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.
18 * @inode: an inode which is searched
19 * @lblock: logical block number to find an extent path
20 * @path: 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 - Acquire two inodes' write lock
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 pgoff_t tmp = index1;
140 mapping[0] = inode2->i_mapping;
141 mapping[1] = inode1->i_mapping;
144 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
148 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
150 unlock_page(page[0]);
155 * grab_cache_page_write_begin() may not wait on page's writeback if
156 * BDI not demand that. But it is reasonable to be very conservative
157 * here and explicitly wait on page's writeback
159 wait_on_page_writeback(page[0]);
160 wait_on_page_writeback(page[1]);
162 swap(page[0], page[1]);
167 /* Force page buffers uptodate w/o dropping page's lock */
169 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
171 struct inode *inode = page->mapping->host;
173 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
174 unsigned int blocksize, block_start, block_end;
175 int i, err, nr = 0, partial = 0;
176 BUG_ON(!PageLocked(page));
177 BUG_ON(PageWriteback(page));
179 if (PageUptodate(page))
182 blocksize = i_blocksize(inode);
183 if (!page_has_buffers(page))
184 create_empty_buffers(page, blocksize, 0);
186 head = page_buffers(page);
187 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
188 for (bh = head, block_start = 0; bh != head || !block_start;
189 block++, block_start = block_end, bh = bh->b_this_page) {
190 block_end = block_start + blocksize;
191 if (block_end <= from || block_start >= to) {
192 if (!buffer_uptodate(bh))
196 if (buffer_uptodate(bh))
198 if (!buffer_mapped(bh)) {
199 err = ext4_get_block(inode, block, bh, 0);
204 if (!buffer_mapped(bh)) {
205 zero_user(page, block_start, blocksize);
206 set_buffer_uptodate(bh);
210 BUG_ON(nr >= MAX_BUF_PER_PAGE);
217 for (i = 0; i < nr; i++) {
219 if (!bh_uptodate_or_lock(bh)) {
220 err = bh_submit_read(bh);
227 SetPageUptodate(page);
232 * move_extent_per_page - Move extent data per page
234 * @o_filp: file structure of original file
235 * @donor_inode: donor inode
236 * @orig_page_offset: page index on original file
237 * @donor_page_offset: page index on donor file
238 * @data_offset_in_page: block index where data swapping starts
239 * @block_len_in_page: the number of blocks to be swapped
240 * @unwritten: orig extent is unwritten or not
241 * @err: pointer to save return value
243 * Save the data in original inode blocks and replace original inode extents
244 * with donor inode extents by calling ext4_swap_extents().
245 * Finally, write out the saved data in new original inode blocks. Return
246 * replaced block count.
249 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
250 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
251 int data_offset_in_page,
252 int block_len_in_page, int unwritten, int *err)
254 struct inode *orig_inode = file_inode(o_filp);
255 struct page *pagep[2] = {NULL, NULL};
257 ext4_lblk_t orig_blk_offset, donor_blk_offset;
258 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
259 unsigned int tmp_data_size, data_size, replaced_size;
260 int i, err2, jblocks, retries = 0;
261 int replaced_count = 0;
262 int from = data_offset_in_page << orig_inode->i_blkbits;
263 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
264 struct super_block *sb = orig_inode->i_sb;
265 struct buffer_head *bh = NULL;
268 * It needs twice the amount of ordinary journal buffers because
269 * inode and donor_inode may change each different metadata blocks.
273 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
274 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
275 if (IS_ERR(handle)) {
276 *err = PTR_ERR(handle);
280 orig_blk_offset = orig_page_offset * blocks_per_page +
283 donor_blk_offset = donor_page_offset * blocks_per_page +
286 /* Calculate data_size */
287 if ((orig_blk_offset + block_len_in_page - 1) ==
288 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
289 /* Replace the last block */
290 tmp_data_size = orig_inode->i_size & (blocksize - 1);
292 * If data_size equal zero, it shows data_size is multiples of
293 * blocksize. So we set appropriate value.
