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_ext_drop_refs(path);
45 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
46 * @first: inode to be locked
47 * @second: inode to be locked
49 * Acquire write lock of i_data_sem of the two inodes
52 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
55 down_write(&EXT4_I(first)->i_data_sem);
56 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
58 down_write(&EXT4_I(second)->i_data_sem);
59 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
65 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
67 * @orig_inode: original inode structure to be released its lock first
68 * @donor_inode: donor inode structure to be released its lock second
69 * Release write lock of i_data_sem of two inodes (orig and donor).
72 ext4_double_up_write_data_sem(struct inode *orig_inode,
73 struct inode *donor_inode)
75 up_write(&EXT4_I(orig_inode)->i_data_sem);
76 up_write(&EXT4_I(donor_inode)->i_data_sem);
80 * mext_check_coverage - Check that all extents in range has the same type
82 * @inode: inode in question
83 * @from: block offset of inode
84 * @count: block count to be checked
85 * @unwritten: extents expected to be unwritten
86 * @err: pointer to save error value
88 * Return 1 if all extents in range has expected type, and zero otherwise.
91 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
92 int unwritten, int *err)
94 struct ext4_ext_path *path = NULL;
95 struct ext4_extent *ext;
97 ext4_lblk_t last = from + count;
99 *err = get_ext_path(inode, from, &path);
102 ext = path[ext_depth(inode)].p_ext;
103 if (unwritten != ext4_ext_is_unwritten(ext))
105 from += ext4_ext_get_actual_len(ext);
106 ext4_ext_drop_refs(path);
110 ext4_ext_drop_refs(path);
116 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
118 * @inode1: the inode structure
119 * @inode2: the inode structure
120 * @index1: page index
121 * @index2: page index
122 * @page: result page vector
124 * Grab two locked pages for inode's by inode order
127 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
128 pgoff_t index1, pgoff_t index2, struct page *page[2])
130 struct address_space *mapping[2];
133 BUG_ON(!inode1 || !inode2);
134 if (inode1 < inode2) {
135 mapping[0] = inode1->i_mapping;
136 mapping[1] = inode2->i_mapping;
138 swap(index1, index2);
139 mapping[0] = inode2->i_mapping;
140 mapping[1] = inode1->i_mapping;
143 flags = memalloc_nofs_save();
144 page[0] = grab_cache_page_write_begin(mapping[0], index1);
146 memalloc_nofs_restore(flags);
150 page[1] = grab_cache_page_write_begin(mapping[1], index2);
151 memalloc_nofs_restore(flags);
153 unlock_page(page[0]);
158 * grab_cache_page_write_begin() may not wait on page's writeback if
159 * BDI not demand that. But it is reasonable to be very conservative
160 * here and explicitly wait on page's writeback
162 wait_on_page_writeback(page[0]);
163 wait_on_page_writeback(page[1]);
165 swap(page[0], page[1]);
170 /* Force page buffers uptodate w/o dropping page's lock */
172 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
174 struct inode *inode = page->mapping->host;
176 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
177 unsigned int blocksize, block_start, block_end;
178 int i, err, nr = 0, partial = 0;
179 BUG_ON(!PageLocked(page));
180 BUG_ON(PageWriteback(page));
182 if (PageUptodate(page))
185 blocksize = i_blocksize(inode);
186 if (!page_has_buffers(page))
187 create_empty_buffers(page, blocksize, 0);
189 head = page_buffers(page);
190 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
191 for (bh = head, block_start = 0; bh != head || !block_start;
192 block++, block_start = block_end, bh = bh->b_this_page) {
193 block_end = block_start + blocksize;
194 if (block_end <= from || block_start >= to) {
195 if (!buffer_uptodate(bh))
199 if (buffer_uptodate(bh))
201 if (!buffer_mapped(bh)) {
202 err = ext4_get_block(inode, block, bh, 0);
207 if (!buffer_mapped(bh)) {
208 zero_user(page, block_start, blocksize);
209 set_buffer_uptodate(bh);
213 BUG_ON(nr >= MAX_BUF_PER_PAGE);
220 for (i = 0; i < nr; i++) {
222 if (!bh_uptodate_or_lock(bh)) {
223 err = ext4_read_bh(bh, 0, NULL);
230 SetPageUptodate(page);
235 * move_extent_per_page - Move extent data per page
237 * @o_filp: file structure of original file
238 * @donor_inode: donor inode
239 * @orig_page_offset: page index on original file
240 * @donor_page_offset: page index on donor file
241 * @data_offset_in_page: block index where data swapping starts
242 * @block_len_in_page: the number of blocks to be swapped
243 * @unwritten: orig extent is unwritten or not
244 * @err: pointer to save return value
246 * Save the data in original inode blocks and replace original inode extents
247 * with donor inode extents by calling ext4_swap_extents().
