xref: /openbmc/linux/fs/ext4/move_extent.c (revision e3d786a3)
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 "ext4_jbd2.h"
12 #include "ext4.h"
13 #include "ext4_extents.h"
14 
15 /**
16  * get_ext_path - Find an extent path for designated logical block number.
17  *
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)
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_ext_drop_refs(path);
36 		kfree(path);
37 		*ppath = NULL;
38 		return -ENODATA;
39 	}
40 	*ppath = path;
41 	return 0;
42 }
43 
44 /**
45  * ext4_double_down_write_data_sem - Acquire two inodes' write lock
46  *                                   of i_data_sem
47  *
48  * Acquire write lock of i_data_sem of the two inodes
49  */
50 void
51 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
52 {
53 	if (first < second) {
54 		down_write(&EXT4_I(first)->i_data_sem);
55 		down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
56 	} else {
57 		down_write(&EXT4_I(second)->i_data_sem);
58 		down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
59 
60 	}
61 }
62 
63 /**
64  * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
65  *
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).
69  */
70 void
71 ext4_double_up_write_data_sem(struct inode *orig_inode,
72 			      struct inode *donor_inode)
73 {
74 	up_write(&EXT4_I(orig_inode)->i_data_sem);
75 	up_write(&EXT4_I(donor_inode)->i_data_sem);
76 }
77 
78 /**
79  * mext_check_coverage - Check that all extents in range has the same type
80  *
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
86  *
87  * Return 1 if all extents in range has expected type, and zero otherwise.
88  */
89 static int
90 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
91 		    int unwritten, int *err)
92 {
93 	struct ext4_ext_path *path = NULL;
94 	struct ext4_extent *ext;
95 	int ret = 0;
96 	ext4_lblk_t last = from + count;
97 	while (from < last) {
98 		*err = get_ext_path(inode, from, &path);
99 		if (*err)
100 			goto out;
101 		ext = path[ext_depth(inode)].p_ext;
102 		if (unwritten != ext4_ext_is_unwritten(ext))
103 			goto out;
104 		from += ext4_ext_get_actual_len(ext);
105 		ext4_ext_drop_refs(path);
106 	}
107 	ret = 1;
108 out:
109 	ext4_ext_drop_refs(path);
110 	kfree(path);
111 	return ret;
112 }
113 
114 /**
115  * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
116  *
117  * @inode1:	the inode structure
118  * @inode2:	the inode structure
119  * @index1:	page index
120  * @index2:	page index
121  * @page:	result page vector
122  *
123  * Grab two locked pages for inode's by inode order
124  */
125 static int
126 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
127 		      pgoff_t index1, pgoff_t index2, struct page *page[2])
128 {
129 	struct address_space *mapping[2];
130 	unsigned fl = AOP_FLAG_NOFS;
131 
132 	BUG_ON(!inode1 || !inode2);
133 	if (inode1 < inode2) {
134 		mapping[0] = inode1->i_mapping;
135 		mapping[1] = inode2->i_mapping;
136 	} else {
137 		swap(index1, index2);
138 		mapping[0] = inode2->i_mapping;
139 		mapping[1] = inode1->i_mapping;
140 	}
141 
142 	page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
143 	if (!page[0])
144 		return -ENOMEM;
145 
146 	page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
147 	if (!page[1]) {
148 		unlock_page(page[0]);
149 		put_page(page[0]);
150 		return -ENOMEM;
151 	}
152 	/*
153 	 * grab_cache_page_write_begin() may not wait on page's writeback if
154 	 * BDI not demand that. But it is reasonable to be very conservative
155 	 * here and explicitly wait on page's writeback
156 	 */
157 	wait_on_page_writeback(page[0]);
