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"
21 #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 **orig_path)
38 struct ext4_ext_path *path;
40 path = ext4_ext_find_extent(inode, lblock, *orig_path, EXT4_EX_NOCACHE);
43 else if (path[ext_depth(inode)].p_ext == NULL)
52 * copy_extent_status - Copy the extent's initialization status
54 * @src: an extent for getting initialize status
55 * @dest: an extent to be set the status
58 copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
60 if (ext4_ext_is_unwritten(src))
61 ext4_ext_mark_unwritten(dest);
63 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
67 * mext_next_extent - Search for the next extent and set it to "extent"
69 * @inode: inode which is searched
70 * @path: this will obtain data for the next extent
71 * @extent: pointer to the next extent we have just gotten
73 * Search the next extent in the array of ext4_ext_path structure (@path)
74 * and set it to ext4_extent structure (@extent). In addition, the member of
75 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
76 * ext4_ext_path structure refers to the last extent, or a negative error
80 mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
81 struct ext4_extent **extent)
83 struct ext4_extent_header *eh;
84 int ppos, leaf_ppos = path->p_depth;
87 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
89 *extent = ++path[ppos].p_ext;
90 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
95 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
101 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
102 if (path[ppos+1].p_bh)
103 brelse(path[ppos+1].p_bh);
105 sb_bread(inode->i_sb, path[ppos].p_block);
106 if (!path[ppos+1].p_bh)
109 ext_block_hdr(path[ppos+1].p_bh);
111 /* Halfway index block */
112 while (++cur_ppos < leaf_ppos) {
113 path[cur_ppos].p_idx =
114 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
115 path[cur_ppos].p_block =
116 ext4_idx_pblock(path[cur_ppos].p_idx);
117 if (path[cur_ppos+1].p_bh)
118 brelse(path[cur_ppos+1].p_bh);
119 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
120 path[cur_ppos].p_block);
121 if (!path[cur_ppos+1].p_bh)
123 path[cur_ppos+1].p_hdr =
124 ext_block_hdr(path[cur_ppos+1].p_bh);
127 path[leaf_ppos].p_ext = *extent = NULL;
129 eh = path[leaf_ppos].p_hdr;
130 if (le16_to_cpu(eh->eh_entries) == 0)
131 /* empty leaf is found */
135 path[leaf_ppos].p_ext = *extent =
136 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
137 path[leaf_ppos].p_block =
138 ext4_ext_pblock(path[leaf_ppos].p_ext);
142 /* We found the last extent */
147 * ext4_double_down_write_data_sem - Acquire two inodes' write lock
150 * Acquire write lock of i_data_sem of the two inodes
153 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
155 if (first < second) {
156 down_write(&EXT4_I(first)->i_data_sem);
157 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
159 down_write(&EXT4_I(second)->i_data_sem);
160 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
166 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
168 * @orig_inode: original inode structure to be released its lock first
169 * @donor_inode: donor inode structure to be released its lock second
170 * Release write lock of i_data_sem of two inodes (orig and donor).
173 ext4_double_up_write_data_sem(struct inode *orig_inode,
174 struct inode *donor_inode)
176 up_write(&EXT4_I(orig_inode)->i_data_sem);
177 up_write(&EXT4_I(donor_inode)->i_data_sem);
181 * mext_insert_across_blocks - Insert extents across leaf block
183 * @handle: journal handle
184 * @orig_inode: original inode
185 * @o_start: first original extent to be changed
186 * @o_end: last original extent to be changed
187 * @start_ext: first new extent to be inserted
188 * @new_ext: middle of new extent to be inserted
189 * @end_ext: last new extent to be inserted
191 * Allocate a new leaf block and insert extents into it. Return 0 on success,
192 * or a negative error value on failure.
