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 *ppath = path;
40 return 0;
41 }
42
43 /**
44 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
45 * @first: inode to be locked
46 * @second: inode to be locked
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 }
106 ret = 1;
107 out:
108 ext4_free_ext_path(path);
109 return ret;
110 }
111
112 /**
113 * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2
114 *
115 * @inode1: the inode structure
116 * @inode2: the inode structure
117 * @index1: folio index
118 * @index2: folio index
119 * @folio: result folio vector
120 *
121 * Grab two locked folio for inode's by inode order
122 */
123 static int
124 mext_folio_double_lock(struct inode *inode1, struct inode *inode2,
125 pgoff_t index1, pgoff_t index2, struct folio *folio[2])
126 {
127 struct address_space *mapping[2];
128 unsigned int flags;
129
130 BUG_ON(!inode1 || !inode2);
131 if (inode1 < inode2) {
132 mapping[0] = inode1->i_mapping;
133 mapping[1] = inode2->i_mapping;
134 } else {
135 swap(index1, index2);
136 mapping[0] = inode2->i_mapping;
137 mapping[1] = inode1->i_mapping;
138 }
139
140 flags = memalloc_nofs_save();
141 folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN,
142 mapping_gfp_mask(mapping[0]));
143 if (IS_ERR(folio[0])) {
144 memalloc_nofs_restore(flags);
145 return PTR_ERR(folio[0]);
146 }
147
148 folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN,
149 mapping_gfp_mask(mapping[1]));
150 memalloc_nofs_restore(flags);
151 if (IS_ERR(folio[1])) {
152 folio_unlock(folio[0]);
153 folio_put(folio[0]);
154 return PTR_ERR(folio[1]);
155 }
156 /*
157 * __filemap_get_folio() may not wait on folio's writeback if
158 * BDI not demand that. But it is reasonable to be very conservative
159 * here and explicitly wait on folio's writeback
160 */
161 folio_wait_writeback(folio[0]);
162 folio_wait_writeback(folio[1]);
163 if (inode1 > inode2)
164 swap(folio[0], folio[1]);
165
166 return 0;
167 }
168
169 /* Force page buffers uptodate w/o dropping page's lock */
170 static int
171 mext_page_mkuptodate(struct folio *folio, unsigned from, unsigned to)
172 {
173 struct inode *inode = folio->mapping->host;
174 sector_t block;
175 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
176 unsigned int blocksize, block_start, block_end;
177 int i, err, nr = 0, partial = 0;
178 BUG_ON(!folio_test_locked(folio));
179 BUG_ON(folio_test_writeback(folio));
180
181 if (folio_test_uptodate(folio))
182 return 0;
183
184 blocksize = i_blocksize(inode);
185 head = folio_buffers(folio);
186 if (!head)
187 head = create_empty_buffers(folio, blocksize, 0);
188
189 block = (sector_t)folio->index << (PAGE_SHIFT - inode->i_blkbits);
190 for (bh = head, block_start = 0; bh != head || !block_start;
191 block++, block_start = block_end, bh = bh->b_this_page) {
192 block_end = block_start + blocksize;
193 if (block_end <= from || block_start >= to) {
194 if (!buffer_uptodate(bh))
195 partial = 1;
196 continue;
197 }
198 if (buffer_uptodate(bh))
199 continue;
200 if (!buffer_mapped(bh)) {
201 err = ext4_get_block(inode, block, bh, 0);
202 if (err)
203 return err;
204 if (!buffer_mapped(bh)) {
205 folio_zero_range(folio, block_start, blocksize);
206 set_buffer_uptodate(bh);
207 continue;
208 }
209 }
210 BUG_ON(nr >= MAX_BUF_PER_PAGE);
211 arr[nr++] = bh;
212 }
213 /* No io required */
214 if (!nr)
215 goto out;
216
217 for (i = 0; i < nr; i++) {
218 bh = arr[i];
219 if (!bh_uptodate_or_lock(bh)) {
220 err = ext4_read_bh(bh, 0, NULL);
221 if (err)
222 return err;
223 }
224 }
225 out:
226 if (!partial)
227 folio_mark_uptodate(folio);
228 return 0;
229 }
230
231 /**
232 * move_extent_per_page - Move extent data per page
233 *
234 * @o_filp: file structure of original file
235 * @donor_inode: donor inode
236 * @orig_page_offset: page index on original file
237 * @donor_page_offset: page index on donor file
238 * @data_offset_in_page: block index where data swapping starts
239 * @block_len_in_page: the number of blocks to be swapped
240 * @unwritten: orig extent is unwritten or not
241 * @err: pointer to save return value
242 *
243 * Save the data in original inode blocks and replace original inode extents
244 * with donor inode extents by calling ext4_swap_extents().
