1 // SPDX-License-Identifier: LGPL-2.1
2 /*
3 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
4 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
5 * Akira Fujita <a-fujita@rs.jp.nec.com>
6 */
7
8 #include <linux/fs.h>
9 #include <linux/quotaops.h>
10 #include <linux/slab.h>
11 #include <linux/sched/mm.h>
12 #include "ext4_jbd2.h"
13 #include "ext4.h"
14 #include "ext4_extents.h"
15
16 /**
17 * get_ext_path() - Find an extent path for designated logical block number.
18 * @inode: inode to be searched
19 * @lblock: logical block number to find an extent path
20 * @path: pointer to an extent path
21 *
22 * ext4_find_extent wrapper. Return an extent path pointer on success,
23 * or an error pointer on failure.
24 */
25 static inline struct ext4_ext_path *
get_ext_path(struct inode * inode,ext4_lblk_t lblock,struct ext4_ext_path * path)26 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
27 struct ext4_ext_path *path)
28 {
29 path = ext4_find_extent(inode, lblock, path, EXT4_EX_NOCACHE);
30 if (IS_ERR(path))
31 return path;
32 if (path[ext_depth(inode)].p_ext == NULL) {
33 ext4_free_ext_path(path);
34 return ERR_PTR(-ENODATA);
35 }
36 return path;
37 }
38
39 /**
40 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
41 * @first: inode to be locked
42 * @second: inode to be locked
43 *
44 * Acquire write lock of i_data_sem of the two inodes
45 */
46 void
ext4_double_down_write_data_sem(struct inode * first,struct inode * second)47 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
48 {
49 if (first < second) {
50 down_write(&EXT4_I(first)->i_data_sem);
51 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
52 } else {
53 down_write(&EXT4_I(second)->i_data_sem);
54 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
55
56 }
57 }
58
59 /**
60 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
61 *
62 * @orig_inode: original inode structure to be released its lock first
63 * @donor_inode: donor inode structure to be released its lock second
64 * Release write lock of i_data_sem of two inodes (orig and donor).
65 */
66 void
ext4_double_up_write_data_sem(struct inode * orig_inode,struct inode * donor_inode)67 ext4_double_up_write_data_sem(struct inode *orig_inode,
68 struct inode *donor_inode)
69 {
70 up_write(&EXT4_I(orig_inode)->i_data_sem);
71 up_write(&EXT4_I(donor_inode)->i_data_sem);
72 }
73
74 /**
75 * mext_check_coverage - Check that all extents in range has the same type
76 *
77 * @inode: inode in question
78 * @from: block offset of inode
79 * @count: block count to be checked
80 * @unwritten: extents expected to be unwritten
81 * @err: pointer to save error value
82 *
83 * Return 1 if all extents in range has expected type, and zero otherwise.
84 */
85 static int
mext_check_coverage(struct inode * inode,ext4_lblk_t from,ext4_lblk_t count,int unwritten,int * err)86 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
87 int unwritten, int *err)
88 {
89 struct ext4_ext_path *path = NULL;
90 struct ext4_extent *ext;
91 int ret = 0;
92 ext4_lblk_t last = from + count;
93 while (from < last) {
94 path = get_ext_path(inode, from, path);
95 if (IS_ERR(path)) {
96 *err = PTR_ERR(path);
97 return ret;
98 }
99 ext = path[ext_depth(inode)].p_ext;
100 if (unwritten != ext4_ext_is_unwritten(ext))
101 goto out;
102 from += ext4_ext_get_actual_len(ext);
103 }
104 ret = 1;
105 out:
106 ext4_free_ext_path(path);
107 return ret;
108 }
109
110 /**
111 * mext_folio_double_lock - Grab and lock folio on both @inode1 and @inode2
112 *
113 * @inode1: the inode structure
114 * @inode2: the inode structure
115 * @index1: folio index
116 * @index2: folio index
117 * @folio: result folio vector
118 *
119 * Grab two locked folio for inode's by inode order
120 */
121 static int
