1 // SPDX-License-Identifier: LGPL-2.1
3 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
4 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
5 * Akira Fujita <a-fujita@rs.jp.nec.com>
9 #include <linux/quotaops.h>
10 #include <linux/slab.h>
11 #include "ext4_jbd2.h"
13 #include "ext4_extents.h"
16 * get_ext_path() - Find an extent path for designated logical block number.
17 * @inode: inode to be searched
18 * @lblock: logical block number to find an extent path
19 * @ppath: pointer to an extent path pointer (for output)
21 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
25 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
26 struct ext4_ext_path **ppath)
28 struct ext4_ext_path *path;
30 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
33 if (path[ext_depth(inode)].p_ext == NULL) {
34 ext4_ext_drop_refs(path);
44 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
45 * @first: inode to be locked
46 * @second: inode to be locked
48 * Acquire write lock of i_data_sem of the two inodes
51 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
54 down_write(&EXT4_I(first)->i_data_sem);
55 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
57 down_write(&EXT4_I(second)->i_data_sem);
58 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
64 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
66 * @orig_inode: original inode structure to be released its lock first
67 * @donor_inode: donor inode structure to be released its lock second
68 * Release write lock of i_data_sem of two inodes (orig and donor).
71 ext4_double_up_write_data_sem(struct inode *orig_inode,
72 struct inode *donor_inode)
74 up_write(&EXT4_I(orig_inode)->i_data_sem);
75 up_write(&EXT4_I(donor_inode)->i_data_sem);
79 * mext_check_coverage - Check that all extents in range has the same type
81 * @inode: inode in question
82 * @from: block offset of inode
83 * @count: block count to be checked
84 * @unwritten: extents expected to be unwritten
85 * @err: pointer to save error value
87 * Return 1 if all extents in range has expected type, and zero otherwise.
90 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
91 int unwritten, int *err)
93 struct ext4_ext_path *path = NULL;
94 struct ext4_extent *ext;
96 ext4_lblk_t last = from + count;
98 *err = get_ext_path(inode, from, &path);
101 ext = path[ext_depth(inode)].p_ext;
102 if (unwritten != ext4_ext_is_unwritten(ext))
104 from += ext4_ext_get_actual_len(ext);
105 ext4_ext_drop_refs(path);
109 ext4_ext_drop_refs(path);
115 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
117 * @inode1: the inode structure
118 * @inode2: the inode structure
119 * @index1: page index
120 * @index2: page index
121 * @page: result page vector
123 * Grab two locked pages for inode's by inode order
126 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
127 pgoff_t index1, pgoff_t index2, struct page *page[2])
129 struct address_space *mapping[2];
130 unsigned fl = AOP_FLAG_NOFS;
132 BUG_ON(!inode1 || !inode2);
133 if (inode1 < inode2) {
134 mapping[0] = inode1->i_mapping;
135 mapping[1] = inode2->i_mapping;
137 swap(index1, index2);
138 mapping[0] = inode2->i_mapping;
139 mapping[1] = inode1->i_mapping;
142 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
146 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
148 unlock_page(page[0]);
153 * grab_cache_page_write_begin() may not wait on page's writeback if
154 * BDI not demand that. But it is reasonable to be very conservative
155 * here and explicitly wait on page's writeback
157 wait_on_page_writeback(page[0]);
158 wait_on_page_writeback(page[1]);
160 swap(page[0], page[1]);
165 /* Force page buffers uptodate w/o dropping page's lock */
167 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
169 struct inode *inode = page->mapping->host;
171 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
172 unsigned int blocksize, block_start, block_end;
173 int i, err, nr = 0, partial = 0;
174 BUG_ON(!PageLocked(page));
175 BUG_ON(PageWriteback(page));
177 if (PageUptodate(page))
180 blocksize = i_blocksize(inode);
181 if (!page_has_buffers(page))
182 create_empty_buffers(page, blocksize, 0);
184 head = page_buffers(page);
185 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
186 for (bh = head, block_start = 0; bh != head || !block_start;
187 block++, block_start = block_end, bh = bh->b_this_page) {
188 block_end = block_start + blocksize;
189 if (block_end <= from || block_start >= to) {
190 if (!buffer_uptodate(bh))
194 if (buffer_uptodate(bh))
196 if (!buffer_mapped(bh)) {
197 err = ext4_get_block(inode, block, bh, 0);
202 if (!buffer_mapped(bh)) {
203 zero_user(page, block_start, blocksize);
204 set_buffer_uptodate(bh);
208 BUG_ON(nr >= MAX_BUF_PER_PAGE);
215 for (i = 0; i < nr; i++) {
217 if (!bh_uptodate_or_lock(bh)) {
218 err = ext4_read_bh(bh, 0, NULL);
225 SetPageUptodate(page);
230 * move_extent_per_page - Move extent data per page
232 * @o_filp: file structure of original file
233 * @donor_inode: donor inode
234 * @orig_page_offset: page index on original file
235 * @donor_page_offset: page index on donor file
236 * @data_offset_in_page: block index where data swapping starts
237 * @block_len_in_page: the number of blocks to be swapped
238 * @unwritten: orig extent is unwritten or not
239 * @err: pointer to save return value
241 * Save the data in original inode blocks and replace original inode extents
242 * with donor inode extents by calling ext4_swap_extents().
