xref: /openbmc/linux/fs/ext4/move_extent.c (revision 8f91116f)
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  * @ppath:	pointer to an extent path pointer (for output)
21  *
22  * ext4_find_extent wrapper. Return 0 on success, or a negative error value
23  * on failure.
24  */
25 static inline int
26 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
27 		struct ext4_ext_path **ppath)
28 {
29 	struct ext4_ext_path *path;
30 
31 	path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
32 	if (IS_ERR(path))
33 		return PTR_ERR(path);
34 	if (path[ext_depth(inode)].p_ext == NULL) {
35 		ext4_free_ext_path(path);
36 		*ppath = NULL;
37 		return -ENODATA;
38 	}
39 	return 0;
40 }
41 
42 /**
43  * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
44  * @first: inode to be locked
45  * @second: inode to be locked
46  *
47  * Acquire write lock of i_data_sem of the two inodes
48  */
49 void
50 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
51 {
52 	if (first < second) {
53 		down_write(&EXT4_I(first)->i_data_sem);
54 		down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
55 	} else {
56 		down_write(&EXT4_I(second)->i_data_sem);
57 		down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
58 
59 	}
60 }
61 
62 /**
63  * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
64  *
65  * @orig_inode:		original inode structure to be released its lock first
66  * @donor_inode:	donor inode structure to be released its lock second
67  * Release write lock of i_data_sem of two inodes (orig and donor).
68  */
69 void
70 ext4_double_up_write_data_sem(struct inode *orig_inode,
71 			      struct inode *donor_inode)
72 {
73 	up_write(&EXT4_I(orig_inode)->i_data_sem);
74 	up_write(&EXT4_I(donor_inode)->i_data_sem);
75 }
76 
77 /**
78  * mext_check_coverage - Check that all extents in range has the same type
79  *
80  * @inode:		inode in question
81  * @from:		block offset of inode
82  * @count:		block count to be checked
83  * @unwritten:		extents expected to be unwritten
84  * @err:		pointer to save error value
85  *
86  * Return 1 if all extents in range has expected type, and zero otherwise.
87  */
88 static int
89 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
90 		    int unwritten, int *err)
91 {
92 	struct ext4_ext_path *path = NULL;
93 	struct ext4_extent *ext;
94 	int ret = 0;
95 	ext4_lblk_t last = from + count;
96 	while (from < last) {
97 		*err = get_ext_path(inode, from, &path);
98 		if (*err)
99 			goto out;
100 		ext = path[ext_depth(inode)].p_ext;
101 		if (unwritten != ext4_ext_is_unwritten(ext))
102 			goto out;
103 		from += ext4_ext_get_actual_len(ext);
104 	}
105 	ret = 1;
106 out:
107 	ext4_free_ext_path(path);
108 	return ret;
109 }
110 
111 /**
112  * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2
113  *
114  * @inode1:	the inode structure
115  * @inode2:	the inode structure
116  * @index1:	folio index
117  * @index2:	folio index
118  * @folio:	result folio vector
119  *
120  * Grab two locked folio for inode's by inode order
121  */
122 static int
123 mext_folio_double_lock(struct inode *inode1, struct inode *inode2,
124 		      pgoff_t index1, pgoff_t index2, struct folio *folio[2])
125 {
126 	struct address_space *mapping[2];
127 	unsigned int flags;
128 
129 	BUG_ON(!inode1 || !inode2);
130 	if (inode1 < inode2) {
131 		mapping[0] = inode1->i_mapping;
132 		mapping[1] = inode2->i_mapping;
133 	} else {
134 		swap(index1, index2);
135 		mapping[0] = inode2->i_mapping;
136 		mapping[1] = inode1->i_mapping;
137 	}
138 
139 	flags = memalloc_nofs_save();
140 	folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN,
141 			mapping_gfp_mask(mapping[0]));
142 	if (IS_ERR(folio[0])) {
143 		memalloc_nofs_restore(flags);