mext_folio_double_lock(struct inode * inode1,struct inode * inode2,pgoff_t index1,pgoff_t index2,struct folio * folio[2])122 mext_folio_double_lock(struct inode *inode1, struct inode *inode2,
123 pgoff_t index1, pgoff_t index2, struct folio *folio[2])
124 {
125 struct address_space *mapping[2];
126 unsigned int flags;
127
128 BUG_ON(!inode1 || !inode2);
129 if (inode1 < inode2) {
130 mapping[0] = inode1->i_mapping;
131 mapping[1] = inode2->i_mapping;
132 } else {
133 swap(index1, index2);
134 mapping[0] = inode2->i_mapping;
135 mapping[1] = inode1->i_mapping;
136 }
137
138 flags = memalloc_nofs_save();
139 folio[0] = __filemap_get_folio(mapping[0], index1, FGP_WRITEBEGIN,
140 mapping_gfp_mask(mapping[0]));
141 if (IS_ERR(folio[0])) {
142 memalloc_nofs_restore(flags);
143 return PTR_ERR(folio[0]);
144 }
145
146 folio[1] = __filemap_get_folio(mapping[1], index2, FGP_WRITEBEGIN,
147 mapping_gfp_mask(mapping[1]));
148 memalloc_nofs_restore(flags);
149 if (IS_ERR(folio[1])) {
150 folio_unlock(folio[0]);
151 folio_put(folio[0]);
152 return PTR_ERR(folio[1]);
153 }
154 /*
155 * __filemap_get_folio() may not wait on folio's writeback if
156 * BDI not demand that. But it is reasonable to be very conservative
157 * here and explicitly wait on folio's writeback
158 */
159 folio_wait_writeback(folio[0]);
160 folio_wait_writeback(folio[1]);
161 if (inode1 > inode2)
162 swap(folio[0], folio[1]);
163
164 return 0;
165 }
166
167 /* Force folio buffers uptodate w/o dropping folio's lock */
mext_page_mkuptodate(struct folio * folio,size_t from,size_t to)168 static int mext_page_mkuptodate(struct folio *folio, size_t from, size_t to)
169 {
170 struct inode *inode = folio->mapping->host;
171 sector_t block;
172 struct buffer_head *bh, *head;
173 unsigned int blocksize, block_start, block_end;
174 int nr = 0;
175 bool partial = false;
176
177 BUG_ON(!folio_test_locked(folio));
178 BUG_ON(folio_test_writeback(folio));
179
180 if (folio_test_uptodate(folio))
181 return 0;
182
183 blocksize = i_blocksize(inode);
184 head = folio_buffers(folio);
185 if (!head)
186 head = create_empty_buffers(folio, blocksize, 0);
187
188 block = folio_pos(folio) >> inode->i_blkbits;
189 block_end = 0;
190 bh = head;
191 do {
192 block_start = block_end;
193 block_end = block_start + blocksize;
194 if (block_end <= from || block_start >= to) {
195 if (!buffer_uptodate(bh))
196 partial = true;
197 continue;
198 }
199 if (buffer_uptodate(bh))
200 continue;
201 if (!buffer_mapped(bh)) {
202 int err = ext4_get_block(inode, block, bh, 0);
203 if (err)
204 return err;
205 if (!buffer_mapped(bh)) {
206 folio_zero_range(folio, block_start, blocksize);
207 set_buffer_uptodate(bh);
208 continue;
209 }
210 }
211 lock_buffer(bh);
212 if (buffer_uptodate(bh)) {
213 unlock_buffer(bh);
214 continue;
215 }
216 ext4_read_bh_nowait(bh, 0, NULL, false);
217 nr++;
218 } while (block++, (bh = bh->b_this_page) != head);
219
220 /* No io required */
221 if (!nr)
222 goto out;
223
224 bh = head;
225 do {
226 if (bh_offset(bh) + blocksize <= from)
227 continue;
228 if (bh_offset(bh) >= to)
229 break;
230 wait_on_buffer(bh);
231 if (buffer_uptodate(bh))
232 continue;
233 return -EIO;
234 } while ((bh = bh->b_this_page) != head);
235 out:
236 if (!partial)
237 folio_mark_uptodate(folio);
238 return 0;
239 }
240
241 /**
242 * move_extent_per_page - Move extent data per page
243 *
244 * @o_filp: file structure of original file
245 * @donor_inode: donor inode
246 * @orig_page_offset: page index on original file
247 * @donor_page_offset: page index on donor file
248 * @data_offset_in_page: block index where data swapping starts
249 * @block_len_in_page: the number of blocks to be swapped
250 * @unwritten: orig extent is unwritten or not
251 * @err: pointer to save return value
252 *
253 * Save the data in original inode blocks and replace original inode extents
254 * with donor inode extents by calling ext4_swap_extents().