243 * Finally, write out the saved data in new original inode blocks. Return
244 * replaced block count.
247 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
248 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
249 int data_offset_in_page,
250 int block_len_in_page, int unwritten, int *err)
252 struct inode *orig_inode = file_inode(o_filp);
253 struct page *pagep[2] = {NULL, NULL};
255 ext4_lblk_t orig_blk_offset, donor_blk_offset;
256 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
257 unsigned int tmp_data_size, data_size, replaced_size;
258 int i, err2, jblocks, retries = 0;
259 int replaced_count = 0;
260 int from = data_offset_in_page << orig_inode->i_blkbits;
261 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
262 struct super_block *sb = orig_inode->i_sb;
263 struct buffer_head *bh = NULL;
266 * It needs twice the amount of ordinary journal buffers because
267 * inode and donor_inode may change each different metadata blocks.
271 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
272 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
273 if (IS_ERR(handle)) {
274 *err = PTR_ERR(handle);
278 orig_blk_offset = orig_page_offset * blocks_per_page +
281 donor_blk_offset = donor_page_offset * blocks_per_page +
284 /* Calculate data_size */
285 if ((orig_blk_offset + block_len_in_page - 1) ==
286 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
287 /* Replace the last block */
288 tmp_data_size = orig_inode->i_size & (blocksize - 1);
290 * If data_size equal zero, it shows data_size is multiples of
291 * blocksize. So we set appropriate value.
293 if (tmp_data_size == 0)
294 tmp_data_size = blocksize;
296 data_size = tmp_data_size +
297 ((block_len_in_page - 1) << orig_inode->i_blkbits);
299 data_size = block_len_in_page << orig_inode->i_blkbits;
301 replaced_size = data_size;
303 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
304 donor_page_offset, pagep);
305 if (unlikely(*err < 0))
308 * If orig extent was unwritten it can become initialized
309 * at any time after i_data_sem was dropped, in order to
310 * serialize with delalloc we have recheck extent while we
311 * hold page's lock, if it is still the case data copy is not
312 * necessary, just swap data blocks between orig and donor.
315 ext4_double_down_write_data_sem(orig_inode, donor_inode);
316 /* If any of extents in range became initialized we have to
317 * fallback to data copying */
318 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
319 block_len_in_page, 1, err);
323 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
324 block_len_in_page, 1, err);
329 ext4_double_up_write_data_sem(orig_inode, donor_inode);
332 if ((page_has_private(pagep[0]) &&
333 !try_to_release_page(pagep[0], 0)) ||
334 (page_has_private(pagep[1]) &&
335 !try_to_release_page(pagep[1], 0))) {
339 replaced_count = ext4_swap_extents(handle, orig_inode,
340 donor_inode, orig_blk_offset,
342 block_len_in_page, 1, err);
344 ext4_double_up_write_data_sem(orig_inode, donor_inode);
348 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
352 /* At this point all buffers in range are uptodate, old mapping layout
353 * is no longer required, try to drop it now. */
354 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
355 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
359 ext4_double_down_write_data_sem(orig_inode, donor_inode);
360 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
361 orig_blk_offset, donor_blk_offset,
362 block_len_in_page, 1, err);
363 ext4_double_up_write_data_sem(orig_inode, donor_inode);
365 if (replaced_count) {
366 block_len_in_page = replaced_count;
368 block_len_in_page << orig_inode->i_blkbits;
372 /* Perform all necessary steps similar write_begin()/write_end()
373 * but keeping in mind that i_size will not change */
374 if (!page_has_buffers(pagep[0]))
375 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
376 bh = page_buffers(pagep[0]);
377 for (i = 0; i < data_offset_in_page; i++)
378 bh = bh->b_this_page;
379 for (i = 0; i < block_len_in_page; i++) {
380 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
383 bh = bh->b_this_page;
386 *err = block_commit_write(pagep[0], from, from + replaced_size);
388 if (unlikely(*err < 0))
389 goto repair_branches;
391 /* Even in case of data=writeback it is reasonable to pin
392 * inode to transaction, to prevent unexpected data loss */
393 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
394 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
397 unlock_page(pagep[0]);
399 unlock_page(pagep[1]);
402 ext4_journal_stop(handle);
403 if (*err == -ENOSPC &&
404 ext4_should_retry_alloc(sb, &retries))
406 /* Buffer was busy because probably is pinned to journal transaction,
407 * force transaction commit may help to free it. */
408 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
409 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
411 return replaced_count;
415 * This should never ever happen!
