xref: /linux/fs/ext4/move_extent.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: LGPL-2.1
2 /*
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>
6  */
7 
8 #include <linux/fs.h>
9 #include <linux/quotaops.h>
10 #include <linux/slab.h>
11 #include <linux/sched/mm.h>
12 #include "ext4_jbd2.h"
13 #include "ext4.h"
14 #include "ext4_extents.h"
15 
16 /**
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  * @path:	pointer to an extent path
21  *
22  * ext4_find_extent wrapper. Return an extent path pointer on success,
23  * or an error pointer on failure.
24  */
25 static inline struct ext4_ext_path *
26 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
27 	     struct ext4_ext_path *path)
28 {
29 	path = ext4_find_extent(inode, lblock, path, EXT4_EX_NOCACHE);
30 	if (IS_ERR(path))
31 		return path;
32 	if (path[ext_depth(inode)].p_ext == NULL) {
33 		ext4_free_ext_path(path);
34 		return ERR_PTR(-ENODATA);
35 	}
36 	return path;
37 }
38 
39 /**
40  * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
41  * @first: inode to be locked
42  * @second: inode to be locked
43  *
44  * Acquire write lock of i_data_sem of the two inodes
45  */
46 void
47 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
48 {
49 	if (first < second) {
50 		down_write(&EXT4_I(first)->i_data_sem);
51 		down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
52 	} else {
53 		down_write(&EXT4_I(second)->i_data_sem);
54 		down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
55 
56 	}
57 }
58 
59 /**
60  * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
61  *
62  * @orig_inode:		original inode structure to be released its lock first
63  * @donor_inode:	donor inode structure to be released its lock second
64  * Release write lock of i_data_sem of two inodes (orig and donor).
65  */
66 void
67 ext4_double_up_write_data_sem(struct inode *orig_inode,
68 			      struct inode *donor_inode)
69 {
70 	up_write(&EXT4_I(orig_inode)->i_data_sem);
71 	up_write(&EXT4_I(donor_inode)->i_data_sem);
72 }
73 
74 /**
75  * mext_check_coverage - Check that all extents in range has the same type
76  *
77  * @inode:		inode in question
78  * @from:		block offset of inode
79  * @count:		block count to be checked
80  * @unwritten:		extents expected to be unwritten
81  * @err:		pointer to save error value
82  *
83  * Return 1 if all extents in range has expected type, and zero otherwise.
84  */
85 static int
86 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
87 		    int unwritten, int *err)
88 {
89 	struct ext4_ext_path *path = NULL;
90 	struct ext4_extent *ext;
91 	int ret = 0;
92 	ext4_lblk_t last = from + count;
93 	while (from < last) {
94 		path = get_ext_path(inode, from, path);
95 		if (IS_ERR(path)) {
96 			*err = PTR_ERR(path);
97 			return ret;
98 		}
99 		ext = path[ext_depth(inode)].p_ext;
100 		if (unwritten != ext4_ext_is_unwritten(ext))
101 			goto out;
102 		from += ext4_ext_get_actual_len(ext);
103 	}
104 	ret = 1;
105 out:
106 	ext4_free_ext_path(path);
107 	return ret;
108 }
109 
110 /**
111  * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2
112  *
113  * @inode1:	the inode structure
114  * @inode2:	the inode structure
115  * @index1:	folio index
116  * @index2:	folio index
117  * @folio:	result folio vector
118  *
119  * Grab two locked folio for inode's by inode order
120  */
121 static int
122 mext_folio_double_lock(struct inode *inode1, struct inode *inode2,
123 		      pgoff_t index1, pgoff_t index2, struct folio *folio[2])
124 {
125 	struct address_space *mapping[2];
126 	unsigned int flags;
127 
128 	BUG_ON(!inode1 || !inode2);
129 	if (inode1 < inode2) {