195 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
196 struct ext4_extent *o_start, struct ext4_extent *o_end,
197 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
198 struct ext4_extent *end_ext)
200 struct ext4_ext_path *orig_path = NULL;
201 ext4_lblk_t eblock = 0;
206 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
207 if (o_start == o_end) {
209 /* start_ext new_ext end_ext
210 * donor |---------|-----------|--------|
211 * orig |------------------------------|
216 /* start_ext new_ext end_ext
217 * donor |---------|----------|---------|
218 * orig |---------------|--------------|
220 o_end->ee_block = end_ext->ee_block;
221 o_end->ee_len = end_ext->ee_len;
222 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
225 o_start->ee_len = start_ext->ee_len;
226 eblock = le32_to_cpu(start_ext->ee_block);
229 } else if (start_ext->ee_len && new_ext->ee_len &&
230 !end_ext->ee_len && o_start == o_end) {
233 * donor |--------------|---------------|
234 * orig |------------------------------|
236 o_start->ee_len = start_ext->ee_len;
237 eblock = le32_to_cpu(start_ext->ee_block);
240 } else if (!start_ext->ee_len && new_ext->ee_len &&
241 end_ext->ee_len && o_start == o_end) {
244 * donor |--------------|---------------|
245 * orig |------------------------------|
247 o_end->ee_block = end_ext->ee_block;
248 o_end->ee_len = end_ext->ee_len;
249 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
252 * Set 0 to the extent block if new_ext was
255 if (new_ext->ee_block)
256 eblock = le32_to_cpu(new_ext->ee_block);
260 ext4_debug("ext4 move extent: Unexpected insert case\n");
265 err = get_ext_path(orig_inode, eblock, &orig_path);
269 if (ext4_ext_insert_extent(handle, orig_inode,
270 orig_path, new_ext, 0))
275 err = get_ext_path(orig_inode,
276 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
280 if (ext4_ext_insert_extent(handle, orig_inode,
281 orig_path, end_ext, 0))
286 ext4_ext_drop_refs(orig_path);
295 * mext_insert_inside_block - Insert new extent to the extent block
297 * @o_start: first original extent to be moved
298 * @o_end: last original extent to be moved
299 * @start_ext: first new extent to be inserted
300 * @new_ext: middle of new extent to be inserted
301 * @end_ext: last new extent to be inserted
302 * @eh: extent header of target leaf block
303 * @range_to_move: used to decide how to insert extent
305 * Insert extents into the leaf block. The extent (@o_start) is overwritten
306 * by inserted extents.
309 mext_insert_inside_block(struct ext4_extent *o_start,
310 struct ext4_extent *o_end,
311 struct ext4_extent *start_ext,
312 struct ext4_extent *new_ext,
313 struct ext4_extent *end_ext,
314 struct ext4_extent_header *eh,
320 /* Move the existing extents */
321 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
322 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
323 (unsigned long)(o_end + 1);
324 memmove(o_end + 1 + range_to_move, o_end + 1, len);
327 /* Insert start entry */
328 if (start_ext->ee_len)
329 o_start[i++].ee_len = start_ext->ee_len;
331 /* Insert new entry */
332 if (new_ext->ee_len) {
333 o_start[i] = *new_ext;
334 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
337 /* Insert end entry */
339 o_start[i] = *end_ext;
341 /* Increment the total entries counter on the extent block */
342 le16_add_cpu(&eh->eh_entries, range_to_move);
346 * mext_insert_extents - Insert new extent
348 * @handle: journal handle
349 * @orig_inode: original inode
350 * @orig_path: path indicates first extent to be changed
351 * @o_start: first original extent to be changed
352 * @o_end: last original extent to be changed
353 * @start_ext: first new extent to be inserted
354 * @new_ext: middle of new extent to be inserted
355 * @end_ext: last new extent to be inserted
357 * Call the function to insert extents. If we cannot add more extents into
358 * the leaf block, we call mext_insert_across_blocks() to create a
359 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
360 * on success, or a negative error value on failure.
363 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
364 struct ext4_ext_path *orig_path,
365 struct ext4_extent *o_start,
366 struct ext4_extent *o_end,
367 struct ext4_extent *start_ext,
368 struct ext4_extent *new_ext,
369 struct ext4_extent *end_ext)
371 struct ext4_extent_header *eh;
372 unsigned long need_slots, slots_range;
373 int range_to_move, depth, ret;
376 * The extents need to be inserted
377 * start_extent + new_extent + end_extent.
379 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
380 (new_ext->ee_len ? 1 : 0);
382 /* The number of slots between start and end */
383 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
384 / sizeof(struct ext4_extent);
386 /* Range to move the end of extent */
387 range_to_move = need_slots - slots_range;
388 depth = orig_path->p_depth;
390 eh = orig_path->p_hdr;
393 /* Register to journal */
394 BUFFER_TRACE(orig_path->p_bh, "get_write_access");
395 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
401 if (range_to_move > 0 &&
402 (range_to_move > le16_to_cpu(eh->eh_max)
403 - le16_to_cpu(eh->eh_entries))) {
405 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
406 o_end, start_ext, new_ext, end_ext);
410 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
411 end_ext, eh, range_to_move);
413 return ext4_ext_dirty(handle, orig_inode, orig_path);
417 * mext_leaf_block - Move one leaf extent block into the inode.
419 * @handle: journal handle
420 * @orig_inode: original inode
421 * @orig_path: path indicates first extent to be changed
422 * @dext: donor extent
423 * @from: start offset on the target file
425 * In order to insert extents into the leaf block, we must divide the extent
426 * in the leaf block into three extents. The one is located to be inserted
427 * extents, and the others are located around it.