144 		return PTR_ERR(folio[0]);
145 	}
146 
147 	folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN,
148 			mapping_gfp_mask(mapping[1]));
149 	memalloc_nofs_restore(flags);
150 	if (IS_ERR(folio[1])) {
151 		folio_unlock(folio[0]);
152 		folio_put(folio[0]);
153 		return PTR_ERR(folio[1]);
154 	}
155 	/*
156 	 * __filemap_get_folio() may not wait on folio's writeback if
157 	 * BDI not demand that. But it is reasonable to be very conservative
158 	 * here and explicitly wait on folio's writeback
159 	 */
160 	folio_wait_writeback(folio[0]);
161 	folio_wait_writeback(folio[1]);
162 	if (inode1 > inode2)
163 		swap(folio[0], folio[1]);
164 
165 	return 0;
166 }
167 
168 /* Force page buffers uptodate w/o dropping page's lock */
169 static int
170 mext_page_mkuptodate(struct folio *folio, unsigned from, unsigned to)
171 {
172 	struct inode *inode = folio->mapping->host;
173 	sector_t block;
174 	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
175 	unsigned int blocksize, block_start, block_end;
176 	int i, err,  nr = 0, partial = 0;
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 		create_empty_buffers(&folio->page, blocksize, 0);
187 		head = folio_buffers(folio);
188 	}
189 
190 	block = (sector_t)folio->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))
196 				partial = 1;
197 			continue;
198 		}
199 		if (buffer_uptodate(bh))
200 			continue;
201 		if (!buffer_mapped(bh)) {
202 			err = ext4_get_block(inode, block, bh, 0);
203 			if (err) {
204 				folio_set_error(folio);
205 				return err;
206 			}
207 			if (!buffer_mapped(bh)) {
208 				folio_zero_range(folio, block_start, blocksize);
209 				set_buffer_uptodate(bh);
210 				continue;
211 			}
212 		}
213 		BUG_ON(nr >= MAX_BUF_PER_PAGE);
214 		arr[nr++] = bh;
215 	}
216 	/* No io required */
217 	if (!nr)
218 		goto out;
219 
220 	for (i = 0; i < nr; i++) {
221 		bh = arr[i];
222 		if (!bh_uptodate_or_lock(bh)) {
223 			err = ext4_read_bh(bh, 0, NULL);
224 			if (err)
225 				return err;
226 		}
227 	}
228 out:
229 	if (!partial)
230 		folio_mark_uptodate(folio);
231 	return 0;
232 }
233 
234 /**
235  * move_extent_per_page - Move extent data per page
236  *
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
245  *
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.
250  */
251 static int
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)
256 {
257 	struct inode *orig_inode = file_inode(o_filp);
258 	struct folio *folio[2] = {NULL, NULL};
259 	handle_t *handle;
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;
269 
270 	/*
271 	 * It needs twice the amount of ordinary journal buffers because
272 	 * inode and donor_inode may change each different metadata blocks.
273 	 */
274 again:
275 	*err = 0;
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);
280 		return 0;
281 	}
282 
283 	orig_blk_offset = orig_page_offset * blocks_per_page +
284 		data_offset_in_page;
285 
286 	donor_blk_offset = donor_page_offset * blocks_per_page +
287 		data_offset_in_page;
288 
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);
294 		/*
295 		 * If data_size equal zero, it shows data_size is multiples of
296 		 * blocksize. So we set appropriate value.
297 		 */
298 		if (tmp_data_size == 0)
299 			tmp_data_size = blocksize;
300 
301 		data_size = tmp_data_size +
302 			((block_len_in_page - 1) << orig_inode->i_blkbits);
303 	} else
304 		data_size = block_len_in_page << orig_inode->i_blkbits;
305 
306 	replaced_size = data_size;
307 
308 	*err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
309 				     donor_page_offset, folio);
310 	if (unlikely(*err < 0))
311 		goto stop_journal;
312 	/*
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.