248 * Finally, write out the saved data in new original inode blocks. Return
249 * replaced block count.
252 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
253 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
254 int data_offset_in_page,
255 int block_len_in_page, int unwritten, int *err)
257 struct inode *orig_inode = file_inode(o_filp);
258 struct page *pagep[2] = {NULL, NULL};
260 ext4_lblk_t orig_blk_offset, donor_blk_offset;
261 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
262 unsigned int tmp_data_size, data_size, replaced_size;
263 int i, err2, jblocks, retries = 0;
264 int replaced_count = 0;
265 int from = data_offset_in_page << orig_inode->i_blkbits;
266 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
267 struct super_block *sb = orig_inode->i_sb;
268 struct buffer_head *bh = NULL;
271 * It needs twice the amount of ordinary journal buffers because
272 * inode and donor_inode may change each different metadata blocks.
276 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
277 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
278 if (IS_ERR(handle)) {
279 *err = PTR_ERR(handle);
283 orig_blk_offset = orig_page_offset * blocks_per_page +
286 donor_blk_offset = donor_page_offset * blocks_per_page +
289 /* Calculate data_size */
290 if ((orig_blk_offset + block_len_in_page - 1) ==
291 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
292 /* Replace the last block */
293 tmp_data_size = orig_inode->i_size & (blocksize - 1);
295 * If data_size equal zero, it shows data_size is multiples of
296 * blocksize. So we set appropriate value.
298 if (tmp_data_size == 0)
299 tmp_data_size = blocksize;
301 data_size = tmp_data_size +
302 ((block_len_in_page - 1) << orig_inode->i_blkbits);
304 data_size = block_len_in_page << orig_inode->i_blkbits;
306 replaced_size = data_size;
308 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
309 donor_page_offset, pagep);
310 if (unlikely(*err < 0))
313 * If orig extent was unwritten it can become initialized
314 * at any time after i_data_sem was dropped, in order to
315 * serialize with delalloc we have recheck extent while we
316 * hold page's lock, if it is still the case data copy is not
317 * necessary, just swap data blocks between orig and donor.
320 ext4_double_down_write_data_sem(orig_inode, donor_inode);
321 /* If any of extents in range became initialized we have to
322 * fallback to data copying */
323 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
324 block_len_in_page, 1, err);
328 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
329 block_len_in_page, 1, err);
334 ext4_double_up_write_data_sem(orig_inode, donor_inode);
337 if ((page_has_private(pagep[0]) &&
338 !try_to_release_page(pagep[0], 0)) ||
339 (page_has_private(pagep[1]) &&
340 !try_to_release_page(pagep[1], 0))) {
344 replaced_count = ext4_swap_extents(handle, orig_inode,
345 donor_inode, orig_blk_offset,
347 block_len_in_page, 1, err);
349 ext4_double_up_write_data_sem(orig_inode, donor_inode);
353 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
357 /* At this point all buffers in range are uptodate, old mapping layout
358 * is no longer required, try to drop it now. */
359 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
360 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
364 ext4_double_down_write_data_sem(orig_inode, donor_inode);
365 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
366 orig_blk_offset, donor_blk_offset,
367 block_len_in_page, 1, err);
368 ext4_double_up_write_data_sem(orig_inode, donor_inode);
370 if (replaced_count) {
371 block_len_in_page = replaced_count;
373 block_len_in_page << orig_inode->i_blkbits;
377 /* Perform all necessary steps similar write_begin()/write_end()
378 * but keeping in mind that i_size will not change */
379 if (!page_has_buffers(pagep[0]))
380 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
381 bh = page_buffers(pagep[0]);
382 for (i = 0; i < data_offset_in_page; i++)
383 bh = bh->b_this_page;
384 for (i = 0; i < block_len_in_page; i++) {
385 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
388 bh = bh->b_this_page;
391 *err = block_commit_write(pagep[0], from, from + replaced_size);
393 if (unlikely(*err < 0))
394 goto repair_branches;
396 /* Even in case of data=writeback it is reasonable to pin
397 * inode to transaction, to prevent unexpected data loss */
398 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
399 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
402 unlock_page(pagep[0]);
404 unlock_page(pagep[1]);
407 ext4_journal_stop(handle);
408 if (*err == -ENOSPC &&
409 ext4_should_retry_alloc(sb, &retries))
411 /* Buffer was busy because probably is pinned to journal transaction,
412 * force transaction commit may help to free it. */
413 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
414 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
416 return replaced_count;
420 * This should never ever happen!