429 * Therefore, this function creates structures to save extents of the leaf
430 * block, and inserts extents by calling mext_insert_extents() with
431 * created extents. Return 0 on success, or a negative error value on failure.
434 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
435 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
438 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
439 struct ext4_extent new_ext, start_ext, end_ext;
440 ext4_lblk_t new_ext_end;
441 int oext_alen, new_ext_alen, end_ext_alen;
442 int depth = ext_depth(orig_inode);
445 start_ext.ee_block = end_ext.ee_block = 0;
446 o_start = o_end = oext = orig_path[depth].p_ext;
447 oext_alen = ext4_ext_get_actual_len(oext);
448 start_ext.ee_len = end_ext.ee_len = 0;
450 new_ext.ee_block = cpu_to_le32(*from);
451 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
452 new_ext.ee_len = dext->ee_len;
453 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
454 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
457 * Case: original extent is first
462 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
463 le32_to_cpu(new_ext.ee_block) <
464 le32_to_cpu(oext->ee_block) + oext_alen) {
465 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
466 le32_to_cpu(oext->ee_block));
467 start_ext.ee_block = oext->ee_block;
468 copy_extent_status(oext, &start_ext);
469 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
472 * We can merge new_ext into previous extent,
473 * if these are contiguous and same extent type.
475 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
478 start_ext.ee_len = cpu_to_le16(
479 ext4_ext_get_actual_len(prev_ext) +
481 start_ext.ee_block = oext->ee_block;
482 copy_extent_status(prev_ext, &start_ext);
488 * Case: new_ext_end must be less than oext
492 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
493 EXT4_ERROR_INODE(orig_inode,
494 "new_ext_end(%u) should be less than or equal to "
495 "oext->ee_block(%u) + oext_alen(%d) - 1",
496 new_ext_end, le32_to_cpu(oext->ee_block),
503 * Case: new_ext is smaller than original extent
504 * oext |---------------|
505 * new_ext |-----------|
508 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
509 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
511 cpu_to_le16(le32_to_cpu(oext->ee_block) +
512 oext_alen - 1 - new_ext_end);
513 copy_extent_status(oext, &end_ext);
514 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
515 ext4_ext_store_pblock(&end_ext,
516 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
518 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
519 oext_alen - end_ext_alen);
522 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
523 o_end, &start_ext, &new_ext, &end_ext);
529 * mext_calc_swap_extents - Calculate extents for extent swapping.
531 * @tmp_dext: the extent that will belong to the original inode
532 * @tmp_oext: the extent that will belong to the donor inode
533 * @orig_off: block offset of original inode
534 * @donor_off: block offset of donor inode
535 * @max_count: the maximum length of extents
537 * Return 0 on success, or a negative error value on failure.
540 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
541 struct ext4_extent *tmp_oext,
542 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
543 ext4_lblk_t max_count)
545 ext4_lblk_t diff, orig_diff;
546 struct ext4_extent dext_old, oext_old;
548 BUG_ON(orig_off != donor_off);
550 /* original and donor extents have to cover the same block offset */
551 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
552 le32_to_cpu(tmp_oext->ee_block) +
553 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
556 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
557 le32_to_cpu(tmp_dext->ee_block) +
558 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
561 dext_old = *tmp_dext;
562 oext_old = *tmp_oext;
564 /* When tmp_dext is too large, pick up the target range. */
565 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
567 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
568 le32_add_cpu(&tmp_dext->ee_block, diff);
569 le16_add_cpu(&tmp_dext->ee_len, -diff);
571 if (max_count < ext4_ext_get_actual_len(tmp_dext))
572 tmp_dext->ee_len = cpu_to_le16(max_count);
574 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
575 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
577 /* Adjust extent length if donor extent is larger than orig */
578 if (ext4_ext_get_actual_len(tmp_dext) >
579 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
580 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
583 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
585 copy_extent_status(&oext_old, tmp_dext);
586 copy_extent_status(&dext_old, tmp_oext);
592 * mext_check_coverage - Check that all extents in range has the same type
594 * @inode: inode in question
595 * @from: block offset of inode
596 * @count: block count to be checked
597 * @unwritten: extents expected to be unwritten
598 * @err: pointer to save error value
600 * Return 1 if all extents in range has expected type, and zero otherwise.