255 * Finally, write out the saved data in new original inode blocks. Return
256 * replaced block count.
257 */
258 static int
move_extent_per_page(struct file * o_filp,struct inode * donor_inode,pgoff_t orig_page_offset,pgoff_t donor_page_offset,int data_offset_in_page,int block_len_in_page,int unwritten,int * err)259 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
260 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
261 int data_offset_in_page,
262 int block_len_in_page, int unwritten, int *err)
263 {
264 struct inode *orig_inode = file_inode(o_filp);
265 struct folio *folio[2] = {NULL, NULL};
266 handle_t *handle;
267 ext4_lblk_t orig_blk_offset, donor_blk_offset;
268 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
269 unsigned int tmp_data_size, data_size, replaced_size;
270 int i, err2, jblocks, retries = 0;
271 int replaced_count = 0;
272 int from;
273 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
274 struct super_block *sb = orig_inode->i_sb;
275 struct buffer_head *bh = NULL;
276
277 /*
278 * It needs twice the amount of ordinary journal buffers because
279 * inode and donor_inode may change each different metadata blocks.
280 */
281 again:
282 *err = 0;
283 jblocks = ext4_meta_trans_blocks(orig_inode, block_len_in_page,
284 block_len_in_page) * 2;
285 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
286 if (IS_ERR(handle)) {
287 *err = PTR_ERR(handle);
288 return 0;
289 }
290
291 orig_blk_offset = orig_page_offset * blocks_per_page +
292 data_offset_in_page;
293
294 donor_blk_offset = donor_page_offset * blocks_per_page +
295 data_offset_in_page;
296
297 /* Calculate data_size */
298 if ((orig_blk_offset + block_len_in_page - 1) ==
299 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
300 /* Replace the last block */
301 tmp_data_size = orig_inode->i_size & (blocksize - 1);
302 /*
303 * If data_size equal zero, it shows data_size is multiples of
304 * blocksize. So we set appropriate value.
305 */
306 if (tmp_data_size == 0)
307 tmp_data_size = blocksize;
308
309 data_size = tmp_data_size +
310 ((block_len_in_page - 1) << orig_inode->i_blkbits);
311 } else
312 data_size = block_len_in_page << orig_inode->i_blkbits;
313
314 replaced_size = data_size;
315
316 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
317 donor_page_offset, folio);
318 if (unlikely(*err < 0))
319 goto stop_journal;
320 /*
321 * If orig extent was unwritten it can become initialized
322 * at any time after i_data_sem was dropped, in order to
323 * serialize with delalloc we have recheck extent while we
324 * hold page's lock, if it is still the case data copy is not
325 * necessary, just swap data blocks between orig and donor.