130 		mapping[0] = inode1->i_mapping;
131 		mapping[1] = inode2->i_mapping;
132 	} else {
133 		swap(index1, index2);
134 		mapping[0] = inode2->i_mapping;
135 		mapping[1] = inode1->i_mapping;
136 	}
137 
138 	flags = memalloc_nofs_save();
139 	folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN,
140 			mapping_gfp_mask(mapping[0]));
141 	if (IS_ERR(folio[0])) {
142 		memalloc_nofs_restore(flags);
143 		return PTR_ERR(folio[0]);
144 	}
145 
146 	folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN,
147 			mapping_gfp_mask(mapping[1]));
148 	memalloc_nofs_restore(flags);
149 	if (IS_ERR(folio[1])) {
150 		folio_unlock(folio[0]);
151 		folio_put(folio[0]);
152 		return PTR_ERR(folio[1]);
153 	}
154 	/*
155 	 * __filemap_get_folio() may not wait on folio's writeback if
156 	 * BDI not demand that. But it is reasonable to be very conservative
157 	 * here and explicitly wait on folio's writeback
158 	 */
159 	folio_wait_writeback(folio[0]);
160 	folio_wait_writeback(folio[1]);
161 	if (inode1 > inode2)
162 		swap(folio[0], folio[1]);
163 
164 	return 0;
165 }
166 
167 /* Force folio buffers uptodate w/o dropping folio's lock */
168 static int mext_page_mkuptodate(struct folio *folio, size_t from, size_t to)
169 {
170 	struct inode *inode = folio->mapping->host;
171 	sector_t block;
172 	struct buffer_head *bh, *head;
173 	unsigned int blocksize, block_start, block_end;
174 	int nr = 0;
175 	bool partial = false;
176 
177 	BUG_ON(!folio_test_locked(folio));
178 	BUG_ON(folio_test_writeback(folio));
179 
180 	if (folio_test_uptodate(folio))
181 		return 0;
182 
183 	blocksize = i_blocksize(inode);
184 	head = folio_buffers(folio);
185 	if (!head)
186 		head = create_empty_buffers(folio, blocksize, 0);
187 
188 	block = folio_pos(folio) >> inode->i_blkbits;
189 	block_end = 0;
190 	bh = head;
191 	do {
192 		block_start = block_end;
193 		block_end = block_start + blocksize;
194 		if (block_end <= from || block_start >= to) {
195 			if (!buffer_uptodate(bh))
196 				partial = true;
197 			continue;
198 		}
199 		if (buffer_uptodate(bh))
200 			continue;
201 		if (!buffer_mapped(bh)) {
202 			int err = ext4_get_block(inode, block, bh, 0);
203 			if (err)
204 				return err;
205 			if (!buffer_mapped(bh)) {
206 				folio_zero_range(folio, block_start, blocksize);
207 				set_buffer_uptodate(bh);
208 				continue;
209 			}
210 		}
211 		lock_buffer(bh);
212 		if (buffer_uptodate(bh)) {
213 			unlock_buffer(bh);
214 			continue;
215 		}
216 		ext4_read_bh_nowait(bh, 0, NULL, false);
217 		nr++;
218 	} while (block++, (bh = bh->b_this_page) != head);
219 
220 	/* No io required */
221 	if (!nr)
222 		goto out;
223 
224 	bh = head;
225 	do {
226 		if (bh_offset(bh) + blocksize <= from)
227 			continue;
228 		if (bh_offset(bh) > to)
229 			break;
230 		wait_on_buffer(bh);
231 		if (buffer_uptodate(bh))
232 			continue;
233 		return -EIO;
234 	} while ((bh = bh->b_this_page) != head);
235 out:
236 	if (!partial)
237 		folio_mark_uptodate(folio);
238 	return 0;
239 }
240 
241 /**
242  * move_extent_per_page - Move extent data per page
243  *
244  * @o_filp:			file structure of original file
245  * @donor_inode:		donor inode
246  * @orig_page_offset:		page index on original file
247  * @donor_page_offset:		page index on donor file
248  * @data_offset_in_page:	block index where data swapping starts
249  * @block_len_in_page:		the number of blocks to be swapped
250  * @unwritten:			orig extent is unwritten or not
251  * @err:			pointer to save return value
252  *
253  * Save the data in original inode blocks and replace original inode extents
254  * with donor inode extents by calling ext4_swap_extents().