603 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
604 int unwritten, int *err)
606 struct ext4_ext_path *path = NULL;
607 struct ext4_extent *ext;
609 ext4_lblk_t last = from + count;
610 while (from < last) {
611 *err = get_ext_path(inode, from, &path);
614 ext = path[ext_depth(inode)].p_ext;
615 if (unwritten != ext4_ext_is_unwritten(ext))
617 from += ext4_ext_get_actual_len(ext);
618 ext4_ext_drop_refs(path);
623 ext4_ext_drop_refs(path);
630 * mext_replace_branches - Replace original extents with new extents
632 * @handle: journal handle
633 * @orig_inode: original inode
634 * @donor_inode: donor inode
635 * @from: block offset of orig_inode
636 * @count: block count to be replaced
637 * @err: pointer to save return value
639 * Replace original inode extents and donor inode extents page by page.
640 * We implement this replacement in the following three steps:
641 * 1. Save the block information of original and donor inodes into
643 * 2. Change the block information of original inode to point at the
644 * donor inode blocks.
645 * 3. Change the block information of donor inode to point at the saved
646 * original inode blocks in the dummy extents.
648 * Return replaced block count.
651 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
652 struct inode *donor_inode, ext4_lblk_t from,
653 ext4_lblk_t count, int *err)
655 struct ext4_ext_path *orig_path = NULL;
656 struct ext4_ext_path *donor_path = NULL;
657 struct ext4_extent *oext, *dext;
658 struct ext4_extent tmp_dext, tmp_oext;
659 ext4_lblk_t orig_off = from, donor_off = from;
661 int replaced_count = 0;
664 *err = ext4_es_remove_extent(orig_inode, from, count);
668 *err = ext4_es_remove_extent(donor_inode, from, count);
672 /* Get the original extent for the block "orig_off" */
673 *err = get_ext_path(orig_inode, orig_off, &orig_path);
677 /* Get the donor extent for the head */
678 *err = get_ext_path(donor_inode, donor_off, &donor_path);
681 depth = ext_depth(orig_inode);
682 oext = orig_path[depth].p_ext;
685 depth = ext_depth(donor_inode);
686 dext = donor_path[depth].p_ext;
688 goto missing_donor_extent;
691 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
696 /* Loop for the donor extents */
698 /* The extent for donor must be found. */
699 if (unlikely(!dext)) {
700 missing_donor_extent:
701 EXT4_ERROR_INODE(donor_inode,
702 "The extent for donor must be found");
705 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
706 EXT4_ERROR_INODE(donor_inode,
707 "Donor offset(%u) and the first block of donor "
708 "extent(%u) should be equal",
710 le32_to_cpu(tmp_dext.ee_block));
715 /* Set donor extent to orig extent */
716 *err = mext_leaf_block(handle, orig_inode,
717 orig_path, &tmp_dext, &orig_off);
721 /* Set orig extent to donor extent */
722 *err = mext_leaf_block(handle, donor_inode,
723 donor_path, &tmp_oext, &donor_off);
727 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
728 replaced_count += dext_alen;
729 donor_off += dext_alen;
730 orig_off += dext_alen;
732 BUG_ON(replaced_count > count);
733 /* Already moved the expected blocks */
734 if (replaced_count >= count)
738 ext4_ext_drop_refs(orig_path);
739 *err = get_ext_path(orig_inode, orig_off, &orig_path);
742 depth = ext_depth(orig_inode);
743 oext = orig_path[depth].p_ext;
747 ext4_ext_drop_refs(donor_path);
748 *err = get_ext_path(donor_inode, donor_off, &donor_path);
751 depth = ext_depth(donor_inode);
752 dext = donor_path[depth].p_ext;
755 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
756 donor_off, count - replaced_count);
763 ext4_ext_drop_refs(orig_path);
767 ext4_ext_drop_refs(donor_path);
771 return replaced_count;
775 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
777 * @inode1: the inode structure
778 * @inode2: the inode structure
780 * @page: result page vector
782 * Grab two locked pages for inode's by inode order
785 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
786 pgoff_t index, struct page *page[2])
788 struct address_space *mapping[2];
789 unsigned fl = AOP_FLAG_NOFS;
791 BUG_ON(!inode1 || !inode2);
792 if (inode1 < inode2) {
793 mapping[0] = inode1->i_mapping;
794 mapping[1] = inode2->i_mapping;
796 mapping[0] = inode2->i_mapping;
797 mapping[1] = inode1->i_mapping;
800 page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
804 page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
