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 <linux/sched/mm.h>
12 #include "ext4_jbd2.h"
14 #include "ext4_extents.h"
17 * get_ext_path() - Find an extent path for designated logical block number.
18 * @inode: inode to be searched
19 * @lblock: logical block number to find an extent path
20 * @ppath: pointer to an extent path pointer (for output)
22 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
26 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
27 struct ext4_ext_path **ppath)
29 struct ext4_ext_path *path;
31 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
34 if (path[ext_depth(inode)].p_ext == NULL) {
35 ext4_free_ext_path(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);
108 ext4_free_ext_path(path);
113 * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2
115 * @inode1: the inode structure
116 * @inode2: the inode structure
117 * @index1: folio index
118 * @index2: folio index
119 * @folio: result folio vector
121 * Grab two locked folio for inode's by inode order
124 mext_folio_double_lock(struct inode *inode1, struct inode *inode2,
125 pgoff_t index1, pgoff_t index2, struct folio *folio[2])
127 struct address_space *mapping[2];
130 BUG_ON(!inode1 || !inode2);
131 if (inode1 < inode2) {
132 mapping[0] = inode1->i_mapping;
133 mapping[1] = inode2->i_mapping;
135 swap(index1, index2);
136 mapping[0] = inode2->i_mapping;
137 mapping[1] = inode1->i_mapping;
140 flags = memalloc_nofs_save();
141 folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN,
142 mapping_gfp_mask(mapping[0]));
143 if (IS_ERR(folio[0])) {
144 memalloc_nofs_restore(flags);
145 return PTR_ERR(folio[0]);
148 folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN,
149 mapping_gfp_mask(mapping[1]));
150 memalloc_nofs_restore(flags);
151 if (IS_ERR(folio[1])) {
152 folio_unlock(folio[0]);
154 return PTR_ERR(folio[1]);
157 * __filemap_get_folio() may not wait on folio's writeback if
158 * BDI not demand that. But it is reasonable to be very conservative
159 * here and explicitly wait on folio's writeback
161 folio_wait_writeback(folio[0]);
162 folio_wait_writeback(folio[1]);
164 swap(folio[0], folio[1]);
169 /* Force page buffers uptodate w/o dropping page's lock */
171 mext_page_mkuptodate(struct folio *folio, unsigned from, unsigned to)
173 struct inode *inode = folio->mapping->host;
175 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
176 unsigned int blocksize, block_start, block_end;
177 int i, err, nr = 0, partial = 0;
178 BUG_ON(!folio_test_locked(folio));
179 BUG_ON(folio_test_writeback(folio));
181 if (folio_test_uptodate(folio))
184 blocksize = i_blocksize(inode);
185 head = folio_buffers(folio);
187 create_empty_buffers(&folio->page, blocksize, 0);
188 head = folio_buffers(folio);
191 block = (sector_t)folio->index << (PAGE_SHIFT - inode->i_blkbits);
192 for (bh = head, block_start = 0; bh != head || !block_start;
193 block++, block_start = block_end, bh = bh->b_this_page) {
194 block_end = block_start + blocksize;
195 if (block_end <= from || block_start >= to) {
196 if (!buffer_uptodate(bh))
200 if (buffer_uptodate(bh))
202 if (!buffer_mapped(bh)) {
203 err = ext4_get_block(inode, block, bh, 0);
205 folio_set_error(folio);
208 if (!buffer_mapped(bh)) {
209 folio_zero_range(folio, block_start, blocksize);
210 set_buffer_uptodate(bh);
214 BUG_ON(nr >= MAX_BUF_PER_PAGE);
221 for (i = 0; i < nr; i++) {
223 if (!bh_uptodate_or_lock(bh)) {
224 err = ext4_read_bh(bh, 0, NULL);
231 folio_mark_uptodate(folio);
236 * move_extent_per_page - Move extent data per page
238 * @o_filp: file structure of original file
239 * @donor_inode: donor inode
240 * @orig_page_offset: page index on original file
241 * @donor_page_offset: page index on donor file
242 * @data_offset_in_page: block index where data swapping starts
243 * @block_len_in_page: the number of blocks to be swapped
244 * @unwritten: orig extent is unwritten or not
245 * @err: pointer to save return value
247 * Save the data in original inode blocks and replace original inode extents
248 * with donor inode extents by calling ext4_swap_extents().
