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 * ext4_double_down_write_data_sem - Acquire two inodes' write lock
55 * Acquire write lock of i_data_sem of the two inodes
58 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
61 down_write(&EXT4_I(first)->i_data_sem);
62 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
64 down_write(&EXT4_I(second)->i_data_sem);
65 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
71 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
73 * @orig_inode: original inode structure to be released its lock first
74 * @donor_inode: donor inode structure to be released its lock second
75 * Release write lock of i_data_sem of two inodes (orig and donor).
78 ext4_double_up_write_data_sem(struct inode *orig_inode,
79 struct inode *donor_inode)
81 up_write(&EXT4_I(orig_inode)->i_data_sem);
82 up_write(&EXT4_I(donor_inode)->i_data_sem);
86 * mext_check_coverage - Check that all extents in range has the same type
88 * @inode: inode in question
89 * @from: block offset of inode
90 * @count: block count to be checked
91 * @unwritten: extents expected to be unwritten
92 * @err: pointer to save error value
94 * Return 1 if all extents in range has expected type, and zero otherwise.
97 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
98 int unwritten, int *err)
100 struct ext4_ext_path *path = NULL;
101 struct ext4_extent *ext;
103 ext4_lblk_t last = from + count;
104 while (from < last) {
105 *err = get_ext_path(inode, from, &path);
108 ext = path[ext_depth(inode)].p_ext;
109 if (unwritten != ext4_ext_is_unwritten(ext))
111 from += ext4_ext_get_actual_len(ext);
112 ext4_ext_drop_refs(path);
117 ext4_ext_drop_refs(path);
124 * mext_replace_branches - Replace original extents with new extents
126 * @handle: journal handle
127 * @orig_inode: original inode
128 * @donor_inode: donor inode
129 * @from: block offset of orig_inode
130 * @count: block count to be replaced
131 * @err: pointer to save return value
133 * Replace original inode extents and donor inode extents page by page.
134 * We implement this replacement in the following three steps:
135 * 1. Save the block information of original and donor inodes into
137 * 2. Change the block information of original inode to point at the
138 * donor inode blocks.
139 * 3. Change the block information of donor inode to point at the saved
140 * original inode blocks in the dummy extents.
142 * Return replaced block count.
146 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
148 * @inode1: the inode structure
149 * @inode2: the inode structure
151 * @page: result page vector
153 * Grab two locked pages for inode's by inode order
156 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
157 pgoff_t index1, pgoff_t index2, struct page *page[2])
159 struct address_space *mapping[2];
160 unsigned fl = AOP_FLAG_NOFS;
162 BUG_ON(!inode1 || !inode2);
163 if (inode1 < inode2) {
164 mapping[0] = inode1->i_mapping;
165 mapping[1] = inode2->i_mapping;
167 pgoff_t tmp = index1;
170 mapping[0] = inode2->i_mapping;
171 mapping[1] = inode1->i_mapping;
174 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
178 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
180 unlock_page(page[0]);
181 page_cache_release(page[0]);
185 * grab_cache_page_write_begin() may not wait on page's writeback if
186 * BDI not demand that. But it is reasonable to be very conservative
187 * here and explicitly wait on page's writeback
189 wait_on_page_writeback(page[0]);
190 wait_on_page_writeback(page[1]);
191 if (inode1 > inode2) {
200 /* Force page buffers uptodate w/o dropping page's lock */
202 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
204 struct inode *inode = page->mapping->host;
206 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
207 unsigned int blocksize, block_start, block_end;
208 int i, err, nr = 0, partial = 0;
209 BUG_ON(!PageLocked(page));
210 BUG_ON(PageWriteback(page));
212 if (PageUptodate(page))
215 blocksize = 1 << inode->i_blkbits;
216 if (!page_has_buffers(page))
217 create_empty_buffers(page, blocksize, 0);
219 head = page_buffers(page);
220 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
221 for (bh = head, block_start = 0; bh != head || !block_start;
222 block++, block_start = block_end, bh = bh->b_this_page) {
223 block_end = block_start + blocksize;
224 if (block_end <= from || block_start >= to) {
225 if (!buffer_uptodate(bh))
229 if (buffer_uptodate(bh))
231 if (!buffer_mapped(bh)) {
232 err = ext4_get_block(inode, block, bh, 0);
237 if (!buffer_mapped(bh)) {
238 zero_user(page, block_start, blocksize);
239 set_buffer_uptodate(bh);
243 BUG_ON(nr >= MAX_BUF_PER_PAGE);
250 for (i = 0; i < nr; i++) {
252 if (!bh_uptodate_or_lock(bh)) {
253 err = bh_submit_read(bh);
260 SetPageUptodate(page);
265 * move_extent_per_page - Move extent data per page
267 * @o_filp: file structure of original file
268 * @donor_inode: donor inode
269 * @orig_page_offset: page index on original file
270 * @data_offset_in_page: block index where data swapping starts
271 * @block_len_in_page: the number of blocks to be swapped
272 * @unwritten: orig extent is unwritten or not
273 * @err: pointer to save return value
275 * Save the data in original inode blocks and replace original inode extents
276 * with donor inode extents by calling mext_replace_branches().