806 unlock_page(page[0]);
807 page_cache_release(page[0]);
811 * grab_cache_page_write_begin() may not wait on page's writeback if
812 * BDI not demand that. But it is reasonable to be very conservative
813 * here and explicitly wait on page's writeback
815 wait_on_page_writeback(page[0]);
816 wait_on_page_writeback(page[1]);
817 if (inode1 > inode2) {
826 /* Force page buffers uptodate w/o dropping page's lock */
828 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
830 struct inode *inode = page->mapping->host;
832 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
833 unsigned int blocksize, block_start, block_end;
834 int i, err, nr = 0, partial = 0;
835 BUG_ON(!PageLocked(page));
836 BUG_ON(PageWriteback(page));
838 if (PageUptodate(page))
841 blocksize = 1 << inode->i_blkbits;
842 if (!page_has_buffers(page))
843 create_empty_buffers(page, blocksize, 0);
845 head = page_buffers(page);
846 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
847 for (bh = head, block_start = 0; bh != head || !block_start;
848 block++, block_start = block_end, bh = bh->b_this_page) {
849 block_end = block_start + blocksize;
850 if (block_end <= from || block_start >= to) {
851 if (!buffer_uptodate(bh))
855 if (buffer_uptodate(bh))
857 if (!buffer_mapped(bh)) {
858 err = ext4_get_block(inode, block, bh, 0);
863 if (!buffer_mapped(bh)) {
864 zero_user(page, block_start, blocksize);
865 set_buffer_uptodate(bh);
869 BUG_ON(nr >= MAX_BUF_PER_PAGE);
876 for (i = 0; i < nr; i++) {
878 if (!bh_uptodate_or_lock(bh)) {
879 err = bh_submit_read(bh);
886 SetPageUptodate(page);
891 * move_extent_per_page - Move extent data per page
893 * @o_filp: file structure of original file
894 * @donor_inode: donor inode
895 * @orig_page_offset: page index on original file
896 * @data_offset_in_page: block index where data swapping starts
897 * @block_len_in_page: the number of blocks to be swapped
898 * @unwritten: orig extent is unwritten or not
899 * @err: pointer to save return value
901 * Save the data in original inode blocks and replace original inode extents
902 * with donor inode extents by calling mext_replace_branches().
903 * Finally, write out the saved data in new original inode blocks. Return
904 * replaced block count.
907 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
908 pgoff_t orig_page_offset, int data_offset_in_page,
909 int block_len_in_page, int unwritten, int *err)
911 struct inode *orig_inode = file_inode(o_filp);
912 struct page *pagep[2] = {NULL, NULL};
914 ext4_lblk_t orig_blk_offset;
915 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
916 unsigned int w_flags = 0;
917 unsigned int tmp_data_size, data_size, replaced_size;
918 int err2, jblocks, retries = 0;
919 int replaced_count = 0;
920 int from = data_offset_in_page << orig_inode->i_blkbits;
921 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
924 * It needs twice the amount of ordinary journal buffers because
925 * inode and donor_inode may change each different metadata blocks.
929 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
930 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
931 if (IS_ERR(handle)) {
932 *err = PTR_ERR(handle);
936 if (segment_eq(get_fs(), KERNEL_DS))
937 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
939 orig_blk_offset = orig_page_offset * blocks_per_page +
942 /* Calculate data_size */
943 if ((orig_blk_offset + block_len_in_page - 1) ==
944 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
945 /* Replace the last block */
946 tmp_data_size = orig_inode->i_size & (blocksize - 1);
948 * If data_size equal zero, it shows data_size is multiples of
949 * blocksize. So we set appropriate value.
951 if (tmp_data_size == 0)
952 tmp_data_size = blocksize;
954 data_size = tmp_data_size +
955 ((block_len_in_page - 1) << orig_inode->i_blkbits);
957 data_size = block_len_in_page << orig_inode->i_blkbits;
959 replaced_size = data_size;
961 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
963 if (unlikely(*err < 0))
966 * If orig extent was unwritten it can become initialized
967 * at any time after i_data_sem was dropped, in order to
968 * serialize with delalloc we have recheck extent while we
969 * hold page's lock, if it is still the case data copy is not
970 * necessary, just swap data blocks between orig and donor.