318 	 */
319 
320 	VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
321 	VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
322 	VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
323 
324 	if (unwritten) {
325 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
326 		/* If any of extents in range became initialized we have to
327 		 * fallback to data copying */
328 		unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
329 						block_len_in_page, 1, err);
330 		if (*err)
331 			goto drop_data_sem;
332 
333 		unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
334 						 block_len_in_page, 1, err);
335 		if (*err)
336 			goto drop_data_sem;
337 
338 		if (!unwritten) {
339 			ext4_double_up_write_data_sem(orig_inode, donor_inode);
340 			goto data_copy;
341 		}
342 		if (!filemap_release_folio(folio[0], 0) ||
343 		    !filemap_release_folio(folio[1], 0)) {
344 			*err = -EBUSY;
345 			goto drop_data_sem;
346 		}
347 		replaced_count = ext4_swap_extents(handle, orig_inode,
348 						   donor_inode, orig_blk_offset,
349 						   donor_blk_offset,
350 						   block_len_in_page, 1, err);
351 	drop_data_sem:
352 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
353 		goto unlock_folios;
354 	}
355 data_copy:
356 	*err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
357 	if (*err)
358 		goto unlock_folios;
359 
360 	/* At this point all buffers in range are uptodate, old mapping layout
361 	 * is no longer required, try to drop it now. */
362 	if (!filemap_release_folio(folio[0], 0) ||
363 	    !filemap_release_folio(folio[1], 0)) {
364 		*err = -EBUSY;
365 		goto unlock_folios;
366 	}
367 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
368 	replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
369 					       orig_blk_offset, donor_blk_offset,
370 					   block_len_in_page, 1, err);
371 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
372 	if (*err) {
373 		if (replaced_count) {
374 			block_len_in_page = replaced_count;
375 			replaced_size =
376 				block_len_in_page << orig_inode->i_blkbits;
377 		} else
378 			goto unlock_folios;
379 	}
380 	/* Perform all necessary steps similar write_begin()/write_end()
381 	 * but keeping in mind that i_size will not change */
382 	if (!folio_buffers(folio[0]))
383 		create_empty_buffers(&folio[0]->page, 1 << orig_inode->i_blkbits, 0);
384 	bh = folio_buffers(folio[0]);
385 	for (i = 0; i < data_offset_in_page; i++)
386 		bh = bh->b_this_page;
387 	for (i = 0; i < block_len_in_page; i++) {
388 		*err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
389 		if (*err < 0)
390 			goto repair_branches;
391 		bh = bh->b_this_page;
392 	}
393 
394 	block_commit_write(&folio[0]->page, from, from + replaced_size);
395 
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);
400 
401 unlock_folios:
402 	folio_unlock(folio[0]);
403 	folio_put(folio[0]);
404 	folio_unlock(folio[1]);
405 	folio_put(folio[1]);
406 stop_journal:
407 	ext4_journal_stop(handle);
408 	if (*err == -ENOSPC &&
409 	    ext4_should_retry_alloc(sb, &retries))
410 		goto again;
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))
415 		goto again;
416 	return replaced_count;
417 
418 repair_branches:
419 	/*
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
423 	 */
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.");
433 		*err = -EIO;
434 	}
435 	replaced_count = 0;
436 	goto unlock_folios;
437 }
438 
439 /**
440  * mext_check_arguments - Check whether move extent can be done
441  *
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
447  *
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.