416 * Extents are swapped already, but we are not able to copy data.
417 * Try to swap extents to it's original places
419 ext4_double_down_write_data_sem(orig_inode, donor_inode);
420 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
421 orig_blk_offset, donor_blk_offset,
422 block_len_in_page, 0, &err2);
423 ext4_double_up_write_data_sem(orig_inode, donor_inode);
424 if (replaced_count != block_len_in_page) {
425 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
426 EIO, "Unable to copy data block,"
427 " data will be lost.");
435 * mext_check_arguments - Check whether move extent can be done
437 * @orig_inode: original inode
438 * @donor_inode: donor inode
439 * @orig_start: logical start offset in block for orig
440 * @donor_start: logical start offset in block for donor
441 * @len: the number of blocks to be moved
443 * Check the arguments of ext4_move_extents() whether the files can be
444 * exchanged with each other.
445 * Return 0 on success, or a negative error value on failure.
448 mext_check_arguments(struct inode *orig_inode,
449 struct inode *donor_inode, __u64 orig_start,
450 __u64 donor_start, __u64 *len)
452 __u64 orig_eof, donor_eof;
453 unsigned int blkbits = orig_inode->i_blkbits;
454 unsigned int blocksize = 1 << blkbits;
456 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
457 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
460 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
461 ext4_debug("ext4 move extent: suid or sgid is set"
462 " to donor file [ino:orig %lu, donor %lu]\n",
463 orig_inode->i_ino, donor_inode->i_ino);
467 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
470 /* Ext4 move extent does not support swapfile */
471 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
472 ext4_debug("ext4 move extent: The argument files should "
473 "not be swapfile [ino:orig %lu, donor %lu]\n",
474 orig_inode->i_ino, donor_inode->i_ino);
478 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
479 ext4_debug("ext4 move extent: The argument files should "
480 "not be quota files [ino:orig %lu, donor %lu]\n",
481 orig_inode->i_ino, donor_inode->i_ino);
485 /* Ext4 move extent supports only extent based file */
486 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
487 ext4_debug("ext4 move extent: orig file is not extents "
488 "based file [ino:orig %lu]\n", orig_inode->i_ino);
490 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
491 ext4_debug("ext4 move extent: donor file is not extents "
492 "based file [ino:donor %lu]\n", donor_inode->i_ino);
496 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
497 ext4_debug("ext4 move extent: File size is 0 byte\n");
501 /* Start offset should be same */
502 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
503 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
504 ext4_debug("ext4 move extent: orig and donor's start "
505 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
506 orig_inode->i_ino, donor_inode->i_ino);
510 if ((orig_start >= EXT_MAX_BLOCKS) ||
511 (donor_start >= EXT_MAX_BLOCKS) ||
512 (*len > EXT_MAX_BLOCKS) ||
513 (donor_start + *len >= EXT_MAX_BLOCKS) ||
514 (orig_start + *len >= EXT_MAX_BLOCKS)) {
515 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
516 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
517 orig_inode->i_ino, donor_inode->i_ino);
520 if (orig_eof <= orig_start)
522 else if (orig_eof < orig_start + *len - 1)
523 *len = orig_eof - orig_start;
524 if (donor_eof <= donor_start)
526 else if (donor_eof < donor_start + *len - 1)
527 *len = donor_eof - donor_start;
529 ext4_debug("ext4 move extent: len should not be 0 "
530 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
539 * ext4_move_extents - Exchange the specified range of a file
541 * @o_filp: file structure of the original file
542 * @d_filp: file structure of the donor file
543 * @orig_blk: start offset in block for orig
544 * @donor_blk: start offset in block for donor
545 * @len: the number of blocks to be moved
546 * @moved_len: moved block length
548 * This function returns 0 and moved block length is set in moved_len
549 * if succeed, otherwise returns error value.