295 if (tmp_data_size == 0)
296 tmp_data_size = blocksize;
298 data_size = tmp_data_size +
299 ((block_len_in_page - 1) << orig_inode->i_blkbits);
301 data_size = block_len_in_page << orig_inode->i_blkbits;
303 replaced_size = data_size;
305 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
306 donor_page_offset, pagep);
307 if (unlikely(*err < 0))
310 * If orig extent was unwritten it can become initialized
311 * at any time after i_data_sem was dropped, in order to
312 * serialize with delalloc we have recheck extent while we
313 * hold page's lock, if it is still the case data copy is not
314 * necessary, just swap data blocks between orig and donor.
317 ext4_double_down_write_data_sem(orig_inode, donor_inode);
318 /* If any of extents in range became initialized we have to
319 * fallback to data copying */
320 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
321 block_len_in_page, 1, err);
325 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
326 block_len_in_page, 1, err);
331 ext4_double_up_write_data_sem(orig_inode, donor_inode);
334 if ((page_has_private(pagep[0]) &&
335 !try_to_release_page(pagep[0], 0)) ||
336 (page_has_private(pagep[1]) &&
337 !try_to_release_page(pagep[1], 0))) {
341 replaced_count = ext4_swap_extents(handle, orig_inode,
342 donor_inode, orig_blk_offset,
344 block_len_in_page, 1, err);
346 ext4_double_up_write_data_sem(orig_inode, donor_inode);
350 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
354 /* At this point all buffers in range are uptodate, old mapping layout
355 * is no longer required, try to drop it now. */
356 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
357 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
361 ext4_double_down_write_data_sem(orig_inode, donor_inode);
362 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
363 orig_blk_offset, donor_blk_offset,
364 block_len_in_page, 1, err);
365 ext4_double_up_write_data_sem(orig_inode, donor_inode);
367 if (replaced_count) {
368 block_len_in_page = replaced_count;
370 block_len_in_page << orig_inode->i_blkbits;
374 /* Perform all necessary steps similar write_begin()/write_end()
375 * but keeping in mind that i_size will not change */
376 if (!page_has_buffers(pagep[0]))
377 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
378 bh = page_buffers(pagep[0]);
379 for (i = 0; i < data_offset_in_page; i++)
380 bh = bh->b_this_page;
381 for (i = 0; i < block_len_in_page; i++) {
382 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
385 bh = bh->b_this_page;
388 *err = block_commit_write(pagep[0], from, from + replaced_size);
390 if (unlikely(*err < 0))
391 goto repair_branches;
393 /* Even in case of data=writeback it is reasonable to pin
394 * inode to transaction, to prevent unexpected data loss */
395 *err = ext4_jbd2_inode_add_write(handle, orig_inode);
398 unlock_page(pagep[0]);
400 unlock_page(pagep[1]);
403 ext4_journal_stop(handle);
404 if (*err == -ENOSPC &&
405 ext4_should_retry_alloc(sb, &retries))
407 /* Buffer was busy because probably is pinned to journal transaction,
408 * force transaction commit may help to free it. */
409 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
410 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
412 return replaced_count;
416 * This should never ever happen!
417 * Extents are swapped already, but we are not able to copy data.
418 * Try to swap extents to it's original places
420 ext4_double_down_write_data_sem(orig_inode, donor_inode);
421 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
422 orig_blk_offset, donor_blk_offset,
423 block_len_in_page, 0, &err2);
424 ext4_double_up_write_data_sem(orig_inode, donor_inode);
425 if (replaced_count != block_len_in_page) {
426 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
427 "Unable to copy data block,"
428 " data will be lost.");
436 * mext_check_arguments - Check whether move extent can be done
438 * @orig_inode: original inode
439 * @donor_inode: donor inode
440 * @orig_start: logical start offset in block for orig
441 * @donor_start: logical start offset in block for donor
442 * @len: the number of blocks to be moved
444 * Check the arguments of ext4_move_extents() whether the files can be
445 * exchanged with each other.