326 */
327 if (unwritten) {
328 ext4_double_down_write_data_sem(orig_inode, donor_inode);
329 /* If any of extents in range became initialized we have to
330 * fallback to data copying */
331 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
332 block_len_in_page, 1, err);
333 if (*err)
334 goto drop_data_sem;
335
336 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
337 block_len_in_page, 1, err);
338 if (*err)
339 goto drop_data_sem;
340
341 if (!unwritten) {
342 ext4_double_up_write_data_sem(orig_inode, donor_inode);
343 goto data_copy;
344 }
345 if (!filemap_release_folio(folio[0], 0) ||
346 !filemap_release_folio(folio[1], 0)) {
347 *err = -EBUSY;
348 goto drop_data_sem;
349 }
350 replaced_count = ext4_swap_extents(handle, orig_inode,
351 donor_inode, orig_blk_offset,
352 donor_blk_offset,
353 block_len_in_page, 1, err);
354 drop_data_sem:
355 ext4_double_up_write_data_sem(orig_inode, donor_inode);
356 goto unlock_folios;
357 }
358 data_copy:
359 from = offset_in_folio(folio[0],
360 orig_blk_offset << orig_inode->i_blkbits);
361 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
362 if (*err)
363 goto unlock_folios;
364
365 /* At this point all buffers in range are uptodate, old mapping layout
366 * is no longer required, try to drop it now. */
367 if (!filemap_release_folio(folio[0], 0) ||
368 !filemap_release_folio(folio[1], 0)) {
369 *err = -EBUSY;
370 goto unlock_folios;
371 }
372 ext4_double_down_write_data_sem(orig_inode, donor_inode);
373 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
374 orig_blk_offset, donor_blk_offset,
375 block_len_in_page, 1, err);
376 ext4_double_up_write_data_sem(orig_inode, donor_inode);
377 if (*err) {
378 if (replaced_count) {
379 block_len_in_page = replaced_count;
380 replaced_size =
381 block_len_in_page << orig_inode->i_blkbits;
382 } else
383 goto unlock_folios;
384 }
385 /* Perform all necessary steps similar write_begin()/write_end()
386 * but keeping in mind that i_size will not change */
387 bh = folio_buffers(folio[0]);
388 if (!bh)
389 bh = create_empty_buffers(folio[0],
390 1 << orig_inode->i_blkbits, 0);
391 for (i = 0; i < from >> orig_inode->i_blkbits; i++)
392 bh = bh->b_this_page;
393 for (i = 0; i < block_len_in_page; i++) {
394 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
395 if (*err < 0)
396 goto repair_branches;
397 bh = bh->b_this_page;
398 }
399
400 block_commit_write(folio[0], from, from + replaced_size);
401
402 /* Even in case of data=writeback it is reasonable to pin
403 * inode to transaction, to prevent unexpected data loss */
404 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
405 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
406
407 unlock_folios:
408 folio_unlock(folio[0]);
409 folio_put(folio[0]);
410 folio_unlock(folio[1]);
411 folio_put(folio[1]);
412 stop_journal:
413 ext4_journal_stop(handle);
414 if (*err == -ENOSPC &&
415 ext4_should_retry_alloc(sb, &retries))
416 goto again;
417 /* Buffer was busy because probably is pinned to journal transaction,
418 * force transaction commit may help to free it. */
419 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
420 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
421 goto again;
422 return replaced_count;
423
424 repair_branches:
425 /*
426 * This should never ever happen!
427 * Extents are swapped already, but we are not able to copy data.
428 * Try to swap extents to it's original places
429 */
430 ext4_double_down_write_data_sem(orig_inode, donor_inode);
431 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
432 orig_blk_offset, donor_blk_offset,
433 block_len_in_page, 0, &err2);
434 ext4_double_up_write_data_sem(orig_inode, donor_inode);
435 if (replaced_count != block_len_in_page) {
436 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
437 EIO, "Unable to copy data block,"
438 " data will be lost.");
439 *err = -EIO;
440 }
441 replaced_count = 0;
442 goto unlock_folios;
443 }
444
445 /**
446 * mext_check_arguments - Check whether move extent can be done
447 *
448 * @orig_inode: original inode
449 * @donor_inode: donor inode
450 * @orig_start: logical start offset in block for orig
451 * @donor_start: logical start offset in block for donor
452 * @len: the number of blocks to be moved
453 *
454 * Check the arguments of ext4_move_extents() whether the files can be
455 * exchanged with each other.
456 * Return 0 on success, or a negative error value on failure.
