Loading...
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 */
25static inline int
26get_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 */
50void
51ext4_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 */
70void
71ext4_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 */
89static int
90mext_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;
107out:
108 ext4_free_ext_path(path);
109 return ret;
110}
111
112/**
113 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
114 *
115 * @inode1: the inode structure
116 * @inode2: the inode structure
117 * @index1: page index
118 * @index2: page index
119 * @page: result page vector
120 *
121 * Grab two locked pages for inode's by inode order
122 */
123static int
124mext_page_double_lock(struct inode *inode1, struct inode *inode2,
125 pgoff_t index1, pgoff_t index2, struct page *page[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 page[0] = grab_cache_page_write_begin(mapping[0], index1);
142 if (!page[0]) {
143 memalloc_nofs_restore(flags);
144 return -ENOMEM;
145 }
146
147 page[1] = grab_cache_page_write_begin(mapping[1], index2);
148 memalloc_nofs_restore(flags);
149 if (!page[1]) {
150 unlock_page(page[0]);
151 put_page(page[0]);
152 return -ENOMEM;
153 }
154 /*
155 * grab_cache_page_write_begin() may not wait on page's writeback if
156 * BDI not demand that. But it is reasonable to be very conservative
157 * here and explicitly wait on page's writeback
158 */
159 wait_on_page_writeback(page[0]);
160 wait_on_page_writeback(page[1]);
161 if (inode1 > inode2)
162 swap(page[0], page[1]);
163
164 return 0;
165}
166
167/* Force page buffers uptodate w/o dropping page's lock */
168static int
169mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
170{
171 struct inode *inode = page->mapping->host;
172 sector_t block;
173 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
174 unsigned int blocksize, block_start, block_end;
175 int i, err, nr = 0, partial = 0;
176 BUG_ON(!PageLocked(page));
177 BUG_ON(PageWriteback(page));
178
179 if (PageUptodate(page))
180 return 0;
181
182 blocksize = i_blocksize(inode);
183 if (!page_has_buffers(page))
184 create_empty_buffers(page, blocksize, 0);
185
186 head = page_buffers(page);
187 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
188 for (bh = head, block_start = 0; bh != head || !block_start;
189 block++, block_start = block_end, bh = bh->b_this_page) {
190 block_end = block_start + blocksize;
191 if (block_end <= from || block_start >= to) {
192 if (!buffer_uptodate(bh))
193 partial = 1;
194 continue;
195 }
196 if (buffer_uptodate(bh))
197 continue;
198 if (!buffer_mapped(bh)) {
199 err = ext4_get_block(inode, block, bh, 0);
200 if (err) {
201 SetPageError(page);
202 return err;
203 }
204 if (!buffer_mapped(bh)) {
205 zero_user(page, 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 }
225out:
226 if (!partial)
227 SetPageUptodate(page);
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 */
248static int
249move_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 page *pagep[2] = {NULL, NULL};
256 struct folio *folio[2] = {NULL, NULL};
257 handle_t *handle;
258 ext4_lblk_t orig_blk_offset, donor_blk_offset;
259 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
260 unsigned int tmp_data_size, data_size, replaced_size;
261 int i, err2, jblocks, retries = 0;
262 int replaced_count = 0;
263 int from = data_offset_in_page << orig_inode->i_blkbits;
264 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
265 struct super_block *sb = orig_inode->i_sb;
266 struct buffer_head *bh = NULL;
267
268 /*
269 * It needs twice the amount of ordinary journal buffers because
270 * inode and donor_inode may change each different metadata blocks.
271 */
272again:
273 *err = 0;
274 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
275 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
276 if (IS_ERR(handle)) {
277 *err = PTR_ERR(handle);
278 return 0;
279 }
280
281 orig_blk_offset = orig_page_offset * blocks_per_page +
282 data_offset_in_page;
283
284 donor_blk_offset = donor_page_offset * blocks_per_page +
285 data_offset_in_page;
286
287 /* Calculate data_size */
288 if ((orig_blk_offset + block_len_in_page - 1) ==
289 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
290 /* Replace the last block */
291 tmp_data_size = orig_inode->i_size & (blocksize - 1);
292 /*
293 * If data_size equal zero, it shows data_size is multiples of
294 * blocksize. So we set appropriate value.