249 * Finally, write out the saved data in new original inode blocks. Return
250 * replaced block count.
253 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
254 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
255 int data_offset_in_page,
256 int block_len_in_page, int unwritten, int *err)
258 struct inode *orig_inode = file_inode(o_filp);
259 struct folio *folio[2] = {NULL, NULL};
261 ext4_lblk_t orig_blk_offset, donor_blk_offset;
262 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
263 unsigned int tmp_data_size, data_size, replaced_size;
264 int i, err2, jblocks, retries = 0;
265 int replaced_count = 0;
266 int from = data_offset_in_page << orig_inode->i_blkbits;
267 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
268 struct super_block *sb = orig_inode->i_sb;
269 struct buffer_head *bh = NULL;
272 * It needs twice the amount of ordinary journal buffers because
273 * inode and donor_inode may change each different metadata blocks.
277 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
278 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
279 if (IS_ERR(handle)) {
280 *err = PTR_ERR(handle);
284 orig_blk_offset = orig_page_offset * blocks_per_page +
287 donor_blk_offset = donor_page_offset * blocks_per_page +
290 /* Calculate data_size */
291 if ((orig_blk_offset + block_len_in_page - 1) ==
292 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
293 /* Replace the last block */
294 tmp_data_size = orig_inode->i_size & (blocksize - 1);
296 * If data_size equal zero, it shows data_size is multiples of
297 * blocksize. So we set appropriate value.
299 if (tmp_data_size == 0)
300 tmp_data_size = blocksize;
302 data_size = tmp_data_size +
303 ((block_len_in_page - 1) << orig_inode->i_blkbits);
305 data_size = block_len_in_page << orig_inode->i_blkbits;
307 replaced_size = data_size;
309 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
310 donor_page_offset, folio);
311 if (unlikely(*err < 0))
314 * If orig extent was unwritten it can become initialized
315 * at any time after i_data_sem was dropped, in order to
316 * serialize with delalloc we have recheck extent while we
317 * hold page's lock, if it is still the case data copy is not
318 * necessary, just swap data blocks between orig and donor.
321 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
322 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
323 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
326 ext4_double_down_write_data_sem(orig_inode, donor_inode);
327 /* If any of extents in range became initialized we have to
328 * fallback to data copying */
329 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
330 block_len_in_page, 1, err);
334 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
335 block_len_in_page, 1, err);
340 ext4_double_up_write_data_sem(orig_inode, donor_inode);
343 if ((folio_has_private(folio[0]) &&
344 !filemap_release_folio(folio[0], 0)) ||
345 (folio_has_private(folio[1]) &&
346 !filemap_release_folio(folio[1], 0))) {
350 replaced_count = ext4_swap_extents(handle, orig_inode,
351 donor_inode, orig_blk_offset,
353 block_len_in_page, 1, err);
355 ext4_double_up_write_data_sem(orig_inode, donor_inode);
359 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
363 /* At this point all buffers in range are uptodate, old mapping layout
364 * is no longer required, try to drop it now. */
365 if ((folio_has_private(folio[0]) &&
366 !filemap_release_folio(folio[0], 0)) ||
367 (folio_has_private(folio[1]) &&
368 !filemap_release_folio(folio[1], 0))) {
372 ext4_double_down_write_data_sem(orig_inode, donor_inode);
373 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
374 orig_blk_offset, donor_blk_offset,
375 block_len_in_page, 1, err);
376 ext4_double_up_write_data_sem(orig_inode, donor_inode);
378 if (replaced_count) {
379 block_len_in_page = replaced_count;
381 block_len_in_page << orig_inode->i_blkbits;
385 /* Perform all necessary steps similar write_begin()/write_end()
386 * but keeping in mind that i_size will not change */
387 if (!folio_buffers(folio[0]))
388 create_empty_buffers(&folio[0]->page, 1 << orig_inode->i_blkbits, 0);
389 bh = folio_buffers(folio[0]);
390 for (i = 0; i < data_offset_in_page; i++)
391 bh = bh->b_this_page;
392 for (i = 0; i < block_len_in_page; i++) {
393 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
396 bh = bh->b_this_page;
399 *err = block_commit_write(&folio[0]->page, from, from + replaced_size);
401 if (unlikely(*err < 0))
402 goto repair_branches;
404 /* Even in case of data=writeback it is reasonable to pin
405 * inode to transaction, to prevent unexpected data loss */
406 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
407 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
410 folio_unlock(folio[0]);
412 folio_unlock(folio[1]);
415 ext4_journal_stop(handle);
416 if (*err == -ENOSPC &&
417 ext4_should_retry_alloc(sb, &retries))
419 /* Buffer was busy because probably is pinned to journal transaction,
420 * force transaction commit may help to free it. */
421 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
422 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
424 return replaced_count;
428 * This should never ever happen!