158 	wait_on_page_writeback(page[1]);
159 	if (inode1 > inode2)
160 		swap(page[0], page[1]);
161 
162 	return 0;
163 }
164 
165 /* Force page buffers uptodate w/o dropping page's lock */
166 static int
167 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
168 {
169 	struct inode *inode = page->mapping->host;
170 	sector_t block;
171 	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
172 	unsigned int blocksize, block_start, block_end;
173 	int i, err,  nr = 0, partial = 0;
174 	BUG_ON(!PageLocked(page));
175 	BUG_ON(PageWriteback(page));
176 
177 	if (PageUptodate(page))
178 		return 0;
179 
180 	blocksize = i_blocksize(inode);
181 	if (!page_has_buffers(page))
182 		create_empty_buffers(page, blocksize, 0);
183 
184 	head = page_buffers(page);
185 	block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
186 	for (bh = head, block_start = 0; bh != head || !block_start;
187 	     block++, block_start = block_end, bh = bh->b_this_page) {
188 		block_end = block_start + blocksize;
189 		if (block_end <= from || block_start >= to) {
190 			if (!buffer_uptodate(bh))
191 				partial = 1;
192 			continue;
193 		}
194 		if (buffer_uptodate(bh))
195 			continue;
196 		if (!buffer_mapped(bh)) {
197 			err = ext4_get_block(inode, block, bh, 0);
198 			if (err) {
199 				SetPageError(page);
200 				return err;
201 			}
202 			if (!buffer_mapped(bh)) {
203 				zero_user(page, block_start, blocksize);
204 				set_buffer_uptodate(bh);
205 				continue;
206 			}
207 		}
208 		BUG_ON(nr >= MAX_BUF_PER_PAGE);
209 		arr[nr++] = bh;
210 	}
211 	/* No io required */
212 	if (!nr)
213 		goto out;
214 
215 	for (i = 0; i < nr; i++) {
216 		bh = arr[i];
217 		if (!bh_uptodate_or_lock(bh)) {
218 			err = bh_submit_read(bh);
219 			if (err)
220 				return err;
221 		}
222 	}
223 out:
224 	if (!partial)
225 		SetPageUptodate(page);
226 	return 0;
227 }
228 
229 /**
230  * move_extent_per_page - Move extent data per page
231  *
232  * @o_filp:			file structure of original file
233  * @donor_inode:		donor inode
234  * @orig_page_offset:		page index on original file
235  * @donor_page_offset:		page index on donor file
236  * @data_offset_in_page:	block index where data swapping starts
237  * @block_len_in_page:		the number of blocks to be swapped
238  * @unwritten:			orig extent is unwritten or not
239  * @err:			pointer to save return value
240  *
241  * Save the data in original inode blocks and replace original inode extents
242  * with donor inode extents by calling ext4_swap_extents().
243  * Finally, write out the saved data in new original inode blocks. Return
244  * replaced block count.
245  */
246 static int
247 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
248 		     pgoff_t orig_page_offset, pgoff_t donor_page_offset,
249 		     int data_offset_in_page,
250 		     int block_len_in_page, int unwritten, int *err)
251 {
252 	struct inode *orig_inode = file_inode(o_filp);
253 	struct page *pagep[2] = {NULL, NULL};
254 	handle_t *handle;
255 	ext4_lblk_t orig_blk_offset, donor_blk_offset;
256 	unsigned long blocksize = orig_inode->i_sb->s_blocksize;
257 	unsigned int tmp_data_size, data_size, replaced_size;
258 	int i, err2, jblocks, retries = 0;
259 	int replaced_count = 0;
260 	int from = data_offset_in_page << orig_inode->i_blkbits;
261 	int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
262 	struct super_block *sb = orig_inode->i_sb;
263 	struct buffer_head *bh = NULL;
264 
265 	/*
266 	 * It needs twice the amount of ordinary journal buffers because
267 	 * inode and donor_inode may change each different metadata blocks.