245 * Finally, write out the saved data in new original inode blocks. Return
246 * replaced block count.
247 */
248 static int
249 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
250 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
251 int data_offset_in_page,
252 int block_len_in_page, int unwritten, int *err)
253 {
254 struct inode *orig_inode = file_inode(o_filp);
255 struct folio *folio[2] = {NULL, NULL};
256 handle_t *handle;
257 ext4_lblk_t orig_blk_offset, donor_blk_offset;
258 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
259 unsigned int tmp_data_size, data_size, replaced_size;
260 int i, err2, jblocks, retries = 0;
261 int replaced_count = 0;
262 int from = data_offset_in_page << orig_inode->i_blkbits;
263 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
264 struct super_block *sb = orig_inode->i_sb;
265 struct buffer_head *bh = NULL;
266
267 /*
268 * It needs twice the amount of ordinary journal buffers because
269 * inode and donor_inode may change each different metadata blocks.
270 */
271 again:
272 *err = 0;
273 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
274 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
275 if (IS_ERR(handle)) {
276 *err = PTR_ERR(handle);
277 return 0;
278 }
279
280 orig_blk_offset = orig_page_offset * blocks_per_page +
281 data_offset_in_page;
282
283 donor_blk_offset = donor_page_offset * blocks_per_page +
284 data_offset_in_page;
285
286 /* Calculate data_size */
287 if ((orig_blk_offset + block_len_in_page - 1) ==
288 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
289 /* Replace the last block */
290 tmp_data_size = orig_inode->i_size & (blocksize - 1);
291 /*
292 * If data_size equal zero, it shows data_size is multiples of
293 * blocksize. So we set appropriate value.
294 */
295 if (tmp_data_size == 0)
296 tmp_data_size = blocksize;
297
298 data_size = tmp_data_size +
299 ((block_len_in_page - 1) << orig_inode->i_blkbits);
300 } else
301 data_size = block_len_in_page << orig_inode->i_blkbits;
302
303 replaced_size = data_size;
304
305 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
306 donor_page_offset, folio);
307 if (unlikely(*err < 0))
308 goto stop_journal;
309 /*
310 * If orig extent was unwritten it can become initialized
311 * at any time after i_data_sem was dropped, in order to
312 * serialize with delalloc we have recheck extent while we
313 * hold page's lock, if it is still the case data copy is not
314 * necessary, just swap data blocks between orig and donor.
