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 head = create_empty_buffers(folio, blocksize, 0);
187
188 block = (sector_t)folio->index << (PAGE_SHIFT - inode->i_blkbits);
189 for (bh = head, block_start = 0; bh != head || !block_start;
190 block++, block_start = block_end, bh = bh->b_this_page) {
191 block_end = block_start + blocksize;
192 if (block_end <= from || block_start >= to) {
193 if (!buffer_uptodate(bh))
194 partial = 1;
195 continue;
196 }
197 if (buffer_uptodate(bh))
198 continue;
199 if (!buffer_mapped(bh)) {
200 err = ext4_get_block(inode, block, bh, 0);
201 if (err)
202 return err;
203 if (!buffer_mapped(bh)) {
204 folio_zero_range(folio, block_start, blocksize);
205 set_buffer_uptodate(bh);
206 continue;
207 }
208 }
209 BUG_ON(nr >= MAX_BUF_PER_PAGE);
210 arr[nr++] = bh;
211 }
212 /* No io required */
213 if (!nr)
214 goto out;
215
216 for (i = 0; i < nr; i++) {
217 bh = arr[i];
218 if (!bh_uptodate_or_lock(bh)) {
219 err = ext4_read_bh(bh, 0, NULL);
220 if (err)
221 return err;
222 }
223 }
224 out:
225 if (!partial)
226 folio_mark_uptodate(folio);
227 return 0;
228 }
229
230 /**
231 * move_extent_per_page - Move extent data per page
232 *
233 * @o_filp: file structure of original file
234 * @donor_inode: donor inode
235 * @orig_page_offset: page index on original file
236 * @donor_page_offset: page index on donor file
237 * @data_offset_in_page: block index where data swapping starts
238 * @block_len_in_page: the number of blocks to be swapped
239 * @unwritten: orig extent is unwritten or not
240 * @err: pointer to save return value
241 *
242 * Save the data in original inode blocks and replace original inode extents
243 * with donor inode extents by calling ext4_swap_extents().
244 * Finally, write out the saved data in new original inode blocks. Return
245 * replaced block count.
246 */
247 static int
248 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
249 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
250 int data_offset_in_page,
251 int block_len_in_page, int unwritten, int *err)
252 {
253 struct inode *orig_inode = file_inode(o_filp);
254 struct folio *folio[2] = {NULL, NULL};
255 handle_t *handle;
256 ext4_lblk_t orig_blk_offset, donor_blk_offset;
257 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
258 unsigned int tmp_data_size, data_size, replaced_size;
259 int i, err2, jblocks, retries = 0;
260 int replaced_count = 0;
261 int from = data_offset_in_page << orig_inode->i_blkbits;
262 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
263 struct super_block *sb = orig_inode->i_sb;
264 struct buffer_head *bh = NULL;
265
266 /*
267 * It needs twice the amount of ordinary journal buffers because
268 * inode and donor_inode may change each different metadata blocks.
269 */
270 again:
271 *err = 0;
272 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
273 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
274 if (IS_ERR(handle)) {
275 *err = PTR_ERR(handle);
276 return 0;
277 }
278
279 orig_blk_offset = orig_page_offset * blocks_per_page +
280 data_offset_in_page;
281
282 donor_blk_offset = donor_page_offset * blocks_per_page +
283 data_offset_in_page;
284
285 /* Calculate data_size */
286 if ((orig_blk_offset + block_len_in_page - 1) ==
287 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
288 /* Replace the last block */
289 tmp_data_size = orig_inode->i_size & (blocksize - 1);
290 /*
291 * If data_size equal zero, it shows data_size is multiples of
292 * blocksize. So we set appropriate value.
293 */
294 if (tmp_data_size == 0)
295 tmp_data_size = blocksize;
296
297 data_size = tmp_data_size +
298 ((block_len_in_page - 1) << orig_inode->i_blkbits);
299 } else
300 data_size = block_len_in_page << orig_inode->i_blkbits;
301
302 replaced_size = data_size;
303
304 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
305 donor_page_offset, folio);
306 if (unlikely(*err < 0))
307 goto stop_journal;
308 /*
309 * If orig extent was unwritten it can become initialized
310 * at any time after i_data_sem was dropped, in order to
311 * serialize with delalloc we have recheck extent while we
312 * hold page's lock, if it is still the case data copy is not
313 * necessary, just swap data blocks between orig and donor.