421 * Extents are swapped already, but we are not able to copy data.
422 * Try to swap extents to it's original places
424 ext4_double_down_write_data_sem(orig_inode, donor_inode);
425 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
426 orig_blk_offset, donor_blk_offset,
427 block_len_in_page, 0, &err2);
428 ext4_double_up_write_data_sem(orig_inode, donor_inode);
429 if (replaced_count != block_len_in_page) {
430 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
431 EIO, "Unable to copy data block,"
432 " data will be lost.");
440 * mext_check_arguments - Check whether move extent can be done
442 * @orig_inode: original inode
443 * @donor_inode: donor inode
444 * @orig_start: logical start offset in block for orig
445 * @donor_start: logical start offset in block for donor
446 * @len: the number of blocks to be moved
448 * Check the arguments of ext4_move_extents() whether the files can be
449 * exchanged with each other.
450 * Return 0 on success, or a negative error value on failure.
453 mext_check_arguments(struct inode *orig_inode,
454 struct inode *donor_inode, __u64 orig_start,
455 __u64 donor_start, __u64 *len)
457 __u64 orig_eof, donor_eof;
458 unsigned int blkbits = orig_inode->i_blkbits;
459 unsigned int blocksize = 1 << blkbits;
461 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
462 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
465 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
466 ext4_debug("ext4 move extent: suid or sgid is set"
467 " to donor file [ino:orig %lu, donor %lu]\n",
468 orig_inode->i_ino, donor_inode->i_ino);
472 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
475 /* Ext4 move extent does not support swapfile */
476 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
477 ext4_debug("ext4 move extent: The argument files should "
478 "not be swapfile [ino:orig %lu, donor %lu]\n",
479 orig_inode->i_ino, donor_inode->i_ino);
483 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
484 ext4_debug("ext4 move extent: The argument files should "
485 "not be quota files [ino:orig %lu, donor %lu]\n",
486 orig_inode->i_ino, donor_inode->i_ino);
490 /* Ext4 move extent supports only extent based file */
491 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
492 ext4_debug("ext4 move extent: orig file is not extents "
493 "based file [ino:orig %lu]\n", orig_inode->i_ino);
495 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
496 ext4_debug("ext4 move extent: donor file is not extents "
497 "based file [ino:donor %lu]\n", donor_inode->i_ino);
501 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
502 ext4_debug("ext4 move extent: File size is 0 byte\n");
506 /* Start offset should be same */
507 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
508 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
509 ext4_debug("ext4 move extent: orig and donor's start "
510 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
511 orig_inode->i_ino, donor_inode->i_ino);
515 if ((orig_start >= EXT_MAX_BLOCKS) ||
516 (donor_start >= EXT_MAX_BLOCKS) ||
517 (*len > EXT_MAX_BLOCKS) ||
518 (donor_start + *len >= EXT_MAX_BLOCKS) ||
519 (orig_start + *len >= EXT_MAX_BLOCKS)) {
520 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
521 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
522 orig_inode->i_ino, donor_inode->i_ino);
525 if (orig_eof <= orig_start)
527 else if (orig_eof < orig_start + *len - 1)
528 *len = orig_eof - orig_start;
529 if (donor_eof <= donor_start)
531 else if (donor_eof < donor_start + *len - 1)
532 *len = donor_eof - donor_start;
534 ext4_debug("ext4 move extent: len should not be 0 "
535 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
544 * ext4_move_extents - Exchange the specified range of a file
546 * @o_filp: file structure of the original file
547 * @d_filp: file structure of the donor file
548 * @orig_blk: start offset in block for orig
549 * @donor_blk: start offset in block for donor
550 * @len: the number of blocks to be moved
551 * @moved_len: moved block length
553 * This function returns 0 and moved block length is set in moved_len
554 * if succeed, otherwise returns error value.