446 * Return 0 on success, or a negative error value on failure.
449 mext_check_arguments(struct inode *orig_inode,
450 struct inode *donor_inode, __u64 orig_start,
451 __u64 donor_start, __u64 *len)
453 __u64 orig_eof, donor_eof;
454 unsigned int blkbits = orig_inode->i_blkbits;
455 unsigned int blocksize = 1 << blkbits;
457 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
458 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
461 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
462 ext4_debug("ext4 move extent: suid or sgid is set"
463 " to donor file [ino:orig %lu, donor %lu]\n",
464 orig_inode->i_ino, donor_inode->i_ino);
468 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
471 /* Ext4 move extent does not support swapfile */
472 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
473 ext4_debug("ext4 move extent: The argument files should "
474 "not be swapfile [ino:orig %lu, donor %lu]\n",
475 orig_inode->i_ino, donor_inode->i_ino);
479 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
480 ext4_debug("ext4 move extent: The argument files should "
481 "not be quota files [ino:orig %lu, donor %lu]\n",
482 orig_inode->i_ino, donor_inode->i_ino);
486 /* Ext4 move extent supports only extent based file */
487 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
488 ext4_debug("ext4 move extent: orig file is not extents "
489 "based file [ino:orig %lu]\n", orig_inode->i_ino);
491 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
492 ext4_debug("ext4 move extent: donor file is not extents "
493 "based file [ino:donor %lu]\n", donor_inode->i_ino);
497 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
498 ext4_debug("ext4 move extent: File size is 0 byte\n");
502 /* Start offset should be same */
503 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
504 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
505 ext4_debug("ext4 move extent: orig and donor's start "
506 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
507 orig_inode->i_ino, donor_inode->i_ino);
511 if ((orig_start >= EXT_MAX_BLOCKS) ||
512 (donor_start >= EXT_MAX_BLOCKS) ||
513 (*len > EXT_MAX_BLOCKS) ||
514 (donor_start + *len >= EXT_MAX_BLOCKS) ||
515 (orig_start + *len >= EXT_MAX_BLOCKS)) {
516 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
517 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
518 orig_inode->i_ino, donor_inode->i_ino);
521 if (orig_eof < orig_start + *len - 1)
522 *len = orig_eof - orig_start;
523 if (donor_eof < donor_start + *len - 1)
524 *len = donor_eof - donor_start;
526 ext4_debug("ext4 move extent: len should not be 0 "
527 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
536 * ext4_move_extents - Exchange the specified range of a file
538 * @o_filp: file structure of the original file
539 * @d_filp: file structure of the donor file
540 * @orig_blk: start offset in block for orig
541 * @donor_blk: start offset in block for donor
542 * @len: the number of blocks to be moved
543 * @moved_len: moved block length
545 * This function returns 0 and moved block length is set in moved_len
546 * if succeed, otherwise returns error value.