429 * Extents are swapped already, but we are not able to copy data.
430 * Try to swap extents to it's original places
432 ext4_double_down_write_data_sem(orig_inode, donor_inode);
433 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
434 orig_blk_offset, donor_blk_offset,
435 block_len_in_page, 0, &err2);
436 ext4_double_up_write_data_sem(orig_inode, donor_inode);
437 if (replaced_count != block_len_in_page) {
438 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
439 EIO, "Unable to copy data block,"
440 " data will be lost.");
448 * mext_check_arguments - Check whether move extent can be done
450 * @orig_inode: original inode
451 * @donor_inode: donor inode
452 * @orig_start: logical start offset in block for orig
453 * @donor_start: logical start offset in block for donor
454 * @len: the number of blocks to be moved
456 * Check the arguments of ext4_move_extents() whether the files can be
457 * exchanged with each other.
458 * Return 0 on success, or a negative error value on failure.
461 mext_check_arguments(struct inode *orig_inode,
462 struct inode *donor_inode, __u64 orig_start,
463 __u64 donor_start, __u64 *len)
465 __u64 orig_eof, donor_eof;
466 unsigned int blkbits = orig_inode->i_blkbits;
467 unsigned int blocksize = 1 << blkbits;
469 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
470 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
473 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
474 ext4_debug("ext4 move extent: suid or sgid is set"
475 " to donor file [ino:orig %lu, donor %lu]\n",
476 orig_inode->i_ino, donor_inode->i_ino);
480 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
483 /* Ext4 move extent does not support swap files */
484 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
485 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
486 orig_inode->i_ino, donor_inode->i_ino);
490 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
491 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
492 orig_inode->i_ino, donor_inode->i_ino);
496 /* Ext4 move extent supports only extent based file */
497 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
498 ext4_debug("ext4 move extent: orig file is not extents "
499 "based file [ino:orig %lu]\n", orig_inode->i_ino);
501 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
502 ext4_debug("ext4 move extent: donor file is not extents "
503 "based file [ino:donor %lu]\n", donor_inode->i_ino);
507 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
508 ext4_debug("ext4 move extent: File size is 0 byte\n");
512 /* Start offset should be same */
513 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
514 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
515 ext4_debug("ext4 move extent: orig and donor's start "
516 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
517 orig_inode->i_ino, donor_inode->i_ino);
521 if ((orig_start >= EXT_MAX_BLOCKS) ||
522 (donor_start >= EXT_MAX_BLOCKS) ||
523 (*len > EXT_MAX_BLOCKS) ||
524 (donor_start + *len >= EXT_MAX_BLOCKS) ||
525 (orig_start + *len >= EXT_MAX_BLOCKS)) {
526 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
527 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
528 orig_inode->i_ino, donor_inode->i_ino);
531 if (orig_eof <= orig_start)
533 else if (orig_eof < orig_start + *len - 1)
534 *len = orig_eof - orig_start;
535 if (donor_eof <= donor_start)
537 else if (donor_eof < donor_start + *len - 1)
538 *len = donor_eof - donor_start;
540 ext4_debug("ext4 move extent: len should not be 0 "
541 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
550 * ext4_move_extents - Exchange the specified range of a file
552 * @o_filp: file structure of the original file
553 * @d_filp: file structure of the donor file
554 * @orig_blk: start offset in block for orig
555 * @donor_blk: start offset in block for donor
556 * @len: the number of blocks to be moved
557 * @moved_len: moved block length
559 * This function returns 0 and moved block length is set in moved_len
560 * if succeed, otherwise returns error value.