553 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
554 __u64 donor_blk, __u64 len, __u64 *moved_len)
556 struct inode *orig_inode = file_inode(o_filp);
557 struct inode *donor_inode = file_inode(d_filp);
558 struct ext4_ext_path *path = NULL;
559 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
560 ext4_lblk_t o_end, o_start = orig_blk;
561 ext4_lblk_t d_start = donor_blk;
564 if (orig_inode->i_sb != donor_inode->i_sb) {
565 ext4_debug("ext4 move extent: The argument files "
566 "should be in same FS [ino:orig %lu, donor %lu]\n",
567 orig_inode->i_ino, donor_inode->i_ino);
571 /* orig and donor should be different inodes */
572 if (orig_inode == donor_inode) {
573 ext4_debug("ext4 move extent: The argument files should not "
574 "be same inode [ino:orig %lu, donor %lu]\n",
575 orig_inode->i_ino, donor_inode->i_ino);
579 /* Regular file check */
580 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
581 ext4_debug("ext4 move extent: The argument files should be "
582 "regular file [ino:orig %lu, donor %lu]\n",
583 orig_inode->i_ino, donor_inode->i_ino);
587 /* TODO: it's not obvious how to swap blocks for inodes with full
588 journaling enabled */
589 if (ext4_should_journal_data(orig_inode) ||
590 ext4_should_journal_data(donor_inode)) {
591 ext4_msg(orig_inode->i_sb, KERN_ERR,
592 "Online defrag not supported with data journaling");
596 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
597 ext4_msg(orig_inode->i_sb, KERN_ERR,
598 "Online defrag not supported for encrypted files");
602 /* Protect orig and donor inodes against a truncate */
603 lock_two_nondirectories(orig_inode, donor_inode);
605 /* Wait for all existing dio workers */
606 inode_dio_wait(orig_inode);
607 inode_dio_wait(donor_inode);
609 /* Protect extent tree against block allocations via delalloc */
610 ext4_double_down_write_data_sem(orig_inode, donor_inode);
611 /* Check the filesystem environment whether move_extent can be done */
612 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
616 o_end = o_start + len;
618 while (o_start < o_end) {
619 struct ext4_extent *ex;
620 ext4_lblk_t cur_blk, next_blk;
621 pgoff_t orig_page_index, donor_page_index;
623 int unwritten, cur_len;
625 ret = get_ext_path(orig_inode, o_start, &path);
628 ex = path[path->p_depth].p_ext;
629 next_blk = ext4_ext_next_allocated_block(path);
630 cur_blk = le32_to_cpu(ex->ee_block);
631 cur_len = ext4_ext_get_actual_len(ex);
632 /* Check hole before the start pos */
633 if (cur_blk + cur_len - 1 < o_start) {
634 if (next_blk == EXT_MAX_BLOCKS) {
639 d_start += next_blk - o_start;
642 /* Check hole after the start pos */
643 } else if (cur_blk > o_start) {
645 d_start += cur_blk - o_start;
647 /* Extent inside requested range ?*/
648 if (cur_blk >= o_end)
650 } else { /* in_range(o_start, o_blk, o_len) */
651 cur_len += cur_blk - o_start;
653 unwritten = ext4_ext_is_unwritten(ex);
654 if (o_end - o_start < cur_len)
655 cur_len = o_end - o_start;
657 orig_page_index = o_start >> (PAGE_SHIFT -
658 orig_inode->i_blkbits);
659 donor_page_index = d_start >> (PAGE_SHIFT -
660 donor_inode->i_blkbits);
661 offset_in_page = o_start % blocks_per_page;
662 if (cur_len > blocks_per_page- offset_in_page)
663 cur_len = blocks_per_page - offset_in_page;
665 * Up semaphore to avoid following problems:
666 * a. transaction deadlock among ext4_journal_start,
667 * ->write_begin via pagefault, and jbd2_journal_commit
668 * b. racing with ->readpage, ->write_begin, and ext4_get_block
669 * in move_extent_per_page
671 ext4_double_up_write_data_sem(orig_inode, donor_inode);
672 /* Swap original branches with new branches */
673 move_extent_per_page(o_filp, donor_inode,
674 orig_page_index, donor_page_index,
675 offset_in_page, cur_len,
677 ext4_double_down_write_data_sem(orig_inode, donor_inode);
683 *moved_len = o_start - orig_blk;
684 if (*moved_len > len)
689 ext4_discard_preallocations(orig_inode, 0);
690 ext4_discard_preallocations(donor_inode, 0);
693 ext4_ext_drop_refs(path);
695 ext4_double_up_write_data_sem(orig_inode, donor_inode);
696 unlock_two_nondirectories(orig_inode, donor_inode);