255  * Finally, write out the saved data in new original inode blocks. Return
256  * replaced block count.
257  */
258 static int
259 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
260 		     pgoff_t orig_page_offset, pgoff_t donor_page_offset,
261 		     int data_offset_in_page,
262 		     int block_len_in_page, int unwritten, int *err)
263 {
264 	struct inode *orig_inode = file_inode(o_filp);
265 	struct folio *folio[2] = {NULL, NULL};
266 	handle_t *handle;
267 	ext4_lblk_t orig_blk_offset, donor_blk_offset;
268 	unsigned long blocksize = orig_inode->i_sb->s_blocksize;
269 	unsigned int tmp_data_size, data_size, replaced_size;
270 	int i, err2, jblocks, retries = 0;
271 	int replaced_count = 0;
272 	int from = data_offset_in_page << orig_inode->i_blkbits;
273 	int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
274 	struct super_block *sb = orig_inode->i_sb;
275 	struct buffer_head *bh = NULL;
276 
277 	/*
278 	 * It needs twice the amount of ordinary journal buffers because
279 	 * inode and donor_inode may change each different metadata blocks.
280 	 */
281 again:
282 	*err = 0;
283 	jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
284 	handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
285 	if (IS_ERR(handle)) {
286 		*err = PTR_ERR(handle);
287 		return 0;
288 	}
289 
290 	orig_blk_offset = orig_page_offset * blocks_per_page +
291 		data_offset_in_page;
292 
293 	donor_blk_offset = donor_page_offset * blocks_per_page +
294 		data_offset_in_page;
295 
296 	/* Calculate data_size */
297 	if ((orig_blk_offset + block_len_in_page - 1) ==
298 	    ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
299 		/* Replace the last block */
300 		tmp_data_size = orig_inode->i_size & (blocksize - 1);
301 		/*
302 		 * If data_size equal zero, it shows data_size is multiples of
303 		 * blocksize. So we set appropriate value.
304 		 */
305 		if (tmp_data_size == 0)
306 			tmp_data_size = blocksize;
307 
308 		data_size = tmp_data_size +
309 			((block_len_in_page - 1) << orig_inode->i_blkbits);
310 	} else
311 		data_size = block_len_in_page << orig_inode->i_blkbits;
312 
313 	replaced_size = data_size;
314 
315 	*err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
316 				     donor_page_offset, folio);
317 	if (unlikely(*err < 0))
318 		goto stop_journal;
319 	/*
320 	 * If orig extent was unwritten it can become initialized
321 	 * at any time after i_data_sem was dropped, in order to
322 	 * serialize with delalloc we have recheck extent while we
323 	 * hold page's lock, if it is still the case data copy is not
324 	 * necessary, just swap data blocks between orig and donor.