277 * Finally, write out the saved data in new original inode blocks. Return
278 * replaced block count.
281 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
282 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
283 int data_offset_in_page,
284 int block_len_in_page, int unwritten, int *err)
286 struct inode *orig_inode = file_inode(o_filp);
287 struct page *pagep[2] = {NULL, NULL};
289 ext4_lblk_t orig_blk_offset, donor_blk_offset;
290 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
291 unsigned int w_flags = 0;
292 unsigned int tmp_data_size, data_size, replaced_size;
293 int err2, jblocks, retries = 0;
294 int replaced_count = 0;
295 int from = data_offset_in_page << orig_inode->i_blkbits;
296 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
299 * It needs twice the amount of ordinary journal buffers because
300 * inode and donor_inode may change each different metadata blocks.
304 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
305 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
306 if (IS_ERR(handle)) {
307 *err = PTR_ERR(handle);
311 if (segment_eq(get_fs(), KERNEL_DS))
312 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
314 orig_blk_offset = orig_page_offset * blocks_per_page +
317 donor_blk_offset = donor_page_offset * blocks_per_page +
320 /* Calculate data_size */
321 if ((orig_blk_offset + block_len_in_page - 1) ==
322 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
323 /* Replace the last block */
324 tmp_data_size = orig_inode->i_size & (blocksize - 1);
326 * If data_size equal zero, it shows data_size is multiples of
327 * blocksize. So we set appropriate value.
329 if (tmp_data_size == 0)
330 tmp_data_size = blocksize;
332 data_size = tmp_data_size +
333 ((block_len_in_page - 1) << orig_inode->i_blkbits);
335 data_size = block_len_in_page << orig_inode->i_blkbits;
337 replaced_size = data_size;
339 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
340 donor_page_offset, pagep);
341 if (unlikely(*err < 0))
344 * If orig extent was unwritten it can become initialized
345 * at any time after i_data_sem was dropped, in order to
346 * serialize with delalloc we have recheck extent while we
347 * hold page's lock, if it is still the case data copy is not
348 * necessary, just swap data blocks between orig and donor.
351 ext4_double_down_write_data_sem(orig_inode, donor_inode);
352 /* If any of extents in range became initialized we have to
353 * fallback to data copying */
354 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
355 block_len_in_page, 1, err);
359 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
360 block_len_in_page, 1, err);
365 ext4_double_up_write_data_sem(orig_inode, donor_inode);
368 if ((page_has_private(pagep[0]) &&
369 !try_to_release_page(pagep[0], 0)) ||
370 (page_has_private(pagep[1]) &&
371 !try_to_release_page(pagep[1], 0))) {
375 replaced_count = ext4_swap_extents(handle, orig_inode,
376 donor_inode, orig_blk_offset,
378 block_len_in_page, 1, err);
380 ext4_double_up_write_data_sem(orig_inode, donor_inode);
384 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
388 /* At this point all buffers in range are uptodate, old mapping layout
389 * is no longer required, try to drop it now. */
390 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
391 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
395 ext4_double_down_write_data_sem(orig_inode, donor_inode);
396 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
397 orig_blk_offset, donor_blk_offset,
398 block_len_in_page, 1, err);
399 ext4_double_up_write_data_sem(orig_inode, donor_inode);
401 if (replaced_count) {
402 block_len_in_page = replaced_count;
404 block_len_in_page << orig_inode->i_blkbits;
408 /* Perform all necessary steps similar write_begin()/write_end()
409 * but keeping in mind that i_size will not change */
410 *err = __block_write_begin(pagep[0], from, replaced_size,
413 *err = block_commit_write(pagep[0], from, from + replaced_size);
415 if (unlikely(*err < 0))
416 goto repair_branches;
418 /* Even in case of data=writeback it is reasonable to pin
419 * inode to transaction, to prevent unexpected data loss */
420 *err = ext4_jbd2_file_inode(handle, orig_inode);
423 unlock_page(pagep[0]);
424 page_cache_release(pagep[0]);
425 unlock_page(pagep[1]);
426 page_cache_release(pagep[1]);
428 ext4_journal_stop(handle);
429 /* Buffer was busy because probably is pinned to journal transaction,
430 * force transaction commit may help to free it. */
431 if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
434 return replaced_count;
438 * This should never ever happen!
439 * Extents are swapped already, but we are not able to copy data.