973 ext4_double_down_write_data_sem(orig_inode, donor_inode);
974 /* If any of extents in range became initialized we have to
975 * fallback to data copying */
976 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
977 block_len_in_page, 1, err);
981 unwritten &= mext_check_coverage(donor_inode, orig_blk_offset,
982 block_len_in_page, 1, err);
987 ext4_double_up_write_data_sem(orig_inode, donor_inode);
990 if ((page_has_private(pagep[0]) &&
991 !try_to_release_page(pagep[0], 0)) ||
992 (page_has_private(pagep[1]) &&
993 !try_to_release_page(pagep[1], 0))) {
997 replaced_count = mext_replace_branches(handle, orig_inode,
998 donor_inode, orig_blk_offset,
999 block_len_in_page, err);
1001 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1005 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1009 /* At this point all buffers in range are uptodate, old mapping layout
1010 * is no longer required, try to drop it now. */
1011 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1012 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1016 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1017 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1019 block_len_in_page, err);
1020 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1022 if (replaced_count) {
1023 block_len_in_page = replaced_count;
1025 block_len_in_page << orig_inode->i_blkbits;
1029 /* Perform all necessary steps similar write_begin()/write_end()
1030 * but keeping in mind that i_size will not change */
1031 *err = __block_write_begin(pagep[0], from, replaced_size,
1034 *err = block_commit_write(pagep[0], from, from + replaced_size);
1036 if (unlikely(*err < 0))
1037 goto repair_branches;
1039 /* Even in case of data=writeback it is reasonable to pin
1040 * inode to transaction, to prevent unexpected data loss */
1041 *err = ext4_jbd2_file_inode(handle, orig_inode);
1044 unlock_page(pagep[0]);
1045 page_cache_release(pagep[0]);
1046 unlock_page(pagep[1]);
1047 page_cache_release(pagep[1]);
1049 ext4_journal_stop(handle);
1050 /* Buffer was busy because probably is pinned to journal transaction,
1051 * force transaction commit may help to free it. */
1052 if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1055 return replaced_count;
1059 * This should never ever happen!
1060 * Extents are swapped already, but we are not able to copy data.
1061 * Try to swap extents to it's original places
1063 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1064 replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1066 block_len_in_page, &err2);
1067 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1068 if (replaced_count != block_len_in_page) {
1069 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1070 "Unable to copy data block,"
1071 " data will be lost.");
1079 * mext_check_arguments - Check whether move extent can be done
1081 * @orig_inode: original inode
1082 * @donor_inode: donor inode
1083 * @orig_start: logical start offset in block for orig
1084 * @donor_start: logical start offset in block for donor
1085 * @len: the number of blocks to be moved
1087 * Check the arguments of ext4_move_extents() whether the files can be
1088 * exchanged with each other.
1089 * Return 0 on success, or a negative error value on failure.
1092 mext_check_arguments(struct inode *orig_inode,
1093 struct inode *donor_inode, __u64 orig_start,
1094 __u64 donor_start, __u64 *len)
1096 ext4_lblk_t orig_blocks, donor_blocks;
1097 unsigned int blkbits = orig_inode->i_blkbits;
1098 unsigned int blocksize = 1 << blkbits;
1100 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1101 ext4_debug("ext4 move extent: suid or sgid is set"
1102 " to donor file [ino:orig %lu, donor %lu]\n",
1103 orig_inode->i_ino, donor_inode->i_ino);
1107 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1110 /* Ext4 move extent does not support swapfile */
1111 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1112 ext4_debug("ext4 move extent: The argument files should "
1113 "not be swapfile [ino:orig %lu, donor %lu]\n",
1114 orig_inode->i_ino, donor_inode->i_ino);
1118 /* Ext4 move extent supports only extent based file */
1119 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1120 ext4_debug("ext4 move extent: orig file is not extents "
1121 "based file [ino:orig %lu]\n", orig_inode->i_ino);
1123 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1124 ext4_debug("ext4 move extent: donor file is not extents "
1125 "based file [ino:donor %lu]\n", donor_inode->i_ino);
1129 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1130 ext4_debug("ext4 move extent: File size is 0 byte\n");
1134 /* Start offset should be same */
1135 if (orig_start != donor_start) {
1136 ext4_debug("ext4 move extent: orig and donor's start "
1137 "offset are not same [ino:orig %lu, donor %lu]\n",
1138 orig_inode->i_ino, donor_inode->i_ino);
1142 if ((orig_start >= EXT_MAX_BLOCKS) ||
1143 (*len > EXT_MAX_BLOCKS) ||
1144 (orig_start + *len >= EXT_MAX_BLOCKS)) {
1145 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1146 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1147 orig_inode->i_ino, donor_inode->i_ino);
1151 if (orig_inode->i_size > donor_inode->i_size) {
1152 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1153 /* TODO: eliminate this artificial restriction */
1154 if (orig_start >= donor_blocks) {
1155 ext4_debug("ext4 move extent: orig start offset "
1156 "[%llu] should be less than donor file blocks "
1157 "[%u] [ino:orig %lu, donor %lu]\n",
1158 orig_start, donor_blocks,
1159 orig_inode->i_ino, donor_inode->i_ino);
1163 /* TODO: eliminate this artificial restriction */
1164 if (orig_start + *len > donor_blocks) {
1165 ext4_debug("ext4 move extent: End offset [%llu] should "
1166 "be less than donor file blocks [%u]."