325 	 */
326 
327 	VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
328 	VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
329 	VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
330 
331 	if (unwritten) {
332 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
333 		/* If any of extents in range became initialized we have to
334 		 * fallback to data copying */
335 		unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
336 						block_len_in_page, 1, err);
337 		if (*err)
338 			goto drop_data_sem;
339 
340 		unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
341 						 block_len_in_page, 1, err);
342 		if (*err)
343 			goto drop_data_sem;
344 
345 		if (!unwritten) {
346 			ext4_double_up_write_data_sem(orig_inode, donor_inode);
347 			goto data_copy;
348 		}
349 		if (!filemap_release_folio(folio[0], 0) ||
350 		    !filemap_release_folio(folio[1], 0)) {
351 			*err = -EBUSY;
352 			goto drop_data_sem;
353 		}
354 		replaced_count = ext4_swap_extents(handle, orig_inode,
355 						   donor_inode, orig_blk_offset,
356 						   donor_blk_offset,
357 						   block_len_in_page, 1, err);
358 	drop_data_sem:
359 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
360 		goto unlock_folios;
361 	}
362 data_copy:
363 	*err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
364 	if (*err)
365 		goto unlock_folios;
366 
367 	/* At this point all buffers in range are uptodate, old mapping layout
368 	 * is no longer required, try to drop it now. */
369 	if (!filemap_release_folio(folio[0], 0) ||
370 	    !filemap_release_folio(folio[1], 0)) {
371 		*err = -EBUSY;
372 		goto unlock_folios;
373 	}
374 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
375 	replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
376 					       orig_blk_offset, donor_blk_offset,
377 					   block_len_in_page, 1, err);
378 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
379 	if (*err) {
380 		if (replaced_count) {
381 			block_len_in_page = replaced_count;
382 			replaced_size =
383 				block_len_in_page << orig_inode->i_blkbits;
384 		} else
385 			goto unlock_folios;
386 	}
387 	/* Perform all necessary steps similar write_begin()/write_end()
388 	 * but keeping in mind that i_size will not change */
389 	bh = folio_buffers(folio[0]);
390 	if (!bh)
391 		bh = create_empty_buffers(folio[0],
392 				1 << orig_inode->i_blkbits, 0);
393 	for (i = 0; i < data_offset_in_page; i++)
394 		bh = bh->b_this_page;
395 	for (i = 0; i < block_len_in_page; i++) {
396 		*err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
397 		if (*err < 0)
398 			goto repair_branches;
399 		bh = bh->b_this_page;
400 	}
401 
402 	block_commit_write(&folio[0]->page, from, from + replaced_size);
403 
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);
408 
409 unlock_folios:
410 	folio_unlock(folio[0]);
411 	folio_put(folio[0]);
412 	folio_unlock(folio[1]);
413 	folio_put(folio[1]);
414 stop_journal:
415 	ext4_journal_stop(handle);
416 	if (*err == -ENOSPC &&
417 	    ext4_should_retry_alloc(sb, &retries))
418 		goto again;
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))
423 		goto again;
424 	return replaced_count;
425 
426 repair_branches:
427 	/*
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
431 	 */
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.");
441 		*err = -EIO;
442 	}
443 	replaced_count = 0;
444 	goto unlock_folios;
445 }
446 
447 /**
448  * mext_check_arguments - Check whether move extent can be done
449  *
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
455  *
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.
459  */
460 static int
461 mext_check_arguments(struct inode *orig_inode,
462 		     struct inode *donor_inode, __u64 orig_start,
463 		     __u64 donor_start, __u64 *len)
464 {
465 	__u64 orig_eof, donor_eof;
466 	unsigned int blkbits = orig_inode->i_blkbits;
467 	unsigned int blocksize = 1 << blkbits;
468 
469 	orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
470 	donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
471 
472 
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);
477 		return -EINVAL;
478 	}
479 
480 	if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
481 		return -EPERM;
482 
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);
487 		return -ETXTBSY;
488 	}
489 
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);
493 		return -EOPNOTSUPP;
494 	}
495 
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);
500 		return -EOPNOTSUPP;
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);
504 		return -EOPNOTSUPP;
505 	}
506 
507 	if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
508 		ext4_debug("ext4 move extent: File size is 0 byte\n");
509 		return -EINVAL;
510 	}
511 
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);
518 		return -EINVAL;
519 	}
520 
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);
529 		return -EINVAL;
530 	}
531 	if (orig_eof <= orig_start)
532 		*len = 0;
533 	else if (orig_eof < orig_start + *len - 1)
534 		*len = orig_eof - orig_start;
535 	if (donor_eof <= donor_start)
536 		*len = 0;
537 	else if (donor_eof < donor_start + *len - 1)
538 		*len = donor_eof - donor_start;
539 	if (!*len) {
540 		ext4_debug("ext4 move extent: len should not be 0 "
541 			"[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
542 			donor_inode->i_ino);
543 		return -EINVAL;
544 	}
545 
546 	return 0;
547 }
548 
549 /**
550  * ext4_move_extents - Exchange the specified range of a file
551  *
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
558  *
559  * This function returns 0 and moved block length is set in moved_len
560  * if succeed, otherwise returns error value.