558 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
559 __u64 donor_blk, __u64 len, __u64 *moved_len)
561 struct inode *orig_inode = file_inode(o_filp);
562 struct inode *donor_inode = file_inode(d_filp);
563 struct ext4_ext_path *path = NULL;
564 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
565 ext4_lblk_t o_end, o_start = orig_blk;
566 ext4_lblk_t d_start = donor_blk;
569 if (orig_inode->i_sb != donor_inode->i_sb) {
570 ext4_debug("ext4 move extent: The argument files "
571 "should be in same FS [ino:orig %lu, donor %lu]\n",
572 orig_inode->i_ino, donor_inode->i_ino);
576 /* orig and donor should be different inodes */
577 if (orig_inode == donor_inode) {
578 ext4_debug("ext4 move extent: The argument files should not "
579 "be same inode [ino:orig %lu, donor %lu]\n",
580 orig_inode->i_ino, donor_inode->i_ino);
584 /* Regular file check */
585 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
586 ext4_debug("ext4 move extent: The argument files should be "
587 "regular file [ino:orig %lu, donor %lu]\n",
588 orig_inode->i_ino, donor_inode->i_ino);
592 /* TODO: it's not obvious how to swap blocks for inodes with full
593 journaling enabled */
594 if (ext4_should_journal_data(orig_inode) ||
595 ext4_should_journal_data(donor_inode)) {
596 ext4_msg(orig_inode->i_sb, KERN_ERR,
597 "Online defrag not supported with data journaling");
601 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
602 ext4_msg(orig_inode->i_sb, KERN_ERR,
603 "Online defrag not supported for encrypted files");
607 /* Protect orig and donor inodes against a truncate */
608 lock_two_nondirectories(orig_inode, donor_inode);
610 /* Wait for all existing dio workers */
611 inode_dio_wait(orig_inode);
612 inode_dio_wait(donor_inode);
614 /* Protect extent tree against block allocations via delalloc */
615 ext4_double_down_write_data_sem(orig_inode, donor_inode);
616 /* Check the filesystem environment whether move_extent can be done */
617 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
621 o_end = o_start + len;
623 while (o_start < o_end) {
624 struct ext4_extent *ex;
625 ext4_lblk_t cur_blk, next_blk;
626 pgoff_t orig_page_index, donor_page_index;
628 int unwritten, cur_len;
630 ret = get_ext_path(orig_inode, o_start, &path);
633 ex = path[path->p_depth].p_ext;
634 next_blk = ext4_ext_next_allocated_block(path);
635 cur_blk = le32_to_cpu(ex->ee_block);
636 cur_len = ext4_ext_get_actual_len(ex);
637 /* Check hole before the start pos */
638 if (cur_blk + cur_len - 1 < o_start) {
639 if (next_blk == EXT_MAX_BLOCKS) {
643 d_start += next_blk - o_start;
646 /* Check hole after the start pos */
647 } else if (cur_blk > o_start) {
649 d_start += cur_blk - o_start;
651 /* Extent inside requested range ?*/
652 if (cur_blk >= o_end)
654 } else { /* in_range(o_start, o_blk, o_len) */
655 cur_len += cur_blk - o_start;
657 unwritten = ext4_ext_is_unwritten(ex);
658 if (o_end - o_start < cur_len)
659 cur_len = o_end - o_start;
661 orig_page_index = o_start >> (PAGE_SHIFT -
662 orig_inode->i_blkbits);
663 donor_page_index = d_start >> (PAGE_SHIFT -
664 donor_inode->i_blkbits);
665 offset_in_page = o_start % blocks_per_page;
666 if (cur_len > blocks_per_page- offset_in_page)
667 cur_len = blocks_per_page - offset_in_page;
669 * Up semaphore to avoid following problems:
670 * a. transaction deadlock among ext4_journal_start,
671 * ->write_begin via pagefault, and jbd2_journal_commit
672 * b. racing with ->read_folio, ->write_begin, and
673 * ext4_get_block in move_extent_per_page
675 ext4_double_up_write_data_sem(orig_inode, donor_inode);
676 /* Swap original branches with new branches */
677 move_extent_per_page(o_filp, donor_inode,
678 orig_page_index, donor_page_index,
679 offset_in_page, cur_len,
681 ext4_double_down_write_data_sem(orig_inode, donor_inode);
687 *moved_len = o_start - orig_blk;
688 if (*moved_len > len)
693 ext4_discard_preallocations(orig_inode, 0);
694 ext4_discard_preallocations(donor_inode, 0);
697 ext4_ext_drop_refs(path);
699 ext4_double_up_write_data_sem(orig_inode, donor_inode);
700 unlock_two_nondirectories(orig_inode, donor_inode);