector_t)(orig_blk_offset + i), bh, 0);
918 if (bh->b_this_page != NULL)
919 bh = bh->b_this_page;
922 *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size,
927 if (unlikely(page)) {
928 if (PageLocked(page))
930 page_cache_release(page);
931 ext4_journal_stop(handle);
934 ext4_journal_stop(handle);
939 return replaced_count;
943 * mext_check_argumants - Check whether move extent can be done
945 * @orig_inode: original inode
946 * @donor_inode: donor inode
947 * @orig_start: logical start offset in block for orig
948 * @donor_start: logical start offset in block for donor
949 * @len: the number of blocks to be moved
951 * Check the arguments of ext4_move_extents() whether the files can be
952 * exchanged with each other.
953 * Return 0 on success, or a negative error value on failure.
956 mext_check_arguments(struct inode *orig_inode,
957 struct inode *donor_inode, __u64 orig_start,
958 __u64 donor_start, __u64 *len)
960 ext4_lblk_t orig_blocks, donor_blocks;
961 unsigned int blkbits = orig_inode->i_blkbits;
962 unsigned int blocksize = 1 << blkbits;
964 /* Regular file check */
965 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
966 ext4_debug("ext4 move extent: The argument files should be "
967 "regular file [ino:orig %lu, donor %lu]\n",
968 orig_inode->i_ino, donor_inode->i_ino);
972 /* Ext4 move extent does not support swapfile */
973 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
974 ext4_debug("ext4 move extent: The argument files should "
975 "not be swapfile [ino:orig %lu, donor %lu]\n",
976 orig_inode->i_ino, donor_inode->i_ino);
980 /* Files should be in the same ext4 FS */
981 if (orig_inode->i_sb != donor_inode->i_sb) {
982 ext4_debug("ext4 move extent: The argument files "
983 "should be in same FS [ino:orig %lu, donor %lu]\n",
984 orig_inode->i_ino, donor_inode->i_ino);
988 /* Ext4 move extent supports only extent based file */
989 if (!(EXT4_I(orig_inode)->i_flags & EXT4_EXTENTS_FL)) {
990 ext4_debug("ext4 move extent: orig file is not extents "
991 "based file [ino:orig %lu]\n", orig_inode->i_ino);
993 } else if (!(EXT4_I(donor_inode)->i_flags & EXT4_EXTENTS_FL)) {
994 ext4_debug("ext4 move extent: donor file is not extents "
995 "based file [ino:donor %lu]\n", donor_inode->i_ino);
999 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1000 ext4_debug("ext4 move extent: File size is 0 byte\n");
1004 /* Start offset should be same */
1005 if (orig_start != donor_start) {
1006 ext4_debug("ext4 move extent: orig and donor's start "
1007 "offset are not same [ino:orig %lu, donor %lu]\n",
1008 orig_inode->i_ino, donor_inode->i_ino);
1012 if ((orig_start > EXT_MAX_BLOCK) ||
1013 (donor_start > EXT_MAX_BLOCK) ||
1014 (*len > EXT_MAX_BLOCK) ||
1015 (orig_start + *len > EXT_MAX_BLOCK)) {
1016 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1017 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCK,
1018 orig_inode->i_ino, donor_inode->i_ino);
1022 if (orig_inode->i_size > donor_inode->i_size) {
1023 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1024 /* TODO: eliminate this artificial restriction */
1025 if (orig_start >= donor_blocks) {
1026 ext4_debug("ext4 move extent: orig start offset "
1027 "[%llu] should be less than donor file blocks "
1028 "[%u] [ino:orig %lu, donor %lu]\n",
1029 orig_start, donor_blocks,
1030 orig_inode->i_ino, donor_inode->i_ino);
1034 /* TODO: eliminate this artificial restriction */
1035 if (orig_start + *len > donor_blocks) {
1036 ext4_debug("ext4 move extent: End offset [%llu] should "
1037 "be less than donor file blocks [%u]."
1038 "So adjust length from %llu to %llu "
1039 "[ino:orig %lu, donor %lu]\n",
1040 orig_start + *len, donor_blocks,
1041 *len, donor_blocks - orig_start,
1042 orig_inode->i_ino, donor_inode->i_ino);
1043 *len = donor_blocks - orig_start;
1046 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1047 if (orig_start >= orig_blocks) {
1048 ext4_debug("ext4 move extent: start offset [%llu] "
1049 "should be less than original file blocks "
1050 "[%u] [ino:orig %lu, donor %lu]\n",
1051 orig_start, orig_blocks,
1052 orig_inode->i_ino, donor_inode->i_ino);
1056 if (orig_start + *len > orig_blocks) {
1057 ext4_debug("ext4 move extent: Adjust length "
1058 "from %llu to %llu. Because it should be "
1059 "less than original file blocks "
1060 "[ino:orig %lu, donor %lu]\n",
1061 *len, orig_blocks - orig_start,
1062 orig_inode->i_ino, donor_inode->i_ino);
1063 *len = orig_blocks - orig_start;
1068 ext4_debug("ext4 move extent: len should not be 0 "
1069 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1070 donor_inode->i_ino);
1078 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1080 * @inode1: the inode structure
1081 * @inode2: the inode structure
1083 * Lock two inodes' i_mutex by i_ino order.