457 */
458 static int
mext_check_arguments(struct inode * orig_inode,struct inode * donor_inode,__u64 orig_start,__u64 donor_start,__u64 * len)459 mext_check_arguments(struct inode *orig_inode,
460 struct inode *donor_inode, __u64 orig_start,
461 __u64 donor_start, __u64 *len)
462 {
463 __u64 orig_eof, donor_eof;
464 unsigned int blkbits = orig_inode->i_blkbits;
465 unsigned int blocksize = 1 << blkbits;
466
467 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
468 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
469
470
471 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
472 ext4_debug("ext4 move extent: suid or sgid is set"
473 " to donor file [ino:orig %lu, donor %lu]\n",
474 orig_inode->i_ino, donor_inode->i_ino);
475 return -EINVAL;
476 }
477
478 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
479 return -EPERM;
480
481 /* Ext4 move extent does not support swap files */
482 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
483 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
484 orig_inode->i_ino, donor_inode->i_ino);
485 return -ETXTBSY;
486 }
487
488 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
489 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
490 orig_inode->i_ino, donor_inode->i_ino);
491 return -EOPNOTSUPP;
492 }
493
494 /* Ext4 move extent supports only extent based file */
495 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
496 ext4_debug("ext4 move extent: orig file is not extents "
497 "based file [ino:orig %lu]\n", orig_inode->i_ino);
498 return -EOPNOTSUPP;
499 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
500 ext4_debug("ext4 move extent: donor file is not extents "
501 "based file [ino:donor %lu]\n", donor_inode->i_ino);
502 return -EOPNOTSUPP;
503 }
504
505 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
506 ext4_debug("ext4 move extent: File size is 0 byte\n");
507 return -EINVAL;
508 }
509
510 /* Start offset should be same */
511 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
512 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
513 ext4_debug("ext4 move extent: orig and donor's start "
514 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
515 orig_inode->i_ino, donor_inode->i_ino);
516 return -EINVAL;
517 }
518
519 if ((orig_start >= EXT_MAX_BLOCKS) ||
520 (donor_start >= EXT_MAX_BLOCKS) ||
521 (*len > EXT_MAX_BLOCKS) ||
522 (donor_start + *len >= EXT_MAX_BLOCKS) ||
523 (orig_start + *len >= EXT_MAX_BLOCKS)) {
524 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
525 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
526 orig_inode->i_ino, donor_inode->i_ino);
527 return -EINVAL;
528 }
529 if (orig_eof <= orig_start)
530 *len = 0;
531 else if (orig_eof < orig_start + *len - 1)
532 *len = orig_eof - orig_start;
533 if (donor_eof <= donor_start)
534 *len = 0;
535 else if (donor_eof < donor_start + *len - 1)
536 *len = donor_eof - donor_start;
537 if (!*len) {
538 ext4_debug("ext4 move extent: len should not be 0 "
539 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
540 donor_inode->i_ino);
541 return -EINVAL;
542 }
543
544 return 0;
545 }
546
547 /**
548 * ext4_move_extents - Exchange the specified range of a file
549 *
550 * @o_filp: file structure of the original file
551 * @d_filp: file structure of the donor file
552 * @orig_blk: start offset in block for orig
553 * @donor_blk: start offset in block for donor
554 * @len: the number of blocks to be moved
555 * @moved_len: moved block length
556 *
557 * This function returns 0 and moved block length is set in moved_len
558 * if succeed, otherwise returns error value.
559 *
560 */
561 int
ext4_move_extents(struct file * o_filp,struct file * d_filp,__u64 orig_blk,__u64 donor_blk,__u64 len,__u64 * moved_len)562 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
563 __u64 donor_blk, __u64 len, __u64 *moved_len)
564 {
565 struct inode *orig_inode = file_inode(o_filp);
566 struct inode *donor_inode = file_inode(d_filp);
567 struct ext4_ext_path *path = NULL;
568 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
569 ext4_lblk_t o_end, o_start = orig_blk;
570 ext4_lblk_t d_start = donor_blk;
571 int ret;
572
573 if (orig_inode->i_sb != donor_inode->i_sb) {
574 ext4_debug("ext4 move extent: The argument files "