268 	 */
269 again:
270 	*err = 0;
271 	jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
272 	handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
273 	if (IS_ERR(handle)) {
274 		*err = PTR_ERR(handle);
275 		return 0;
276 	}
277 
278 	orig_blk_offset = orig_page_offset * blocks_per_page +
279 		data_offset_in_page;
280 
281 	donor_blk_offset = donor_page_offset * blocks_per_page +
282 		data_offset_in_page;
283 
284 	/* Calculate data_size */
285 	if ((orig_blk_offset + block_len_in_page - 1) ==
286 	    ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
287 		/* Replace the last block */
288 		tmp_data_size = orig_inode->i_size & (blocksize - 1);
289 		/*
290 		 * If data_size equal zero, it shows data_size is multiples of
291 		 * blocksize. So we set appropriate value.
292 		 */
293 		if (tmp_data_size == 0)
294 			tmp_data_size = blocksize;
295 
296 		data_size = tmp_data_size +
297 			((block_len_in_page - 1) << orig_inode->i_blkbits);
298 	} else
299 		data_size = block_len_in_page << orig_inode->i_blkbits;
300 
301 	replaced_size = data_size;
302 
303 	*err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
304 				     donor_page_offset, pagep);
305 	if (unlikely(*err < 0))
306 		goto stop_journal;
307 	/*
308 	 * If orig extent was unwritten it can become initialized
309 	 * at any time after i_data_sem was dropped, in order to
310 	 * serialize with delalloc we have recheck extent while we
311 	 * hold page's lock, if it is still the case data copy is not
312 	 * necessary, just swap data blocks between orig and donor.
313 	 */
314 	if (unwritten) {
315 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
316 		/* If any of extents in range became initialized we have to
317 		 * fallback to data copying */
318 		unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
319 						block_len_in_page, 1, err);
320 		if (*err)
321 			goto drop_data_sem;
322 
323 		unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
324 						 block_len_in_page, 1, err);
325 		if (*err)
326 			goto drop_data_sem;
327 
328 		if (!unwritten) {
329 			ext4_double_up_write_data_sem(orig_inode, donor_inode);
330 			goto data_copy;
331 		}
332 		if ((page_has_private(pagep[0]) &&
333 		     !try_to_release_page(pagep[0], 0)) ||
334 		    (page_has_private(pagep[1]) &&
335 		     !try_to_release_page(pagep[1], 0))) {
336 			*err = -EBUSY;
337 			goto drop_data_sem;
338 		}
339 		replaced_count = ext4_swap_extents(handle, orig_inode,
340 						   donor_inode, orig_blk_offset,
341 						   donor_blk_offset,
342 						   block_len_in_page, 1, err);
343 	drop_data_sem:
344 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
345 		goto unlock_pages;
346 	}
347 data_copy:
348 	*err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
349 	if (*err)
350 		goto unlock_pages;
351 
352 	/* At this point all buffers in range are uptodate, old mapping layout
353 	 * is no longer required, try to drop it now. */
354 	if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
355 	    (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
356 		*err = -EBUSY;
357 		goto unlock_pages;
358 	}
359 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
360 	replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
361 					       orig_blk_offset, donor_blk_offset,
362 					   block_len_in_page, 1, err);
363 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
364 	if (*err) {
365 		if (replaced_count) {
366 			block_len_in_page = replaced_count;
367 			replaced_size =
368 				block_len_in_page << orig_inode->i_blkbits;
369 		} else
370 			goto unlock_pages;
371 	}
372 	/* Perform all necessary steps similar write_begin()/write_end()
373 	 * but keeping in mind that i_size will not change */
374 	if (!page_has_buffers(pagep[0]))
375 		create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
376 	bh = page_buffers(pagep[0]);
377 	for (i = 0; i < data_offset_in_page; i++)
378 		bh = bh->b_this_page;
379 	for (i = 0; i < block_len_in_page; i++) {
380 		*err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
381 		if (*err < 0)
382 			break;
383 		bh = bh->b_this_page;
384 	}
385 	if (!*err)
386 		*err = block_commit_write(pagep[0], from, from + replaced_size);
387 
388 	if (unlikely(*err < 0))
389 		goto repair_branches;
390 
391 	/* Even in case of data=writeback it is reasonable to pin
392 	 * inode to transaction, to prevent unexpected data loss */
393 	*err = ext4_jbd2_inode_add_write(handle, orig_inode);
394 
395 unlock_pages:
396 	unlock_page(pagep[0]);
397 	put_page(pagep[0]);
398 	unlock_page(pagep[1]);
399 	put_page(pagep[1]);
400 stop_journal:
401 	ext4_journal_stop(handle);
402 	if (*err == -ENOSPC &&
403 	    ext4_should_retry_alloc(sb, &retries))
404 		goto again;
405 	/* Buffer was busy because probably is pinned to journal transaction,
406 	 * force transaction commit may help to free it. */
407 	if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
408 	    jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
409 		goto again;
410 	return replaced_count;
411 
412 repair_branches:
413 	/*
414 	 * This should never ever happen!