1084 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1087 mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1091 BUG_ON(inode1 == NULL && inode2 == NULL);
1093 ret = mext_check_null_inode(inode1, inode2, __func__);
1097 if (inode1 == inode2) {
1098 mutex_lock(&inode1->i_mutex);
1102 if (inode1->i_ino < inode2->i_ino) {
1103 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1104 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1106 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1107 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1115 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1117 * @inode1: the inode that is released first
1118 * @inode2: the inode that is released second
1120 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1124 mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1128 BUG_ON(inode1 == NULL && inode2 == NULL);
1130 ret = mext_check_null_inode(inode1, inode2, __func__);
1135 mutex_unlock(&inode1->i_mutex);
1137 if (inode2 && inode2 != inode1)
1138 mutex_unlock(&inode2->i_mutex);
1145 * ext4_move_extents - Exchange the specified range of a file
1147 * @o_filp: file structure of the original file
1148 * @d_filp: file structure of the donor file
1149 * @orig_start: start offset in block for orig
1150 * @donor_start: start offset in block for donor
1151 * @len: the number of blocks to be moved
1152 * @moved_len: moved block length
1154 * This function returns 0 and moved block length is set in moved_len
1155 * if succeed, otherwise returns error value.
1157 * Note: ext4_move_extents() proceeds the following order.
1158 * 1:ext4_move_extents() calculates the last block number of moving extent
1159 * function by the start block number (orig_start) and the number of blocks
1160 * to be moved (len) specified as arguments.
1161 * If the {orig, donor}_start points a hole, the extent's start offset
1162 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1163 * after hole behind.
1164 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1165 * or the ext_cur exceeds the block_end which is last logical block number.
1166 * 3:To get the length of continues area, call mext_next_extent()
1167 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1168 * until find un-continuous extent, the start logical block number exceeds
1169 * the block_end or the extent points to the last extent.
1170 * 4:Exchange the original inode data with donor inode data
1171 * from orig_page_offset to seq_end_page.
1172 * The start indexes of data are specified as arguments.
1173 * That of the original inode is orig_page_offset,
1174 * and the donor inode is also orig_page_offset
1175 * (To easily handle blocksize != pagesize case, the offset for the
1176 * donor inode is block unit).
1177 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1178 * then returns to step 2.
1179 * 6:Release holecheck_path, orig_path and set the len to moved_len
1180 * which shows the number of moved blocks.
1181 * The moved_len is useful for the command to calculate the file offset
1182 * for starting next move extent ioctl.
1183 * 7:Return 0 on success, or a negative error value on failure.
1186 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1187 __u64 orig_start, __u64 donor_start, __u64 len,
1190 struct inode *orig_inode = o_filp->f_dentry->d_inode;
1191 struct inode *donor_inode = d_filp->f_dentry->d_inode;
1192 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1193 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1194 ext4_lblk_t block_start = orig_start;
1195 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1196 ext4_lblk_t rest_blocks;
1197 pgoff_t orig_page_offset = 0, seq_end_page;
1198 int ret1, ret2, depth, last_extent = 0;
1199 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1200 int data_offset_in_page;
1201 int block_len_in_page;
1204 /* orig and donor should be different file */
1205 if (orig_inode->i_ino == donor_inode->i_ino) {
1206 ext4_debug("ext4 move extent: The argument files should not "
1207 "be same file [ino:orig %lu, donor %lu]\n",
1208 orig_inode->i_ino, donor_inode->i_ino);
1212 /* Protect orig and donor inodes against a truncate */
1213 ret1 = mext_inode_double_lock(orig_inode, donor_inode);
1217 /* Protect extent tree against block allocations via delalloc */
1218 double_down_write_data_sem(orig_inode, donor_inode);
1219 /* Check the filesystem environment whether move_extent can be done */
1220 ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start,
1225 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1226 block_end = block_start + len - 1;