315 */
316
317 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
318 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
319 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
320
321 if (unwritten) {
322 ext4_double_down_write_data_sem(orig_inode, donor_inode);
323 /* If any of extents in range became initialized we have to
324 * fallback to data copying */
325 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
326 block_len_in_page, 1, err);
327 if (*err)
328 goto drop_data_sem;
329
330 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
331 block_len_in_page, 1, err);
332 if (*err)
333 goto drop_data_sem;
334
335 if (!unwritten) {
336 ext4_double_up_write_data_sem(orig_inode, donor_inode);
337 goto data_copy;
338 }
339 if (!filemap_release_folio(folio[0], 0) ||
340 !filemap_release_folio(folio[1], 0)) {
341 *err = -EBUSY;
342 goto drop_data_sem;
343 }
344 replaced_count = ext4_swap_extents(handle, orig_inode,
345 donor_inode, orig_blk_offset,
346 donor_blk_offset,
347 block_len_in_page, 1, err);
348 drop_data_sem:
349 ext4_double_up_write_data_sem(orig_inode, donor_inode);
350 goto unlock_folios;
351 }
352 data_copy:
353 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
354 if (*err)
355 goto unlock_folios;
356
357 /* At this point all buffers in range are uptodate, old mapping layout
358 * is no longer required, try to drop it now. */
359 if (!filemap_release_folio(folio[0], 0) ||
360 !filemap_release_folio(folio[1], 0)) {
361 *err = -EBUSY;
362 goto unlock_folios;
363 }
364 ext4_double_down_write_data_sem(orig_inode, donor_inode);
365 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
366 orig_blk_offset, donor_blk_offset,
367 block_len_in_page, 1, err);
368 ext4_double_up_write_data_sem(orig_inode, donor_inode);
369 if (*err) {
370 if (replaced_count) {
371 block_len_in_page = replaced_count;
372 replaced_size =
373 block_len_in_page << orig_inode->i_blkbits;
374 } else
375 goto unlock_folios;
376 }
377 /* Perform all necessary steps similar write_begin()/write_end()
378 * but keeping in mind that i_size will not change */
379 bh = folio_buffers(folio[0]);
380 if (!bh)
381 bh = create_empty_buffers(folio[0],
382 1 << orig_inode->i_blkbits, 0);
383 for (i = 0; i < data_offset_in_page; i++)
384 bh = bh->b_this_page;
385 for (i = 0; i < block_len_in_page; i++) {
386 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
387 if (*err < 0)
388 goto repair_branches;
389 bh = bh->b_this_page;
390 }
391
392 block_commit_write(&folio[0]->page, from, from + replaced_size);
393
394 /* Even in case of data=writeback it is reasonable to pin
395 * inode to transaction, to prevent unexpected data loss */
396 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
397 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
398
399 unlock_folios:
400 folio_unlock(folio[0]);
401 folio_put(folio[0]);
402 folio_unlock(folio[1]);
403 folio_put(folio[1]);
404 stop_journal:
405 ext4_journal_stop(handle);
406 if (*err == -ENOSPC &&
407 ext4_should_retry_alloc(sb, &retries))
408 goto again;
409 /* Buffer was busy because probably is pinned to journal transaction,
410 * force transaction commit may help to free it. */
411 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
412 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
413 goto again;
414 return replaced_count;
415
416 repair_branches:
417 /*
418 * This should never ever happen!
419 * Extents are swapped already, but we are not able to copy data.
420 * Try to swap extents to it's original places
421 */
422 ext4_double_down_write_data_sem(orig_inode, donor_inode);
423 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
424 orig_blk_offset, donor_blk_offset,
425 block_len_in_page, 0, &err2);
426 ext4_double_up_write_data_sem(orig_inode, donor_inode);
427 if (replaced_count != block_len_in_page) {
428 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
429 EIO, "Unable to copy data block,"
430 " data will be lost.");
431 *err = -EIO;
432 }
433 replaced_count = 0;
434 goto unlock_folios;
435 }
436
437 /**
438 * mext_check_arguments - Check whether move extent can be done
439 *
440 * @orig_inode: original inode
441 * @donor_inode: donor inode
442 * @orig_start: logical start offset in block for orig
443 * @donor_start: logical start offset in block for donor
444 * @len: the number of blocks to be moved
445 *
446 * Check the arguments of ext4_move_extents() whether the files can be
447 * exchanged with each other.
448 * Return 0 on success, or a negative error value on failure.