550 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
551 __u64 donor_blk, __u64 len, __u64 *moved_len)
553 struct inode *orig_inode = file_inode(o_filp);
554 struct inode *donor_inode = file_inode(d_filp);
555 struct ext4_ext_path *path = NULL;
556 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
557 ext4_lblk_t o_end, o_start = orig_blk;
558 ext4_lblk_t d_start = donor_blk;
561 if (orig_inode->i_sb != donor_inode->i_sb) {
562 ext4_debug("ext4 move extent: The argument files "
563 "should be in same FS [ino:orig %lu, donor %lu]\n",
564 orig_inode->i_ino, donor_inode->i_ino);
568 /* orig and donor should be different inodes */
569 if (orig_inode == donor_inode) {
570 ext4_debug("ext4 move extent: The argument files should not "
571 "be same inode [ino:orig %lu, donor %lu]\n",
572 orig_inode->i_ino, donor_inode->i_ino);
576 /* Regular file check */
577 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
578 ext4_debug("ext4 move extent: The argument files should be "
579 "regular file [ino:orig %lu, donor %lu]\n",
580 orig_inode->i_ino, donor_inode->i_ino);
584 /* TODO: it's not obvious how to swap blocks for inodes with full
585 journaling enabled */
586 if (ext4_should_journal_data(orig_inode) ||
587 ext4_should_journal_data(donor_inode)) {
588 ext4_msg(orig_inode->i_sb, KERN_ERR,
589 "Online defrag not supported with data journaling");
593 if (ext4_encrypted_inode(orig_inode) ||
594 ext4_encrypted_inode(donor_inode)) {
595 ext4_msg(orig_inode->i_sb, KERN_ERR,
596 "Online defrag not supported for encrypted files");
600 /* Protect orig and donor inodes against a truncate */
601 lock_two_nondirectories(orig_inode, donor_inode);
603 /* Wait for all existing dio workers */
604 inode_dio_wait(orig_inode);
605 inode_dio_wait(donor_inode);
607 /* Protect extent tree against block allocations via delalloc */
608 ext4_double_down_write_data_sem(orig_inode, donor_inode);
609 /* Check the filesystem environment whether move_extent can be done */
610 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
614 o_end = o_start + len;
616 while (o_start < o_end) {
617 struct ext4_extent *ex;
618 ext4_lblk_t cur_blk, next_blk;
619 pgoff_t orig_page_index, donor_page_index;
621 int unwritten, cur_len;
623 ret = get_ext_path(orig_inode, o_start, &path);
626 ex = path[path->p_depth].p_ext;
627 next_blk = ext4_ext_next_allocated_block(path);
628 cur_blk = le32_to_cpu(ex->ee_block);
629 cur_len = ext4_ext_get_actual_len(ex);
630 /* Check hole before the start pos */
631 if (cur_blk + cur_len - 1 < o_start) {
632 if (next_blk == EXT_MAX_BLOCKS) {
637 d_start += next_blk - o_start;
640 /* Check hole after the start pos */
641 } else if (cur_blk > o_start) {
643 d_start += cur_blk - o_start;
645 /* Extent inside requested range ?*/
646 if (cur_blk >= o_end)
648 } else { /* in_range(o_start, o_blk, o_len) */
649 cur_len += cur_blk - o_start;
651 unwritten = ext4_ext_is_unwritten(ex);
652 if (o_end - o_start < cur_len)
653 cur_len = o_end - o_start;
655 orig_page_index = o_start >> (PAGE_SHIFT -
656 orig_inode->i_blkbits);
657 donor_page_index = d_start >> (PAGE_SHIFT -
658 donor_inode->i_blkbits);
659 offset_in_page = o_start % blocks_per_page;
660 if (cur_len > blocks_per_page- offset_in_page)
661 cur_len = blocks_per_page - offset_in_page;
663 * Up semaphore to avoid following problems:
664 * a. transaction deadlock among ext4_journal_start,
665 * ->write_begin via pagefault, and jbd2_journal_commit
666 * b. racing with ->readpage, ->write_begin, and ext4_get_block
667 * in move_extent_per_page
669 ext4_double_up_write_data_sem(orig_inode, donor_inode);
670 /* Swap original branches with new branches */
671 move_extent_per_page(o_filp, donor_inode,
672 orig_page_index, donor_page_index,
673 offset_in_page, cur_len,
675 ext4_double_down_write_data_sem(orig_inode, donor_inode);
681 *moved_len = o_start - orig_blk;
682 if (*moved_len > len)
687 ext4_discard_preallocations(orig_inode);
688 ext4_discard_preallocations(donor_inode);
691 ext4_ext_drop_refs(path);
693 ext4_double_up_write_data_sem(orig_inode, donor_inode);
694 unlock_two_nondirectories(orig_inode, donor_inode);