314 */
315
316 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
317 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
318 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
319
320 if (unwritten) {
321 ext4_double_down_write_data_sem(orig_inode, donor_inode);
322 /* If any of extents in range became initialized we have to
323 * fallback to data copying */
324 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
325 block_len_in_page, 1, err);
326 if (*err)
327 goto drop_data_sem;
328
329 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
330 block_len_in_page, 1, err);
331 if (*err)
332 goto drop_data_sem;
333
334 if (!unwritten) {
335 ext4_double_up_write_data_sem(orig_inode, donor_inode);
336 goto data_copy;
337 }
338 if (!filemap_release_folio(folio[0], 0) ||
339 !filemap_release_folio(folio[1], 0)) {
340 *err = -EBUSY;
341 goto drop_data_sem;
342 }
343 replaced_count = ext4_swap_extents(handle, orig_inode,
344 donor_inode, orig_blk_offset,
345 donor_blk_offset,
346 block_len_in_page, 1, err);
347 drop_data_sem:
348 ext4_double_up_write_data_sem(orig_inode, donor_inode);
349 goto unlock_folios;
350 }
351 data_copy:
352 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
353 if (*err)
354 goto unlock_folios;
355
356 /* At this point all buffers in range are uptodate, old mapping layout
357 * is no longer required, try to drop it now. */
358 if (!filemap_release_folio(folio[0], 0) ||
359 !filemap_release_folio(folio[1], 0)) {
360 *err = -EBUSY;
361 goto unlock_folios;
362 }
363 ext4_double_down_write_data_sem(orig_inode, donor_inode);
364 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
365 orig_blk_offset, donor_blk_offset,
366 block_len_in_page, 1, err);
367 ext4_double_up_write_data_sem(orig_inode, donor_inode);
368 if (*err) {
369 if (replaced_count) {
370 block_len_in_page = replaced_count;
371 replaced_size =
372 block_len_in_page << orig_inode->i_blkbits;
373 } else
374 goto unlock_folios;
375 }
376 /* Perform all necessary steps similar write_begin()/write_end()
377 * but keeping in mind that i_size will not change */
378 bh = folio_buffers(folio[0]);
379 if (!bh)
380 bh = create_empty_buffers(folio[0],
381 1 << orig_inode->i_blkbits, 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);
386 if (*err < 0)
387 goto repair_branches;
388 bh = bh->b_this_page;
389 }
390
391 block_commit_write(&folio[0]->page, from, from + replaced_size);
392
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,
396 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
397
398 unlock_folios:
399 folio_unlock(folio[0]);
400 folio_put(folio[0]);
401 folio_unlock(folio[1]);
402 folio_put(folio[1]);
403 stop_journal:
404 ext4_journal_stop(handle);
405 if (*err == -ENOSPC &&
406 ext4_should_retry_alloc(sb, &retries))
407 goto again;
408 /* Buffer was busy because probably is pinned to journal transaction,
409 * force transaction commit may help to free it. */
410 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
411 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
412 goto again;
413 return replaced_count;
414
415 repair_branches:
416 /*
417 * This should never ever happen!
418 * Extents are swapped already, but we are not able to copy data.
419 * Try to swap extents to it's original places
420 */
421 ext4_double_down_write_data_sem(orig_inode, donor_inode);
422 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
423 orig_blk_offset, donor_blk_offset,
424 block_len_in_page, 0, &err2);
425 ext4_double_up_write_data_sem(orig_inode, donor_inode);
426 if (replaced_count != block_len_in_page) {
427 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
428 EIO, "Unable to copy data block,"
429 " data will be lost.");
430 *err = -EIO;
431 }
432 replaced_count = 0;
433 goto unlock_folios;
434 }
435
436 /**
437 * mext_check_arguments - Check whether move extent can be done
438 *
439 * @orig_inode: original inode
440 * @donor_inode: donor inode
441 * @orig_start: logical start offset in block for orig
442 * @donor_start: logical start offset in block for donor
443 * @len: the number of blocks to be moved
444 *
445 * Check the arguments of ext4_move_extents() whether the files can be
446 * exchanged with each other.
447 * Return 0 on success, or a negative error value on failure.