449 */
450 static int
451 mext_check_arguments(struct inode *orig_inode,
452 struct inode *donor_inode, __u64 orig_start,
453 __u64 donor_start, __u64 *len)
454 {
455 __u64 orig_eof, donor_eof;
456 unsigned int blkbits = orig_inode->i_blkbits;
457 unsigned int blocksize = 1 << blkbits;
458
459 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
460 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
461
462
463 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
464 ext4_debug("ext4 move extent: suid or sgid is set"
465 " to donor file [ino:orig %lu, donor %lu]\n",
466 orig_inode->i_ino, donor_inode->i_ino);
467 return -EINVAL;
468 }
469
470 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
471 return -EPERM;
472
473 /* Ext4 move extent does not support swap files */
474 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
475 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
476 orig_inode->i_ino, donor_inode->i_ino);
477 return -ETXTBSY;
478 }
479
480 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
481 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
482 orig_inode->i_ino, donor_inode->i_ino);
483 return -EOPNOTSUPP;
484 }
485
486 /* Ext4 move extent supports only extent based file */
487 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
488 ext4_debug("ext4 move extent: orig file is not extents "
489 "based file [ino:orig %lu]\n", orig_inode->i_ino);
490 return -EOPNOTSUPP;
491 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
492 ext4_debug("ext4 move extent: donor file is not extents "
493 "based file [ino:donor %lu]\n", donor_inode->i_ino);
494 return -EOPNOTSUPP;
495 }
496
497 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
498 ext4_debug("ext4 move extent: File size is 0 byte\n");
499 return -EINVAL;
500 }
501
502 /* Start offset should be same */
503 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
504 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
505 ext4_debug("ext4 move extent: orig and donor's start "
506 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
507 orig_inode->i_ino, donor_inode->i_ino);
508 return -EINVAL;
509 }
510
511 if ((orig_start >= EXT_MAX_BLOCKS) ||
512 (donor_start >= EXT_MAX_BLOCKS) ||
513 (*len > EXT_MAX_BLOCKS) ||
514 (donor_start + *len >= EXT_MAX_BLOCKS) ||
515 (orig_start + *len >= EXT_MAX_BLOCKS)) {
516 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
517 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
518 orig_inode->i_ino, donor_inode->i_ino);
519 return -EINVAL;
520 }
521 if (orig_eof <= orig_start)
522 *len = 0;
523 else if (orig_eof < orig_start + *len - 1)
524 *len = orig_eof - orig_start;
525 if (donor_eof <= donor_start)
526 *len = 0;
527 else if (donor_eof < donor_start + *len - 1)
528 *len = donor_eof - donor_start;
529 if (!*len) {
530 ext4_debug("ext4 move extent: len should not be 0 "
531 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
532 donor_inode->i_ino);
533 return -EINVAL;
534 }
535
536 return 0;
537 }
538
539 /**
540 * ext4_move_extents - Exchange the specified range of a file
541 *
542 * @o_filp: file structure of the original file
543 * @d_filp: file structure of the donor file
544 * @orig_blk: start offset in block for orig
545 * @donor_blk: start offset in block for donor
546 * @len: the number of blocks to be moved
547 * @moved_len: moved block length
548 *
549 * This function returns 0 and moved block length is set in moved_len
550 * if succeed, otherwise returns error value.
551 *
552 */
553 int
554 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
555 __u64 donor_blk, __u64 len, __u64 *moved_len)
556 {
557 struct inode *orig_inode = file_inode(o_filp);
558 struct inode *donor_inode = file_inode(d_filp);
559 struct ext4_ext_path *path = NULL;
560 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
561 ext4_lblk_t o_end, o_start = orig_blk;
562 ext4_lblk_t d_start = donor_blk;
563 int ret;
564
565 if (orig_inode->i_sb != donor_inode->i_sb) {
566 ext4_debug("ext4 move extent: The argument files "
567 "should be in same FS [ino:orig %lu, donor %lu]\n",
568 orig_inode->i_ino, donor_inode->i_ino);
569 return -EINVAL;