295 */
296 if (tmp_data_size == 0)
297 tmp_data_size = blocksize;
298
299 data_size = tmp_data_size +
300 ((block_len_in_page - 1) << orig_inode->i_blkbits);
301 } else
302 data_size = block_len_in_page << orig_inode->i_blkbits;
303
304 replaced_size = data_size;
305
306 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
307 donor_page_offset, pagep);
308 if (unlikely(*err < 0))
309 goto stop_journal;
310 /*
311 * If orig extent was unwritten it can become initialized
312 * at any time after i_data_sem was dropped, in order to
313 * serialize with delalloc we have recheck extent while we
314 * hold page's lock, if it is still the case data copy is not
315 * necessary, just swap data blocks between orig and donor.
316 */
317 folio[0] = page_folio(pagep[0]);
318 folio[1] = page_folio(pagep[1]);
319
320 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
321 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
322 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
323
324 if (unwritten) {
325 ext4_double_down_write_data_sem(orig_inode, donor_inode);
326 /* If any of extents in range became initialized we have to
327 * fallback to data copying */
328 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
329 block_len_in_page, 1, err);
330 if (*err)
331 goto drop_data_sem;
332
333 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
334 block_len_in_page, 1, err);
335 if (*err)
336 goto drop_data_sem;
337
338 if (!unwritten) {
339 ext4_double_up_write_data_sem(orig_inode, donor_inode);
340 goto data_copy;
341 }
342 if ((folio_has_private(folio[0]) &&
343 !filemap_release_folio(folio[0], 0)) ||
344 (folio_has_private(folio[1]) &&
345 !filemap_release_folio(folio[1], 0))) {
346 *err = -EBUSY;
347 goto drop_data_sem;
348 }
349 replaced_count = ext4_swap_extents(handle, orig_inode,
350 donor_inode, orig_blk_offset,
351 donor_blk_offset,
352 block_len_in_page, 1, err);
353 drop_data_sem:
354 ext4_double_up_write_data_sem(orig_inode, donor_inode);
355 goto unlock_folios;
356 }
357data_copy:
358 *err = mext_page_mkuptodate(&folio[0]->page, from, from + replaced_size);
359 if (*err)
360 goto unlock_folios;
361
362 /* At this point all buffers in range are uptodate, old mapping layout
363 * is no longer required, try to drop it now. */
364 if ((folio_has_private(folio[0]) &&
365 !filemap_release_folio(folio[0], 0)) ||
366 (folio_has_private(folio[1]) &&
367 !filemap_release_folio(folio[1], 0))) {
368 *err = -EBUSY;
369 goto unlock_folios;
370 }
371 ext4_double_down_write_data_sem(orig_inode, donor_inode);
372 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
373 orig_blk_offset, donor_blk_offset,
374 block_len_in_page, 1, err);
375 ext4_double_up_write_data_sem(orig_inode, donor_inode);
376 if (*err) {
377 if (replaced_count) {
378 block_len_in_page = replaced_count;
379 replaced_size =
380 block_len_in_page << orig_inode->i_blkbits;
381 } else
382 goto unlock_folios;
383 }
384 /* Perform all necessary steps similar write_begin()/write_end()
385 * but keeping in mind that i_size will not change */
386 if (!folio_buffers(folio[0]))
387 create_empty_buffers(&folio[0]->page, 1 << orig_inode->i_blkbits, 0);
388 bh = folio_buffers(folio[0]);
389 for (i = 0; i < data_offset_in_page; i++)
390 bh = bh->b_this_page;
391 for (i = 0; i < block_len_in_page; i++) {
392 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
393 if (*err < 0)
394 break;
395 bh = bh->b_this_page;
396 }
397 if (!*err)
398 *err = block_commit_write(&folio[0]->page, from, from + replaced_size);
399
400 if (unlikely(*err < 0))
401 goto repair_branches;
402
403 /* Even in case of data=writeback it is reasonable to pin
404 * inode to transaction, to prevent unexpected data loss */
405 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
406 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
407
408unlock_folios:
409 folio_unlock(folio[0]);
410 folio_put(folio[0]);
411 folio_unlock(folio[1]);
412 folio_put(folio[1]);
413stop_journal:
414 ext4_journal_stop(handle);
415 if (*err == -ENOSPC &&
416 ext4_should_retry_alloc(sb, &retries))
417 goto again;
418 /* Buffer was busy because probably is pinned to journal transaction,
419 * force transaction commit may help to free it. */
420 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
421 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
422 goto again;
423 return replaced_count;
424
425repair_branches:
426 /*
427 * This should never ever happen!