451  */
452 static int
453 mext_check_arguments(struct inode *orig_inode,
454 		     struct inode *donor_inode, __u64 orig_start,
455 		     __u64 donor_start, __u64 *len)
456 {
457 	__u64 orig_eof, donor_eof;
458 	unsigned int blkbits = orig_inode->i_blkbits;
459 	unsigned int blocksize = 1 << blkbits;
460 
461 	orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
462 	donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
463 
464 
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);
469 		return -EINVAL;
470 	}
471 
472 	if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
473 		return -EPERM;
474 
475 	/* Ext4 move extent does not support swap files */
476 	if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
477 		ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
478 			orig_inode->i_ino, donor_inode->i_ino);
479 		return -ETXTBSY;
480 	}
481 
482 	if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
483 		ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
484 			orig_inode->i_ino, donor_inode->i_ino);
485 		return -EOPNOTSUPP;
486 	}
487 
488 	/* Ext4 move extent supports only extent based file */
489 	if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
490 		ext4_debug("ext4 move extent: orig file is not extents "
491 			"based file [ino:orig %lu]\n", orig_inode->i_ino);
492 		return -EOPNOTSUPP;
493 	} else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
494 		ext4_debug("ext4 move extent: donor file is not extents "
495 			"based file [ino:donor %lu]\n", donor_inode->i_ino);
496 		return -EOPNOTSUPP;
497 	}
498 
499 	if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
500 		ext4_debug("ext4 move extent: File size is 0 byte\n");
501 		return -EINVAL;
502 	}
503 
504 	/* Start offset should be same */
505 	if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
506 	    (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
507 		ext4_debug("ext4 move extent: orig and donor's start "
508 			"offsets are not aligned [ino:orig %lu, donor %lu]\n",
509 			orig_inode->i_ino, donor_inode->i_ino);
510 		return -EINVAL;
511 	}
512 
513 	if ((orig_start >= EXT_MAX_BLOCKS) ||
514 	    (donor_start >= EXT_MAX_BLOCKS) ||
515 	    (*len > EXT_MAX_BLOCKS) ||
516 	    (donor_start + *len >= EXT_MAX_BLOCKS) ||
517 	    (orig_start + *len >= EXT_MAX_BLOCKS))  {
518 		ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
519 			"[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
520 			orig_inode->i_ino, donor_inode->i_ino);
521 		return -EINVAL;
522 	}
523 	if (orig_eof <= orig_start)
524 		*len = 0;
525 	else if (orig_eof < orig_start + *len - 1)
526 		*len = orig_eof - orig_start;
527 	if (donor_eof <= donor_start)
528 		*len = 0;
529 	else if (donor_eof < donor_start + *len - 1)
530 		*len = donor_eof - donor_start;
531 	if (!*len) {
532 		ext4_debug("ext4 move extent: len should not be 0 "
533 			"[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
534 			donor_inode->i_ino);
535 		return -EINVAL;
536 	}
537 
538 	return 0;
539 }
540 
541 /**
542  * ext4_move_extents - Exchange the specified range of a file
543  *
544  * @o_filp:		file structure of the original file
545  * @d_filp:		file structure of the donor file
546  * @orig_blk:		start offset in block for orig
547  * @donor_blk:		start offset in block for donor
548  * @len:		the number of blocks to be moved
549  * @moved_len:		moved block length
550  *
551  * This function returns 0 and moved block length is set in moved_len
552  * if succeed, otherwise returns error value.
553  *
554  */
555 int
556 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
557 		  __u64 donor_blk, __u64 len, __u64 *moved_len)
558 {
559 	struct inode *orig_inode = file_inode(o_filp);
560 	struct inode *donor_inode = file_inode(d_filp);
561 	struct ext4_ext_path *path = NULL;
562 	int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
563 	ext4_lblk_t o_end, o_start = orig_blk;
564 	ext4_lblk_t d_start = donor_blk;
565 	int ret;
566 
567 	if (orig_inode->i_sb != donor_inode->i_sb) {
568 		ext4_debug("ext4 move extent: The argument files "
569 			"should be in same FS [ino:orig %lu, donor %lu]\n",
570 			orig_inode->i_ino, donor_inode->i_ino);
571 		return -EINVAL;
572 	}
573 
574 	/* orig and donor should be different inodes */