575 "should be in same FS [ino:orig %lu, donor %lu]\n",
576 orig_inode->i_ino, donor_inode->i_ino);
577 return -EINVAL;
578 }
579
580 /* orig and donor should be different inodes */
581 if (orig_inode == donor_inode) {
582 ext4_debug("ext4 move extent: The argument files should not "
583 "be same inode [ino:orig %lu, donor %lu]\n",
584 orig_inode->i_ino, donor_inode->i_ino);
585 return -EINVAL;
586 }
587
588 /* Regular file check */
589 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
590 ext4_debug("ext4 move extent: The argument files should be "
591 "regular file [ino:orig %lu, donor %lu]\n",
592 orig_inode->i_ino, donor_inode->i_ino);
593 return -EINVAL;
594 }
595
596 /* TODO: it's not obvious how to swap blocks for inodes with full
597 journaling enabled */
598 if (ext4_should_journal_data(orig_inode) ||
599 ext4_should_journal_data(donor_inode)) {
600 ext4_msg(orig_inode->i_sb, KERN_ERR,
601 "Online defrag not supported with data journaling");
602 return -EOPNOTSUPP;
603 }
604
605 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
606 ext4_msg(orig_inode->i_sb, KERN_ERR,
607 "Online defrag not supported for encrypted files");
608 return -EOPNOTSUPP;
609 }
610
611 /* Protect orig and donor inodes against a truncate */
612 lock_two_nondirectories(orig_inode, donor_inode);
613
614 /* Wait for all existing dio workers */
615 inode_dio_wait(orig_inode);
616 inode_dio_wait(donor_inode);
617
618 /* Protect extent tree against block allocations via delalloc */
619 ext4_double_down_write_data_sem(orig_inode, donor_inode);
620 /* Check the filesystem environment whether move_extent can be done */
621 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
622 donor_blk, &len);
623 if (ret)
624 goto out;
625 o_end = o_start + len;
626
627 *moved_len = 0;
628 while (o_start < o_end) {
629 struct ext4_extent *ex;
630 ext4_lblk_t cur_blk, next_blk;
631 pgoff_t orig_page_index, donor_page_index;
632 int offset_in_page;
633 int unwritten, cur_len;
634
635 path = get_ext_path(orig_inode, o_start, path);
636 if (IS_ERR(path)) {
637 ret = PTR_ERR(path);
638 goto out;
639 }
640 ex = path[path->p_depth].p_ext;
641 cur_blk = le32_to_cpu(ex->ee_block);
642 cur_len = ext4_ext_get_actual_len(ex);
643 /* Check hole before the start pos */
644 if (cur_blk + cur_len - 1 < o_start) {
645 next_blk = ext4_ext_next_allocated_block(path);
646 if (next_blk == EXT_MAX_BLOCKS) {
647 ret = -ENODATA;
648 goto out;
649 }
650 d_start += next_blk - o_start;
651 o_start = next_blk;
652 continue;
653 /* Check hole after the start pos */
654 } else if (cur_blk > o_start) {
655 /* Skip hole */
656 d_start += cur_blk - o_start;
657 o_start = cur_blk;
658 /* Extent inside requested range ?*/
659 if (cur_blk >= o_end)
660 goto out;
661 } else { /* in_range(o_start, o_blk, o_len) */
662 cur_len += cur_blk - o_start;
663 }
664 unwritten = ext4_ext_is_unwritten(ex);
665 if (o_end - o_start < cur_len)
666 cur_len = o_end - o_start;
667
668 orig_page_index = o_start >> (PAGE_SHIFT -
669 orig_inode->i_blkbits);
670 donor_page_index = d_start >> (PAGE_SHIFT -
671 donor_inode->i_blkbits);
672 offset_in_page = o_start % blocks_per_page;
673 if (cur_len > blocks_per_page - offset_in_page)
674 cur_len = blocks_per_page - offset_in_page;
675 /*
676 * Up semaphore to avoid following problems:
677 * a. transaction deadlock among ext4_journal_start,
678 * ->write_begin via pagefault, and jbd2_journal_commit
679 * b. racing with ->read_folio, ->write_begin, and
680 * ext4_get_block in move_extent_per_page
681 */
682 ext4_double_up_write_data_sem(orig_inode, donor_inode);
683 /* Swap original branches with new branches */
684 *moved_len += move_extent_per_page(o_filp, donor_inode,
685 orig_page_index, donor_page_index,
686 offset_in_page, cur_len,
687 unwritten, &ret);
688 ext4_double_down_write_data_sem(orig_inode, donor_inode);
689 if (ret < 0)
690 break;
691 o_start += cur_len;
692 d_start += cur_len;
693 }
694
695 out:
696 if (*moved_len) {
697 ext4_discard_preallocations(orig_inode);
698 ext4_discard_preallocations(donor_inode);
699 }
700
701 ext4_free_ext_path(path);
702 ext4_double_up_write_data_sem(orig_inode, donor_inode);
703 unlock_two_nondirectories(orig_inode, donor_inode);
704
705 return ret;
706 }
707