440 * Try to swap extents to it's original places
442 ext4_double_down_write_data_sem(orig_inode, donor_inode);
443 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
444 orig_blk_offset, donor_blk_offset,
445 block_len_in_page, 0, &err2);
446 ext4_double_up_write_data_sem(orig_inode, donor_inode);
447 if (replaced_count != block_len_in_page) {
448 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
449 "Unable to copy data block,"
450 " data will be lost.");
458 * mext_check_arguments - Check whether move extent can be done
460 * @orig_inode: original inode
461 * @donor_inode: donor inode
462 * @orig_start: logical start offset in block for orig
463 * @donor_start: logical start offset in block for donor
464 * @len: the number of blocks to be moved
466 * Check the arguments of ext4_move_extents() whether the files can be
467 * exchanged with each other.
468 * Return 0 on success, or a negative error value on failure.
471 mext_check_arguments(struct inode *orig_inode,
472 struct inode *donor_inode, __u64 orig_start,
473 __u64 donor_start, __u64 *len)
475 __u64 orig_eof, donor_eof;
476 unsigned int blkbits = orig_inode->i_blkbits;
477 unsigned int blocksize = 1 << blkbits;
479 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
480 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
483 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
484 ext4_debug("ext4 move extent: suid or sgid is set"
485 " to donor file [ino:orig %lu, donor %lu]\n",
486 orig_inode->i_ino, donor_inode->i_ino);
490 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
493 /* Ext4 move extent does not support swapfile */
494 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
495 ext4_debug("ext4 move extent: The argument files should "
496 "not be swapfile [ino:orig %lu, donor %lu]\n",
497 orig_inode->i_ino, donor_inode->i_ino);
501 /* Ext4 move extent supports only extent based file */
502 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
503 ext4_debug("ext4 move extent: orig file is not extents "
504 "based file [ino:orig %lu]\n", orig_inode->i_ino);
506 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
507 ext4_debug("ext4 move extent: donor file is not extents "
508 "based file [ino:donor %lu]\n", donor_inode->i_ino);
512 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
513 ext4_debug("ext4 move extent: File size is 0 byte\n");
517 /* Start offset should be same */
518 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
519 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
520 ext4_debug("ext4 move extent: orig and donor's start "
521 "offset are not alligned [ino:orig %lu, donor %lu]\n",
522 orig_inode->i_ino, donor_inode->i_ino);
526 if ((orig_start >= EXT_MAX_BLOCKS) ||
527 (donor_start >= EXT_MAX_BLOCKS) ||
528 (*len > EXT_MAX_BLOCKS) ||
529 (donor_start + *len >= EXT_MAX_BLOCKS) ||
530 (orig_start + *len >= EXT_MAX_BLOCKS)) {
531 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
532 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
533 orig_inode->i_ino, donor_inode->i_ino);
536 if (orig_eof < orig_start + *len - 1)
537 *len = orig_eof - orig_start;
538 if (donor_eof < donor_start + *len - 1)
539 *len = donor_eof - donor_start;
541 ext4_debug("ext4 move extent: len should not be 0 "
542 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
551 * ext4_move_extents - Exchange the specified range of a file
553 * @o_filp: file structure of the original file
554 * @d_filp: file structure of the donor file
555 * @orig_start: start offset in block for orig
556 * @donor_start: start offset in block for donor
557 * @len: the number of blocks to be moved
558 * @moved_len: moved block length
560 * This function returns 0 and moved block length is set in moved_len
561 * if succeed, otherwise returns error value.
563 * Note: ext4_move_extents() proceeds the following order.
564 * 1:ext4_move_extents() calculates the last block number of moving extent
565 * function by the start block number (orig_start) and the number of blocks
566 * to be moved (len) specified as arguments.
567 * If the {orig, donor}_start points a hole, the extent's start offset
568 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
570 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
571 * or the ext_cur exceeds the block_end which is last logical block number.
572 * 3:To get the length of continues area, call mext_next_extent()
573 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
574 * until find un-continuous extent, the start logical block number exceeds
575 * the block_end or the extent points to the last extent.
576 * 4:Exchange the original inode data with donor inode data
577 * from orig_page_offset to seq_end_page.
578 * The start indexes of data are specified as arguments.
579 * That of the original inode is orig_page_offset,
580 * and the donor inode is also orig_page_offset
581 * (To easily handle blocksize != pagesize case, the offset for the
582 * donor inode is block unit).
583 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
584 * then returns to step 2.
585 * 6:Release holecheck_path, orig_path and set the len to moved_len
586 * which shows the number of moved blocks.
587 * The moved_len is useful for the command to calculate the file offset
588 * for starting next move extent ioctl.
589 * 7:Return 0 on success, or a negative error value on failure.