570 }
571
572 /* orig and donor should be different inodes */
573 if (orig_inode == donor_inode) {
574 ext4_debug("ext4 move extent: The argument files should not "
575 "be same inode [ino:orig %lu, donor %lu]\n",
576 orig_inode->i_ino, donor_inode->i_ino);
577 return -EINVAL;
578 }
579
580 /* Regular file check */
581 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
582 ext4_debug("ext4 move extent: The argument files should be "
583 "regular file [ino:orig %lu, donor %lu]\n",
584 orig_inode->i_ino, donor_inode->i_ino);
585 return -EINVAL;
586 }
587
588 /* TODO: it's not obvious how to swap blocks for inodes with full
589 journaling enabled */
590 if (ext4_should_journal_data(orig_inode) ||
591 ext4_should_journal_data(donor_inode)) {
592 ext4_msg(orig_inode->i_sb, KERN_ERR,
593 "Online defrag not supported with data journaling");
594 return -EOPNOTSUPP;
595 }
596
597 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
598 ext4_msg(orig_inode->i_sb, KERN_ERR,
599 "Online defrag not supported for encrypted files");
600 return -EOPNOTSUPP;
601 }
602
603 /* Protect orig and donor inodes against a truncate */
604 lock_two_nondirectories(orig_inode, donor_inode);
605
606 /* Wait for all existing dio workers */
607 inode_dio_wait(orig_inode);
608 inode_dio_wait(donor_inode);
609
610 /* Protect extent tree against block allocations via delalloc */
611 ext4_double_down_write_data_sem(orig_inode, donor_inode);
612 /* Check the filesystem environment whether move_extent can be done */
613 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
614 donor_blk, &len);
615 if (ret)
616 goto out;
617 o_end = o_start + len;
618
619 *moved_len = 0;
620 while (o_start < o_end) {
621 struct ext4_extent *ex;
622 ext4_lblk_t cur_blk, next_blk;
623 pgoff_t orig_page_index, donor_page_index;
624 int offset_in_page;
625 int unwritten, cur_len;
626
627 ret = get_ext_path(orig_inode, o_start, &path);
628 if (ret)
629 goto out;
630 ex = path[path->p_depth].p_ext;
631 cur_blk = le32_to_cpu(ex->ee_block);
632 cur_len = ext4_ext_get_actual_len(ex);
633 /* Check hole before the start pos */
634 if (cur_blk + cur_len - 1 < o_start) {
635 next_blk = ext4_ext_next_allocated_block(path);
636 if (next_blk == EXT_MAX_BLOCKS) {
637 ret = -ENODATA;
638 goto out;
639 }
640 d_start += next_blk - o_start;
641 o_start = next_blk;
642 continue;
643 /* Check hole after the start pos */
644 } else if (cur_blk > o_start) {
645 /* Skip hole */
646 d_start += cur_blk - o_start;
647 o_start = cur_blk;
648 /* Extent inside requested range ?*/
649 if (cur_blk >= o_end)
650 goto out;
651 } else { /* in_range(o_start, o_blk, o_len) */
652 cur_len += cur_blk - o_start;
653 }
654 unwritten = ext4_ext_is_unwritten(ex);
655 if (o_end - o_start < cur_len)
656 cur_len = o_end - o_start;
657
658 orig_page_index = o_start >> (PAGE_SHIFT -
659 orig_inode->i_blkbits);
660 donor_page_index = d_start >> (PAGE_SHIFT -
661 donor_inode->i_blkbits);
662 offset_in_page = o_start % blocks_per_page;
663 if (cur_len > blocks_per_page - offset_in_page)
664 cur_len = blocks_per_page - offset_in_page;
665 /*
666 * Up semaphore to avoid following problems:
667 * a. transaction deadlock among ext4_journal_start,
668 * ->write_begin via pagefault, and jbd2_journal_commit
669 * b. racing with ->read_folio, ->write_begin, and
670 * ext4_get_block in move_extent_per_page
671 */
672 ext4_double_up_write_data_sem(orig_inode, donor_inode);
673 /* Swap original branches with new branches */
674 *moved_len += move_extent_per_page(o_filp, donor_inode,
675 orig_page_index, donor_page_index,
676 offset_in_page, cur_len,
677 unwritten, &ret);
678 ext4_double_down_write_data_sem(orig_inode, donor_inode);
679 if (ret < 0)
680 break;
681 o_start += cur_len;
682 d_start += cur_len;
683 }
684
685 out:
686 if (*moved_len) {
687 ext4_discard_preallocations(orig_inode);
688 ext4_discard_preallocations(donor_inode);
689 }
690
691 ext4_free_ext_path(path);
692 ext4_double_up_write_data_sem(orig_inode, donor_inode);
693 unlock_two_nondirectories(orig_inode, donor_inode);
694
695 return ret;
696 }
697
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.