1167 "So adjust length from %llu to %llu "
1168 "[ino:orig %lu, donor %lu]\n",
1169 orig_start + *len, donor_blocks,
1170 *len, donor_blocks - orig_start,
1171 orig_inode->i_ino, donor_inode->i_ino);
1172 *len = donor_blocks - orig_start;
1175 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1176 if (orig_start >= orig_blocks) {
1177 ext4_debug("ext4 move extent: start offset [%llu] "
1178 "should be less than original file blocks "
1179 "[%u] [ino:orig %lu, donor %lu]\n",
1180 orig_start, orig_blocks,
1181 orig_inode->i_ino, donor_inode->i_ino);
1185 if (orig_start + *len > orig_blocks) {
1186 ext4_debug("ext4 move extent: Adjust length "
1187 "from %llu to %llu. Because it should be "
1188 "less than original file blocks "
1189 "[ino:orig %lu, donor %lu]\n",
1190 *len, orig_blocks - orig_start,
1191 orig_inode->i_ino, donor_inode->i_ino);
1192 *len = orig_blocks - orig_start;
1197 ext4_debug("ext4 move extent: len should not be 0 "
1198 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1199 donor_inode->i_ino);
1207 * ext4_move_extents - Exchange the specified range of a file
1209 * @o_filp: file structure of the original file
1210 * @d_filp: file structure of the donor file
1211 * @orig_start: start offset in block for orig
1212 * @donor_start: start offset in block for donor
1213 * @len: the number of blocks to be moved
1214 * @moved_len: moved block length
1216 * This function returns 0 and moved block length is set in moved_len
1217 * if succeed, otherwise returns error value.
1219 * Note: ext4_move_extents() proceeds the following order.
1220 * 1:ext4_move_extents() calculates the last block number of moving extent
1221 * function by the start block number (orig_start) and the number of blocks
1222 * to be moved (len) specified as arguments.
1223 * If the {orig, donor}_start points a hole, the extent's start offset
1224 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1225 * after hole behind.
1226 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1227 * or the ext_cur exceeds the block_end which is last logical block number.
1228 * 3:To get the length of continues area, call mext_next_extent()
1229 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1230 * until find un-continuous extent, the start logical block number exceeds
1231 * the block_end or the extent points to the last extent.
1232 * 4:Exchange the original inode data with donor inode data
1233 * from orig_page_offset to seq_end_page.
1234 * The start indexes of data are specified as arguments.
1235 * That of the original inode is orig_page_offset,
1236 * and the donor inode is also orig_page_offset
1237 * (To easily handle blocksize != pagesize case, the offset for the
1238 * donor inode is block unit).
1239 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1240 * then returns to step 2.
1241 * 6:Release holecheck_path, orig_path and set the len to moved_len
1242 * which shows the number of moved blocks.
1243 * The moved_len is useful for the command to calculate the file offset
1244 * for starting next move extent ioctl.
1245 * 7:Return 0 on success, or a negative error value on failure.
1248 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1249 __u64 orig_start, __u64 donor_start, __u64 len,
1252 struct inode *orig_inode = file_inode(o_filp);
1253 struct inode *donor_inode = file_inode(d_filp);
1254 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1255 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1256 ext4_lblk_t block_start = orig_start;
1257 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1258 ext4_lblk_t rest_blocks;
1259 pgoff_t orig_page_offset = 0, seq_end_page;
1260 int ret, depth, last_extent = 0;
1261 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1262 int data_offset_in_page;
1263 int block_len_in_page;
1266 if (orig_inode->i_sb != donor_inode->i_sb) {
1267 ext4_debug("ext4 move extent: The argument files "
1268 "should be in same FS [ino:orig %lu, donor %lu]\n",
1269 orig_inode->i_ino, donor_inode->i_ino);
1273 /* orig and donor should be different inodes */
1274 if (orig_inode == donor_inode) {
1275 ext4_debug("ext4 move extent: The argument files should not "
1276 "be same inode [ino:orig %lu, donor %lu]\n",
1277 orig_inode->i_ino, donor_inode->i_ino);
1281 /* Regular file check */
1282 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1283 ext4_debug("ext4 move extent: The argument files should be "
1284 "regular file [ino:orig %lu, donor %lu]\n",
1285 orig_inode->i_ino, donor_inode->i_ino);
1288 /* TODO: This is non obvious task to swap blocks for inodes with full
1289 jornaling enabled */
1290 if (ext4_should_journal_data(orig_inode) ||
1291 ext4_should_journal_data(donor_inode)) {
1294 /* Protect orig and donor inodes against a truncate */
1295 lock_two_nondirectories(orig_inode, donor_inode);
1297 /* Wait for all existing dio workers */
1298 ext4_inode_block_unlocked_dio(orig_inode);
1299 ext4_inode_block_unlocked_dio(donor_inode);