592 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
593 __u64 donor_blk, __u64 len, __u64 *moved_len)
595 struct inode *orig_inode = file_inode(o_filp);
596 struct inode *donor_inode = file_inode(d_filp);
597 struct ext4_ext_path *path = NULL;
598 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
599 ext4_lblk_t o_end, o_start = orig_blk;
600 ext4_lblk_t d_start = donor_blk;
603 if (orig_inode->i_sb != donor_inode->i_sb) {
604 ext4_debug("ext4 move extent: The argument files "
605 "should be in same FS [ino:orig %lu, donor %lu]\n",
606 orig_inode->i_ino, donor_inode->i_ino);
610 /* orig and donor should be different inodes */
611 if (orig_inode == donor_inode) {
612 ext4_debug("ext4 move extent: The argument files should not "
613 "be same inode [ino:orig %lu, donor %lu]\n",
614 orig_inode->i_ino, donor_inode->i_ino);
618 /* Regular file check */
619 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
620 ext4_debug("ext4 move extent: The argument files should be "
621 "regular file [ino:orig %lu, donor %lu]\n",
622 orig_inode->i_ino, donor_inode->i_ino);
625 /* TODO: This is non obvious task to swap blocks for inodes with full
627 if (ext4_should_journal_data(orig_inode) ||
628 ext4_should_journal_data(donor_inode)) {
631 /* Protect orig and donor inodes against a truncate */
632 lock_two_nondirectories(orig_inode, donor_inode);
634 /* Wait for all existing dio workers */
635 ext4_inode_block_unlocked_dio(orig_inode);
636 ext4_inode_block_unlocked_dio(donor_inode);
637 inode_dio_wait(orig_inode);
638 inode_dio_wait(donor_inode);
640 /* Protect extent tree against block allocations via delalloc */
641 ext4_double_down_write_data_sem(orig_inode, donor_inode);
642 /* Check the filesystem environment whether move_extent can be done */
643 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
647 o_end = o_start + len;
649 while (o_start < o_end) {
650 struct ext4_extent *ex;
651 ext4_lblk_t cur_blk, next_blk;
652 pgoff_t orig_page_index, donor_page_index;
654 int unwritten, cur_len;
656 ret = get_ext_path(orig_inode, o_start, &path);
659 ex = path[path->p_depth].p_ext;
660 next_blk = ext4_ext_next_allocated_block(path);
661 cur_blk = le32_to_cpu(ex->ee_block);
662 cur_len = ext4_ext_get_actual_len(ex);
663 /* Check hole before the start pos */
664 if (cur_blk + cur_len - 1 < o_start) {
665 if (next_blk == EXT_MAX_BLOCKS) {
670 d_start += next_blk - o_start;
673 /* Check hole after the start pos */
674 } else if (cur_blk > o_start) {
676 d_start += cur_blk - o_start;
678 /* Extent inside requested range ?*/
679 if (cur_blk >= o_end)
681 } else { /* in_range(o_start, o_blk, o_len) */
682 cur_len += cur_blk - o_start;
684 unwritten = ext4_ext_is_unwritten(ex);
685 if (o_end - o_start < cur_len)
686 cur_len = o_end - o_start;
688 orig_page_index = o_start >> (PAGE_CACHE_SHIFT -
689 orig_inode->i_blkbits);
690 donor_page_index = d_start >> (PAGE_CACHE_SHIFT -
691 donor_inode->i_blkbits);
692 offset_in_page = o_start % blocks_per_page;
693 if (cur_len > blocks_per_page- offset_in_page)
694 cur_len = blocks_per_page - offset_in_page;
696 * Up semaphore to avoid following problems:
697 * a. transaction deadlock among ext4_journal_start,
698 * ->write_begin via pagefault, and jbd2_journal_commit
699 * b. racing with ->readpage, ->write_begin, and ext4_get_block
700 * in move_extent_per_page
702 ext4_double_up_write_data_sem(orig_inode, donor_inode);
703 /* Swap original branches with new branches */
704 move_extent_per_page(o_filp, donor_inode,
705 orig_page_index, donor_page_index,
706 offset_in_page, cur_len,
708 ext4_double_down_write_data_sem(orig_inode, donor_inode);
715 ext4_ext_drop_refs(path);
720 *moved_len = o_start - orig_blk;
721 if (*moved_len > len)
726 ext4_discard_preallocations(orig_inode);
727 ext4_discard_preallocations(donor_inode);
731 ext4_ext_drop_refs(path);
734 ext4_double_up_write_data_sem(orig_inode, donor_inode);
735 ext4_inode_resume_unlocked_dio(orig_inode);
736 ext4_inode_resume_unlocked_dio(donor_inode);
737 unlock_two_nondirectories(orig_inode, donor_inode);