575 	if (orig_inode == donor_inode) {
576 		ext4_debug("ext4 move extent: The argument files should not "
577 			"be same inode [ino:orig %lu, donor %lu]\n",
578 			orig_inode->i_ino, donor_inode->i_ino);
579 		return -EINVAL;
580 	}
581 
582 	/* Regular file check */
583 	if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
584 		ext4_debug("ext4 move extent: The argument files should be "
585 			"regular file [ino:orig %lu, donor %lu]\n",
586 			orig_inode->i_ino, donor_inode->i_ino);
587 		return -EINVAL;
588 	}
589 
590 	/* TODO: it's not obvious how to swap blocks for inodes with full
591 	   journaling enabled */
592 	if (ext4_should_journal_data(orig_inode) ||
593 	    ext4_should_journal_data(donor_inode)) {
594 		ext4_msg(orig_inode->i_sb, KERN_ERR,
595 			 "Online defrag not supported with data journaling");
596 		return -EOPNOTSUPP;
597 	}
598 
599 	if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
600 		ext4_msg(orig_inode->i_sb, KERN_ERR,
601 			 "Online defrag not supported for encrypted files");
602 		return -EOPNOTSUPP;
603 	}
604 
605 	/* Protect orig and donor inodes against a truncate */
606 	lock_two_nondirectories(orig_inode, donor_inode);
607 
608 	/* Wait for all existing dio workers */
609 	inode_dio_wait(orig_inode);
610 	inode_dio_wait(donor_inode);
611 
612 	/* Protect extent tree against block allocations via delalloc */
613 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
614 	/* Check the filesystem environment whether move_extent can be done */
615 	ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
616 				    donor_blk, &len);
617 	if (ret)
618 		goto out;
619 	o_end = o_start + len;
620 
621 	*moved_len = 0;
622 	while (o_start < o_end) {
623 		struct ext4_extent *ex;
624 		ext4_lblk_t cur_blk, next_blk;
625 		pgoff_t orig_page_index, donor_page_index;
626 		int offset_in_page;
627 		int unwritten, cur_len;
628 
629 		ret = get_ext_path(orig_inode, o_start, &path);
630 		if (ret)
631 			goto out;
632 		ex = path[path->p_depth].p_ext;
633 		cur_blk = le32_to_cpu(ex->ee_block);
634 		cur_len = ext4_ext_get_actual_len(ex);
635 		/* Check hole before the start pos */
636 		if (cur_blk + cur_len - 1 < o_start) {
637 			next_blk = ext4_ext_next_allocated_block(path);
638 			if (next_blk == EXT_MAX_BLOCKS) {
639 				ret = -ENODATA;
640 				goto out;
641 			}
642 			d_start += next_blk - o_start;
643 			o_start = next_blk;
644 			continue;
645 		/* Check hole after the start pos */
646 		} else if (cur_blk > o_start) {
647 			/* Skip hole */
648 			d_start += cur_blk - o_start;
649 			o_start = cur_blk;
650 			/* Extent inside requested range ?*/
651 			if (cur_blk >= o_end)
652 				goto out;
653 		} else { /* in_range(o_start, o_blk, o_len) */
654 			cur_len += cur_blk - o_start;
655 		}
656 		unwritten = ext4_ext_is_unwritten(ex);
657 		if (o_end - o_start < cur_len)
658 			cur_len = o_end - o_start;
659 
660 		orig_page_index = o_start >> (PAGE_SHIFT -
661 					       orig_inode->i_blkbits);
662 		donor_page_index = d_start >> (PAGE_SHIFT -
663 					       donor_inode->i_blkbits);
664 		offset_in_page = o_start % blocks_per_page;
665 		if (cur_len > blocks_per_page - offset_in_page)
666 			cur_len = blocks_per_page - offset_in_page;
667 		/*
668 		 * Up semaphore to avoid following problems:
669 		 * a. transaction deadlock among ext4_journal_start,
670 		 *    ->write_begin via pagefault, and jbd2_journal_commit
671 		 * b. racing with ->read_folio, ->write_begin, and
672 		 *    ext4_get_block in move_extent_per_page
673 		 */
674 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
675 		/* Swap original branches with new branches */
676 		*moved_len += move_extent_per_page(o_filp, donor_inode,
677 				     orig_page_index, donor_page_index,
678 				     offset_in_page, cur_len,
679 				     unwritten, &ret);
680 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
681 		if (ret < 0)
682 			break;
683 		o_start += cur_len;
684 		d_start += cur_len;
685 	}
686 
687 out:
688 	if (*moved_len) {
689 		ext4_discard_preallocations(orig_inode, 0);
690 		ext4_discard_preallocations(donor_inode, 0);
691 	}
692 
693 	ext4_free_ext_path(path);
694 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
695 	unlock_two_nondirectories(orig_inode, donor_inode);
696 
697 	return ret;
698 }
699