448 */
449 static int
450 mext_check_arguments(struct inode *orig_inode,
451 struct inode *donor_inode, __u64 orig_start,
452 __u64 donor_start, __u64 *len)
453 {
454 __u64 orig_eof, donor_eof;
455 unsigned int blkbits = orig_inode->i_blkbits;
456 unsigned int blocksize = 1 << blkbits;
457
458 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
459 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
460
461
462 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
463 ext4_debug("ext4 move extent: suid or sgid is set"
464 " to donor file [ino:orig %lu, donor %lu]\n",
465 orig_inode->i_ino, donor_inode->i_ino);
466 return -EINVAL;
467 }
468
469 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
470 return -EPERM;
471
472 /* Ext4 move extent does not support swap files */
473 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
474 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
475 orig_inode->i_ino, donor_inode->i_ino);
476 return -ETXTBSY;
477 }
478
479 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
480 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
481 orig_inode->i_ino, donor_inode->i_ino);
482 return -EOPNOTSUPP;
483 }
484
485 /* Ext4 move extent supports only extent based file */
486 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
487 ext4_debug("ext4 move extent: orig file is not extents "
488 "based file [ino:orig %lu]\n", orig_inode->i_ino);
489 return -EOPNOTSUPP;
490 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
491 ext4_debug("ext4 move extent: donor file is not extents "
492 "based file [ino:donor %lu]\n", donor_inode->i_ino);
493 return -EOPNOTSUPP;
494 }
495
496 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
497 ext4_debug("ext4 move extent: File size is 0 byte\n");
498 return -EINVAL;
499 }
500
501 /* Start offset should be same */
502 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
503 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
504 ext4_debug("ext4 move extent: orig and donor's start "
505 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
506 orig_inode->i_ino, donor_inode->i_ino);
507 return -EINVAL;
508 }
509
510 if ((orig_start >= EXT_MAX_BLOCKS) ||
511 (donor_start >= EXT_MAX_BLOCKS) ||
512 (*len > EXT_MAX_BLOCKS) ||
513 (donor_start + *len >= EXT_MAX_BLOCKS) ||
514 (orig_start + *len >= EXT_MAX_BLOCKS)) {
515 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
516 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
517 orig_inode->i_ino, donor_inode->i_ino);
518 return -EINVAL;
519 }
520 if (orig_eof <= orig_start)
521 *len = 0;
522 else if (orig_eof < orig_start + *len - 1)
523 *len = orig_eof - orig_start;
524 if (donor_eof <= donor_start)
525 *len = 0;
526 else if (donor_eof < donor_start + *len - 1)
527 *len = donor_eof - donor_start;
528 if (!*len) {
529 ext4_debug("ext4 move extent: len should not be 0 "
530 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
531 donor_inode->i_ino);
532 return -EINVAL;
533 }
534
535 return 0;
536 }
537
538 /**
539 * ext4_move_extents - Exchange the specified range of a file
540 *
541 * @o_filp: file structure of the original file
542 * @d_filp: file structure of the donor file
543 * @orig_blk: start offset in block for orig
544 * @donor_blk: start offset in block for donor
545 * @len: the number of blocks to be moved
546 * @moved_len: moved block length
547 *
548 * This function returns 0 and moved block length is set in moved_len
549 * if succeed, otherwise returns error value.
550 *
551 */
552 int
553 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
554 __u64 donor_blk, __u64 len, __u64 *moved_len)
555 {
556 struct inode *orig_inode = file_inode(o_filp);
557 struct inode *donor_inode = file_inode(d_filp);
558 struct ext4_ext_path *path = NULL;
559 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
560 ext4_lblk_t o_end, o_start = orig_blk;
561 ext4_lblk_t d_start = donor_blk;
562 int ret;
563
564 if (orig_inode->i_sb != donor_inode->i_sb) {
565 ext4_debug("ext4 move extent: The argument files "
566 "should be in same FS [ino:orig %lu, donor %lu]\n",
567 orig_inode->i_ino, donor_inode->i_ino);
568 return -EINVAL;
569 }
570
571 /* orig and donor should be different inodes */
572 if (orig_inode == donor_inode) {
573 ext4_debug("ext4 move extent: The argument files should not "
574 "be same inode [ino:orig %lu, donor %lu]\n",
575 orig_inode->i_ino, donor_inode->i_ino);
576 return -EINVAL;
577 }
578
579 /* Regular file check */
580 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
581 ext4_debug("ext4 move extent: The argument files should be "
582 "regular file [ino:orig %lu, donor %lu]\n",
583 orig_inode->i_ino, donor_inode->i_ino);
584 return -EINVAL;
585 }
586
587 /* TODO: it's not obvious how to swap blocks for inodes with full
588 journaling enabled */
589 if (ext4_should_journal_data(orig_inode) ||
590 ext4_should_journal_data(donor_inode)) {
591 ext4_msg(orig_inode->i_sb, KERN_ERR,
592 "Online defrag not supported with data journaling");
593 return -EOPNOTSUPP;
594 }
595
596 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(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 *moved_len = 0;
619 while (o_start < o_end) {
620 struct ext4_extent *ex;
621 ext4_lblk_t cur_blk, next_blk;
622 pgoff_t orig_page_index, donor_page_index;
623 int offset_in_page;
624 int unwritten, cur_len;
625
626 ret = get_ext_path(orig_inode, o_start, &path);
627 if (ret)
628 goto out;
629 ex = path[path->p_depth].p_ext;
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 next_blk = ext4_ext_next_allocated_block(path);
635 if (next_blk == EXT_MAX_BLOCKS) {
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 ->read_folio, ->write_begin, and
669 * ext4_get_block 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 *moved_len += 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
684 out:
685 if (*moved_len) {
686 ext4_discard_preallocations(orig_inode);
687 ext4_discard_preallocations(donor_inode);
688 }
689
690 ext4_free_ext_path(path);
691 ext4_double_up_write_data_sem(orig_inode, donor_inode);
692 unlock_two_nondirectories(orig_inode, donor_inode);
693
694 return ret;
695 }
696
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.