1227 if (file_end < block_end)
1228 len -= block_end - file_end;
1230 ret1 = get_ext_path(orig_inode, block_start, &orig_path);
1234 /* Get path structure to check the hole */
1235 ret1 = get_ext_path(orig_inode, block_start, &holecheck_path);
1239 depth = ext_depth(orig_inode);
1240 ext_cur = holecheck_path[depth].p_ext;
1243 * Get proper starting location of block replacement if block_start was
1246 if (le32_to_cpu(ext_cur->ee_block) +
1247 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1249 * The hole exists between extents or the tail of
1252 last_extent = mext_next_extent(orig_inode,
1253 holecheck_path, &ext_cur);
1254 if (last_extent < 0) {
1258 last_extent = mext_next_extent(orig_inode, orig_path,
1260 if (last_extent < 0) {
1264 seq_start = le32_to_cpu(ext_cur->ee_block);
1265 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1266 /* The hole exists at the beginning of original file. */
1267 seq_start = le32_to_cpu(ext_cur->ee_block);
1269 seq_start = block_start;
1271 /* No blocks within the specified range. */
1272 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1273 ext4_debug("ext4 move extent: The specified range of file "
1274 "may be the hole\n");
1279 /* Adjust start blocks */
1280 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1281 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1282 max(le32_to_cpu(ext_cur->ee_block), block_start);
1284 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1285 seq_blocks += add_blocks;
1287 /* Adjust tail blocks */
1288 if (seq_start + seq_blocks - 1 > block_end)
1289 seq_blocks = block_end - seq_start + 1;
1292 last_extent = mext_next_extent(orig_inode, holecheck_path,
1294 if (last_extent < 0) {
1298 add_blocks = ext4_ext_get_actual_len(ext_cur);
1301 * Extend the length of contiguous block (seq_blocks)
1302 * if extents are contiguous.
1304 if (ext4_can_extents_be_merged(orig_inode,
1305 ext_prev, ext_cur) &&
1306 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1310 /* Is original extent is uninitialized */
1311 uninit = ext4_ext_is_uninitialized(ext_prev);
1313 data_offset_in_page = seq_start % blocks_per_page;
1316 * Calculate data blocks count that should be swapped
1317 * at the first page.
1319 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1320 /* Swapped blocks are across pages */
1322 blocks_per_page - data_offset_in_page;
1324 /* Swapped blocks are in a page */
1325 block_len_in_page = seq_blocks;
1328 orig_page_offset = seq_start >>
1329 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1330 seq_end_page = (seq_start + seq_blocks - 1) >>
1331 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1332 seq_start = le32_to_cpu(ext_cur->ee_block);
1333 rest_blocks = seq_blocks;
1336 * Up semaphore to avoid following problems:
1337 * a. transaction deadlock among ext4_journal_start,
1338 * ->write_begin via pagefault, and jbd2_journal_commit
1339 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1340 * in move_extent_per_page
1342 double_up_write_data_sem(orig_inode, donor_inode);
1344 while (orig_page_offset <= seq_end_page) {
1346 /* Swap original branches with new branches */
1347 block_len_in_page = move_extent_per_page(
1348 o_filp, donor_inode,
1350 data_offset_in_page,
1351 block_len_in_page, uninit,
1354 /* Count how many blocks we have exchanged */
1355 *moved_len += block_len_in_page;
1358 if (*moved_len > len) {
1359 ext4_error(orig_inode->i_sb, __func__,
1360 "We replaced blocks too much! "
1361 "sum of replaced: %llu requested: %llu",
1368 data_offset_in_page = 0;
1369 rest_blocks -= block_len_in_page;
1370 if (rest_blocks > blocks_per_page)
1371 block_len_in_page = blocks_per_page;
1373 block_len_in_page = rest_blocks;
1376 double_down_write_data_sem(orig_inode, donor_inode);
1380 /* Decrease buffer counter */
1382 ext4_ext_drop_refs(holecheck_path);
1383 ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path);
1386 depth = holecheck_path->p_depth;
1388 /* Decrease buffer counter */
1390 ext4_ext_drop_refs(orig_path);
1391 ret1 = get_ext_path(orig_inode, seq_start, &orig_path);
1395 ext_cur = holecheck_path[depth].p_ext;
1396 add_blocks = ext4_ext_get_actual_len(ext_cur);
1402 ext4_discard_preallocations(orig_inode);
1403 ext4_discard_preallocations(donor_inode);
1407 ext4_ext_drop_refs(orig_path);
1410 if (holecheck_path) {
1411 ext4_ext_drop_refs(holecheck_path);
1412 kfree(holecheck_path);
1414 double_up_write_data_sem(orig_inode, donor_inode);
1415 ret2 = mext_inode_double_unlock(orig_inode, donor_inode);