415 	 * Extents are swapped already, but we are not able to copy data.
416 	 * Try to swap extents to it's original places
417 	 */
418 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
419 	replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
420 					       orig_blk_offset, donor_blk_offset,
421 					   block_len_in_page, 0, &err2);
422 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
423 	if (replaced_count != block_len_in_page) {
424 		EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
425 				       "Unable to copy data block,"
426 				       " data will be lost.");
427 		*err = -EIO;
428 	}
429 	replaced_count = 0;
430 	goto unlock_pages;
431 }
432 
433 /**
434  * mext_check_arguments - Check whether move extent can be done
435  *
436  * @orig_inode:		original inode
437  * @donor_inode:	donor inode
438  * @orig_start:		logical start offset in block for orig
439  * @donor_start:	logical start offset in block for donor
440  * @len:		the number of blocks to be moved
441  *
442  * Check the arguments of ext4_move_extents() whether the files can be
443  * exchanged with each other.
444  * Return 0 on success, or a negative error value on failure.
445  */
446 static int
447 mext_check_arguments(struct inode *orig_inode,
448 		     struct inode *donor_inode, __u64 orig_start,
449 		     __u64 donor_start, __u64 *len)
450 {
451 	__u64 orig_eof, donor_eof;
452 	unsigned int blkbits = orig_inode->i_blkbits;
453 	unsigned int blocksize = 1 << blkbits;
454 
455 	orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
456 	donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
457 
458 
459 	if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
460 		ext4_debug("ext4 move extent: suid or sgid is set"
461 			   " to donor file [ino:orig %lu, donor %lu]\n",
462 			   orig_inode->i_ino, donor_inode->i_ino);
463 		return -EINVAL;
464 	}
465 
466 	if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
467 		return -EPERM;
468 
469 	/* Ext4 move extent does not support swapfile */
470 	if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
471 		ext4_debug("ext4 move extent: The argument files should "
472 			"not be swapfile [ino:orig %lu, donor %lu]\n",
473 			orig_inode->i_ino, donor_inode->i_ino);
474 		return -EBUSY;
475 	}
476 
477 	if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
478 		ext4_debug("ext4 move extent: The argument files should "
479 			"not be quota files [ino:orig %lu, donor %lu]\n",
480 			orig_inode->i_ino, donor_inode->i_ino);
481 		return -EBUSY;
482 	}
483 
484 	/* Ext4 move extent supports only extent based file */
485 	if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
486 		ext4_debug("ext4 move extent: orig file is not extents "
487 			"based file [ino:orig %lu]\n", orig_inode->i_ino);
488 		return -EOPNOTSUPP;
489 	} else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
490 		ext4_debug("ext4 move extent: donor file is not extents "
491 			"based file [ino:donor %lu]\n", donor_inode->i_ino);
492 		return -EOPNOTSUPP;
493 	}
494 
495 	if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
496 		ext4_debug("ext4 move extent: File size is 0 byte\n");
497 		return -EINVAL;
498 	}
499 
500 	/* Start offset should be same */
501 	if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
502 	    (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
503 		ext4_debug("ext4 move extent: orig and donor's start "