561  *
562  */
563 int
564 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
565 		  __u64 donor_blk, __u64 len, __u64 *moved_len)
566 {
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;
573 	int ret;
574 
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);
579 		return -EINVAL;
580 	}
581 
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);
587 		return -EINVAL;
588 	}
589 
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);
595 		return -EINVAL;
596 	}
597 
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");
604 		return -EOPNOTSUPP;
605 	}
606 
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");
610 		return -EOPNOTSUPP;
611 	}
612 
613 	/* Protect orig and donor inodes against a truncate */
614 	lock_two_nondirectories(orig_inode, donor_inode);
615 
616 	/* Wait for all existing dio workers */
617 	inode_dio_wait(orig_inode);
618 	inode_dio_wait(donor_inode);
619 
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,
624 				    donor_blk, &len);
625 	if (ret)
626 		goto out;
627 	o_end = o_start + len;
628 
629 	*moved_len = 0;
630 	while (o_start < o_end) {
631 		struct ext4_extent *ex;
632 		ext4_lblk_t cur_blk, next_blk;
633 		pgoff_t orig_page_index, donor_page_index;
634 		int offset_in_page;
635 		int unwritten, cur_len;
636 
637 		path = get_ext_path(orig_inode, o_start, path);
638 		if (IS_ERR(path)) {
639 			ret = PTR_ERR(path);
640 			goto out;
641 		}
642 		ex = path[path->p_depth].p_ext;
643 		cur_blk = le32_to_cpu(ex->ee_block);
644 		cur_len = ext4_ext_get_actual_len(ex);
645 		/* Check hole before the start pos */
646 		if (cur_blk + cur_len - 1 < o_start) {
647 			next_blk = ext4_ext_next_allocated_block(path);
648 			if (next_blk == EXT_MAX_BLOCKS) {
649 				ret = -ENODATA;
650 				goto out;
651 			}
652 			d_start += next_blk - o_start;
653 			o_start = next_blk;
654 			continue;
655 		/* Check hole after the start pos */
656 		} else if (cur_blk > o_start) {
657 			/* Skip hole */
658 			d_start += cur_blk - o_start;
659 			o_start = cur_blk;
660 			/* Extent inside requested range ?*/
661 			if (cur_blk >= o_end)
662 				goto out;
663 		} else { /* in_range(o_start, o_blk, o_len) */
664 			cur_len += cur_blk - o_start;
665 		}
666 		unwritten = ext4_ext_is_unwritten(ex);
667 		if (o_end - o_start < cur_len)
668 			cur_len = o_end - o_start;
669 
670 		orig_page_index = o_start >> (PAGE_SHIFT -
671 					       orig_inode->i_blkbits);
672 		donor_page_index = d_start >> (PAGE_SHIFT -
673 					       donor_inode->i_blkbits);
674 		offset_in_page = o_start % blocks_per_page;
675 		if (cur_len > blocks_per_page - offset_in_page)
676 			cur_len = blocks_per_page - offset_in_page;
677 		/*
678 		 * Up semaphore to avoid following problems:
679 		 * a. transaction deadlock among ext4_journal_start,
680 		 *    ->write_begin via pagefault, and jbd2_journal_commit
681 		 * b. racing with ->read_folio, ->write_begin, and
682 		 *    ext4_get_block in move_extent_per_page
683 		 */
684 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
685 		/* Swap original branches with new branches */
686 		*moved_len += move_extent_per_page(o_filp, donor_inode,
687 				     orig_page_index, donor_page_index,
688 				     offset_in_page, cur_len,
689 				     unwritten, &ret);
690 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
691 		if (ret < 0)
692 			break;
693 		o_start += cur_len;
694 		d_start += cur_len;
695 	}
696 
697 out:
698 	if (*moved_len) {
699 		ext4_discard_preallocations(orig_inode);
700 		ext4_discard_preallocations(donor_inode);
701 	}
702 
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);
706 
707 	return ret;
708 }
709