428 * Extents are swapped already, but we are not able to copy data.
429 * Try to swap extents to it's original places
430 */
431 ext4_double_down_write_data_sem(orig_inode, donor_inode);
432 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
433 orig_blk_offset, donor_blk_offset,
434 block_len_in_page, 0, &err2);
435 ext4_double_up_write_data_sem(orig_inode, donor_inode);
436 if (replaced_count != block_len_in_page) {
437 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
438 EIO, "Unable to copy data block,"
439 " data will be lost.");
440 *err = -EIO;
441 }
442 replaced_count = 0;
443 goto unlock_folios;
444}
445
446/**
447 * mext_check_arguments - Check whether move extent can be done
448 *
449 * @orig_inode: original inode
450 * @donor_inode: donor inode
451 * @orig_start: logical start offset in block for orig
452 * @donor_start: logical start offset in block for donor
453 * @len: the number of blocks to be moved
454 *
455 * Check the arguments of ext4_move_extents() whether the files can be
456 * exchanged with each other.
457 * Return 0 on success, or a negative error value on failure.
458 */
459static int
460mext_check_arguments(struct inode *orig_inode,
461 struct inode *donor_inode, __u64 orig_start,
462 __u64 donor_start, __u64 *len)
463{
464 __u64 orig_eof, donor_eof;
465 unsigned int blkbits = orig_inode->i_blkbits;
466 unsigned int blocksize = 1 << blkbits;
467
468 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
469 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
470
471
472 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
473 ext4_debug("ext4 move extent: suid or sgid is set"
474 " to donor file [ino:orig %lu, donor %lu]\n",
475 orig_inode->i_ino, donor_inode->i_ino);
476 return -EINVAL;
477 }
478
479 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
480 return -EPERM;
481
482 /* Ext4 move extent does not support swap files */
483 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
484 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
485 orig_inode->i_ino, donor_inode->i_ino);
486 return -ETXTBSY;
487 }
488
489 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
490 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
491 orig_inode->i_ino, donor_inode->i_ino);
492 return -EOPNOTSUPP;
493 }
494
495 /* Ext4 move extent supports only extent based file */
496 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
497 ext4_debug("ext4 move extent: orig file is not extents "
498 "based file [ino:orig %lu]\n", orig_inode->i_ino);
499 return -EOPNOTSUPP;
500 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
501 ext4_debug("ext4 move extent: donor file is not extents "
502 "based file [ino:donor %lu]\n", donor_inode->i_ino);
503 return -EOPNOTSUPP;
504 }
505
506 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
507 ext4_debug("ext4 move extent: File size is 0 byte\n");
508 return -EINVAL;
509 }
510
511 /* Start offset should be same */
512 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
513 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
514 ext4_debug("ext4 move extent: orig and donor's start "
515 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
516 orig_inode->i_ino, donor_inode->i_ino);
517 return -EINVAL;
518 }
519
520 if ((orig_start >= EXT_MAX_BLOCKS) ||
521 (donor_start >= EXT_MAX_BLOCKS) ||
522 (*len > EXT_MAX_BLOCKS) ||
523 (donor_start + *len >= EXT_MAX_BLOCKS) ||
524 (orig_start + *len >= EXT_MAX_BLOCKS)) {
525 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
526 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
527 orig_inode->i_ino, donor_inode->i_ino);
528 return -EINVAL;
529 }
530 if (orig_eof <= orig_start)
531 *len = 0;
532 else if (orig_eof < orig_start + *len - 1)
533 *len = orig_eof - orig_start;
534 if (donor_eof <= donor_start)
535 *len = 0;
536 else if (donor_eof < donor_start + *len - 1)
537 *len = donor_eof - donor_start;
538 if (!*len) {
539 ext4_debug("ext4 move extent: len should not be 0 "
540 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
541 donor_inode->i_ino);
542 return -EINVAL;
543 }
544
545 return 0;
546}
547
548/**
549 * ext4_move_extents - Exchange the specified range of a file
550 *
551 * @o_filp: file structure of the original file
552 * @d_filp: file structure of the donor file
553 * @orig_blk: start offset in block for orig
554 * @donor_blk: start offset in block for donor
555 * @len: the number of blocks to be moved
556 * @moved_len: moved block length
557 *
558 * This function returns 0 and moved block length is set in moved_len
559 * if succeed, otherwise returns error value.