504 			"offsets are not aligned [ino:orig %lu, donor %lu]\n",
505 			orig_inode->i_ino, donor_inode->i_ino);
506 		return -EINVAL;
507 	}
508 
509 	if ((orig_start >= EXT_MAX_BLOCKS) ||
510 	    (donor_start >= EXT_MAX_BLOCKS) ||
511 	    (*len > EXT_MAX_BLOCKS) ||
512 	    (donor_start + *len >= EXT_MAX_BLOCKS) ||
513 	    (orig_start + *len >= EXT_MAX_BLOCKS))  {
514 		ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
515 			"[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
516 			orig_inode->i_ino, donor_inode->i_ino);
517 		return -EINVAL;
518 	}
519 	if (orig_eof <= orig_start)
520 		*len = 0;
521 	else if (orig_eof < orig_start + *len - 1)
522 		*len = orig_eof - orig_start;
523 	if (donor_eof <= donor_start)
524 		*len = 0;
525 	else if (donor_eof < donor_start + *len - 1)
526 		*len = donor_eof - donor_start;
527 	if (!*len) {
528 		ext4_debug("ext4 move extent: len should not be 0 "
529 			"[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
530 			donor_inode->i_ino);
531 		return -EINVAL;
532 	}
533 
534 	return 0;
535 }
536 
537 /**
538  * ext4_move_extents - Exchange the specified range of a file
539  *
540  * @o_filp:		file structure of the original file
541  * @d_filp:		file structure of the donor file
542  * @orig_blk:		start offset in block for orig
543  * @donor_blk:		start offset in block for donor
544  * @len:		the number of blocks to be moved
545  * @moved_len:		moved block length
546  *
547  * This function returns 0 and moved block length is set in moved_len
548  * if succeed, otherwise returns error value.
549  *
550  */
551 int
552 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
553 		  __u64 donor_blk, __u64 len, __u64 *moved_len)
554 {
555 	struct inode *orig_inode = file_inode(o_filp);
556 	struct inode *donor_inode = file_inode(d_filp);
557 	struct ext4_ext_path *path = NULL;
558 	int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
559 	ext4_lblk_t o_end, o_start = orig_blk;
560 	ext4_lblk_t d_start = donor_blk;
561 	int ret;
562 
563 	if (orig_inode->i_sb != donor_inode->i_sb) {
564 		ext4_debug("ext4 move extent: The argument files "
565 			"should be in same FS [ino:orig %lu, donor %lu]\n",
566 			orig_inode->i_ino, donor_inode->i_ino);
567 		return -EINVAL;
568 	}
569 
570 	/* orig and donor should be different inodes */
571 	if (orig_inode == donor_inode) {
572 		ext4_debug("ext4 move extent: The argument files should not "
573 			"be same inode [ino:orig %lu, donor %lu]\n",
574 			orig_inode->i_ino, donor_inode->i_ino);
575 		return -EINVAL;
576 	}
577 
578 	/* Regular file check */
579 	if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
580 		ext4_debug("ext4 move extent: The argument files should be "
581 			"regular file [ino:orig %lu, donor %lu]\n",
582 			orig_inode->i_ino, donor_inode->i_ino);
583 		return -EINVAL;
584 	}
585 
586 	/* TODO: it's not obvious how to swap blocks for inodes with full
587 	   journaling enabled */
588 	if (ext4_should_journal_data(orig_inode) ||
589 	    ext4_should_journal_data(donor_inode)) {
590 		ext4_msg(orig_inode->i_sb, KERN_ERR,
591 			 "Online defrag not supported with data journaling");
592 		return -EOPNOTSUPP;
593 	}
594 
595 	if (ext4_encrypted_inode(orig_inode) ||
596 	    ext4_encrypted_inode(donor_inode)) {
597 		ext4_msg(orig_inode->i_sb, KERN_ERR,
598 			 "Online defrag not supported for encrypted files");