1300 inode_dio_wait(orig_inode);
1301 inode_dio_wait(donor_inode);
1303 /* Protect extent tree against block allocations via delalloc */
1304 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1305 /* Check the filesystem environment whether move_extent can be done */
1306 ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1311 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1312 block_end = block_start + len - 1;
1313 if (file_end < block_end)
1314 len -= block_end - file_end;
1316 ret = get_ext_path(orig_inode, block_start, &orig_path);
1320 /* Get path structure to check the hole */
1321 ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1325 depth = ext_depth(orig_inode);
1326 ext_cur = holecheck_path[depth].p_ext;
1329 * Get proper starting location of block replacement if block_start was
1332 if (le32_to_cpu(ext_cur->ee_block) +
1333 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1335 * The hole exists between extents or the tail of
1338 last_extent = mext_next_extent(orig_inode,
1339 holecheck_path, &ext_cur);
1340 if (last_extent < 0) {
1344 last_extent = mext_next_extent(orig_inode, orig_path,
1346 if (last_extent < 0) {
1350 seq_start = le32_to_cpu(ext_cur->ee_block);
1351 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1352 /* The hole exists at the beginning of original file. */
1353 seq_start = le32_to_cpu(ext_cur->ee_block);
1355 seq_start = block_start;
1357 /* No blocks within the specified range. */
1358 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1359 ext4_debug("ext4 move extent: The specified range of file "
1360 "may be the hole\n");
1365 /* Adjust start blocks */
1366 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1367 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1368 max(le32_to_cpu(ext_cur->ee_block), block_start);
1370 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1371 seq_blocks += add_blocks;
1373 /* Adjust tail blocks */
1374 if (seq_start + seq_blocks - 1 > block_end)
1375 seq_blocks = block_end - seq_start + 1;
1378 last_extent = mext_next_extent(orig_inode, holecheck_path,
1380 if (last_extent < 0) {
1384 add_blocks = ext4_ext_get_actual_len(ext_cur);
1387 * Extend the length of contiguous block (seq_blocks)
1388 * if extents are contiguous.
1390 if (ext4_can_extents_be_merged(orig_inode,
1391 ext_prev, ext_cur) &&
1392 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1396 /* Is original extent is unwritten */
1397 unwritten = ext4_ext_is_unwritten(ext_prev);
1399 data_offset_in_page = seq_start % blocks_per_page;
1402 * Calculate data blocks count that should be swapped
1403 * at the first page.
1405 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1406 /* Swapped blocks are across pages */
1408 blocks_per_page - data_offset_in_page;
1410 /* Swapped blocks are in a page */
1411 block_len_in_page = seq_blocks;
1414 orig_page_offset = seq_start >>
1415 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1416 seq_end_page = (seq_start + seq_blocks - 1) >>
1417 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1418 seq_start = le32_to_cpu(ext_cur->ee_block);
1419 rest_blocks = seq_blocks;
1422 * Up semaphore to avoid following problems:
1423 * a. transaction deadlock among ext4_journal_start,
1424 * ->write_begin via pagefault, and jbd2_journal_commit
1425 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1426 * in move_extent_per_page
1428 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1430 while (orig_page_offset <= seq_end_page) {
1432 /* Swap original branches with new branches */
1433 block_len_in_page = move_extent_per_page(
1434 o_filp, donor_inode,
1436 data_offset_in_page,
1440 /* Count how many blocks we have exchanged */
1441 *moved_len += block_len_in_page;
1444 if (*moved_len > len) {
1445 EXT4_ERROR_INODE(orig_inode,
1446 "We replaced blocks too much! "
1447 "sum of replaced: %llu requested: %llu",
1454 data_offset_in_page = 0;
1455 rest_blocks -= block_len_in_page;
1456 if (rest_blocks > blocks_per_page)
1457 block_len_in_page = blocks_per_page;
1459 block_len_in_page = rest_blocks;
1462 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1466 /* Decrease buffer counter */
1468 ext4_ext_drop_refs(holecheck_path);
1469 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1472 depth = holecheck_path->p_depth;
1474 /* Decrease buffer counter */
1476 ext4_ext_drop_refs(orig_path);
1477 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1481 ext_cur = holecheck_path[depth].p_ext;
1482 add_blocks = ext4_ext_get_actual_len(ext_cur);
1488 ext4_discard_preallocations(orig_inode);
1489 ext4_discard_preallocations(donor_inode);
1493 ext4_ext_drop_refs(orig_path);
1496 if (holecheck_path) {
1497 ext4_ext_drop_refs(holecheck_path);
1498 kfree(holecheck_path);
1500 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1501 ext4_inode_resume_unlocked_dio(orig_inode);
1502 ext4_inode_resume_unlocked_dio(donor_inode);
1503 unlock_two_nondirectories(orig_inode, donor_inode);