564 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
565 __u64 donor_blk, __u64 len, __u64 *moved_len)
567 struct inode *orig_inode = file_inode(o_filp);
568 struct inode *donor_inode = file_inode(d_filp);
569 struct ext4_ext_path *path = NULL;
570 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
571 ext4_lblk_t o_end, o_start = orig_blk;
572 ext4_lblk_t d_start = donor_blk;
575 if (orig_inode->i_sb != donor_inode->i_sb) {
576 ext4_debug("ext4 move extent: The argument files "
577 "should be in same FS [ino:orig %lu, donor %lu]\n",
578 orig_inode->i_ino, donor_inode->i_ino);
582 /* orig and donor should be different inodes */
583 if (orig_inode == donor_inode) {
584 ext4_debug("ext4 move extent: The argument files should not "
585 "be same inode [ino:orig %lu, donor %lu]\n",
586 orig_inode->i_ino, donor_inode->i_ino);
590 /* Regular file check */
591 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
592 ext4_debug("ext4 move extent: The argument files should be "
593 "regular file [ino:orig %lu, donor %lu]\n",
594 orig_inode->i_ino, donor_inode->i_ino);
598 /* TODO: it's not obvious how to swap blocks for inodes with full
599 journaling enabled */
600 if (ext4_should_journal_data(orig_inode) ||
601 ext4_should_journal_data(donor_inode)) {
602 ext4_msg(orig_inode->i_sb, KERN_ERR,
603 "Online defrag not supported with data journaling");
607 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
608 ext4_msg(orig_inode->i_sb, KERN_ERR,
609 "Online defrag not supported for encrypted files");
613 /* Protect orig and donor inodes against a truncate */
614 lock_two_nondirectories(orig_inode, donor_inode);
616 /* Wait for all existing dio workers */
617 inode_dio_wait(orig_inode);
618 inode_dio_wait(donor_inode);
620 /* Protect extent tree against block allocations via delalloc */
621 ext4_double_down_write_data_sem(orig_inode, donor_inode);
622 /* Check the filesystem environment whether move_extent can be done */
623 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
627 o_end = o_start + len;
629 while (o_start < o_end) {
630 struct ext4_extent *ex;
631 ext4_lblk_t cur_blk, next_blk;
632 pgoff_t orig_page_index, donor_page_index;
634 int unwritten, cur_len;
636 ret = get_ext_path(orig_inode, o_start, &path);
639 ex = path[path->p_depth].p_ext;
640 cur_blk = le32_to_cpu(ex->ee_block);
641 cur_len = ext4_ext_get_actual_len(ex);
642 /* Check hole before the start pos */
643 if (cur_blk + cur_len - 1 < o_start) {
644 next_blk = ext4_ext_next_allocated_block(path);
645 if (next_blk == EXT_MAX_BLOCKS) {
649 d_start += next_blk - o_start;
652 /* Check hole after the start pos */
653 } else if (cur_blk > o_start) {
655 d_start += cur_blk - o_start;
657 /* Extent inside requested range ?*/
658 if (cur_blk >= o_end)
660 } else { /* in_range(o_start, o_blk, o_len) */
661 cur_len += cur_blk - o_start;
663 unwritten = ext4_ext_is_unwritten(ex);
664 if (o_end - o_start < cur_len)
665 cur_len = o_end - o_start;
667 orig_page_index = o_start >> (PAGE_SHIFT -
668 orig_inode->i_blkbits);
669 donor_page_index = d_start >> (PAGE_SHIFT -
670 donor_inode->i_blkbits);
671 offset_in_page = o_start % blocks_per_page;
672 if (cur_len > blocks_per_page - offset_in_page)
673 cur_len = blocks_per_page - offset_in_page;
675 * Up semaphore to avoid following problems:
676 * a. transaction deadlock among ext4_journal_start,
677 * ->write_begin via pagefault, and jbd2_journal_commit
678 * b. racing with ->read_folio, ->write_begin, and
679 * ext4_get_block in move_extent_per_page
681 ext4_double_up_write_data_sem(orig_inode, donor_inode);
682 /* Swap original branches with new branches */
683 move_extent_per_page(o_filp, donor_inode,
684 orig_page_index, donor_page_index,
685 offset_in_page, cur_len,
687 ext4_double_down_write_data_sem(orig_inode, donor_inode);
693 *moved_len = o_start - orig_blk;
694 if (*moved_len > len)
699 ext4_discard_preallocations(orig_inode, 0);
700 ext4_discard_preallocations(donor_inode, 0);
703 ext4_free_ext_path(path);
704 ext4_double_up_write_data_sem(orig_inode, donor_inode);
705 unlock_two_nondirectories(orig_inode, donor_inode);