599 		return -EOPNOTSUPP;
600 	}
601 
602 	/* Protect orig and donor inodes against a truncate */
603 	lock_two_nondirectories(orig_inode, donor_inode);
604 
605 	/* Wait for all existing dio workers */
606 	inode_dio_wait(orig_inode);
607 	inode_dio_wait(donor_inode);
608 
609 	/* Protect extent tree against block allocations via delalloc */
610 	ext4_double_down_write_data_sem(orig_inode, donor_inode);
611 	/* Check the filesystem environment whether move_extent can be done */
612 	ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
613 				    donor_blk, &len);
614 	if (ret)
615 		goto out;
616 	o_end = o_start + len;
617 
618 	while (o_start < o_end) {
619 		struct ext4_extent *ex;
620 		ext4_lblk_t cur_blk, next_blk;
621 		pgoff_t orig_page_index, donor_page_index;
622 		int offset_in_page;
623 		int unwritten, cur_len;
624 
625 		ret = get_ext_path(orig_inode, o_start, &path);
626 		if (ret)
627 			goto out;
628 		ex = path[path->p_depth].p_ext;
629 		next_blk = ext4_ext_next_allocated_block(path);
630 		cur_blk = le32_to_cpu(ex->ee_block);
631 		cur_len = ext4_ext_get_actual_len(ex);
632 		/* Check hole before the start pos */
633 		if (cur_blk + cur_len - 1 < o_start) {
634 			if (next_blk == EXT_MAX_BLOCKS) {
635 				o_start = o_end;
636 				ret = -ENODATA;
637 				goto out;
638 			}
639 			d_start += next_blk - o_start;
640 			o_start = next_blk;
641 			continue;
642 		/* Check hole after the start pos */
643 		} else if (cur_blk > o_start) {
644 			/* Skip hole */
645 			d_start += cur_blk - o_start;
646 			o_start = cur_blk;
647 			/* Extent inside requested range ?*/
648 			if (cur_blk >= o_end)
649 				goto out;
650 		} else { /* in_range(o_start, o_blk, o_len) */
651 			cur_len += cur_blk - o_start;
652 		}
653 		unwritten = ext4_ext_is_unwritten(ex);
654 		if (o_end - o_start < cur_len)
655 			cur_len = o_end - o_start;
656 
657 		orig_page_index = o_start >> (PAGE_SHIFT -
658 					       orig_inode->i_blkbits);
659 		donor_page_index = d_start >> (PAGE_SHIFT -
660 					       donor_inode->i_blkbits);
661 		offset_in_page = o_start % blocks_per_page;
662 		if (cur_len > blocks_per_page- offset_in_page)
663 			cur_len = blocks_per_page - offset_in_page;
664 		/*
665 		 * Up semaphore to avoid following problems:
666 		 * a. transaction deadlock among ext4_journal_start,
667 		 *    ->write_begin via pagefault, and jbd2_journal_commit
668 		 * b. racing with ->readpage, ->write_begin, and ext4_get_block
669 		 *    in move_extent_per_page
670 		 */
671 		ext4_double_up_write_data_sem(orig_inode, donor_inode);
672 		/* Swap original branches with new branches */
673 		move_extent_per_page(o_filp, donor_inode,
674 				     orig_page_index, donor_page_index,
675 				     offset_in_page, cur_len,
676 				     unwritten, &ret);
677 		ext4_double_down_write_data_sem(orig_inode, donor_inode);
678 		if (ret < 0)
679 			break;
680 		o_start += cur_len;
681 		d_start += cur_len;
682 	}
683 	*moved_len = o_start - orig_blk;
684 	if (*moved_len > len)
685 		*moved_len = len;
686 
687 out:
688 	if (*moved_len) {
689 		ext4_discard_preallocations(orig_inode);
690 		ext4_discard_preallocations(donor_inode);
691 	}
692 
693 	ext4_ext_drop_refs(path);
694 	kfree(path);
695 	ext4_double_up_write_data_sem(orig_inode, donor_inode);
696 	unlock_two_nondirectories(orig_inode, donor_inode);
697 
698 	return ret;
699 }
700