560 *
561 */
562int
563ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
564 __u64 donor_blk, __u64 len, __u64 *moved_len)
565{
566 struct inode *orig_inode = file_inode(o_filp);
567 struct inode *donor_inode = file_inode(d_filp);
568 struct ext4_ext_path *path = NULL;
569 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
570 ext4_lblk_t o_end, o_start = orig_blk;
571 ext4_lblk_t d_start = donor_blk;
572 int ret;
573
574 if (orig_inode->i_sb != donor_inode->i_sb) {
575 ext4_debug("ext4 move extent: The argument files "
576 "should be in same FS [ino:orig %lu, donor %lu]\n",
577 orig_inode->i_ino, donor_inode->i_ino);
578 return -EINVAL;
579 }
580
581 /* orig and donor should be different inodes */
582 if (orig_inode == donor_inode) {
583 ext4_debug("ext4 move extent: The argument files should not "
584 "be same inode [ino:orig %lu, donor %lu]\n",
585 orig_inode->i_ino, donor_inode->i_ino);
586 return -EINVAL;
587 }
588
589 /* Regular file check */
590 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
591 ext4_debug("ext4 move extent: The argument files should be "
592 "regular file [ino:orig %lu, donor %lu]\n",
593 orig_inode->i_ino, donor_inode->i_ino);
594 return -EINVAL;
595 }
596
597 /* TODO: it's not obvious how to swap blocks for inodes with full
598 journaling enabled */
599 if (ext4_should_journal_data(orig_inode) ||
600 ext4_should_journal_data(donor_inode)) {
601 ext4_msg(orig_inode->i_sb, KERN_ERR,
602 "Online defrag not supported with data journaling");
603 return -EOPNOTSUPP;
604 }
605
606 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
607 ext4_msg(orig_inode->i_sb, KERN_ERR,
608 "Online defrag not supported for encrypted files");
609 return -EOPNOTSUPP;
610 }
611
612 /* Protect orig and donor inodes against a truncate */
613 lock_two_nondirectories(orig_inode, donor_inode);
614
615 /* Wait for all existing dio workers */
616 inode_dio_wait(orig_inode);
617 inode_dio_wait(donor_inode);
618
619 /* Protect extent tree against block allocations via delalloc */
620 ext4_double_down_write_data_sem(orig_inode, donor_inode);
621 /* Check the filesystem environment whether move_extent can be done */
622 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
623 donor_blk, &len);
624 if (ret)
625 goto out;
626 o_end = o_start + len;
627
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 ret = get_ext_path(orig_inode, o_start, &path);
636 if (ret)
637 goto out;
638 ex = path[path->p_depth].p_ext;
639 cur_blk = le32_to_cpu(ex->ee_block);
640 cur_len = ext4_ext_get_actual_len(ex);
641 /* Check hole before the start pos */
642 if (cur_blk + cur_len - 1 < o_start) {
643 next_blk = ext4_ext_next_allocated_block(path);
644 if (next_blk == EXT_MAX_BLOCKS) {
645 ret = -ENODATA;
646 goto out;
647 }
648 d_start += next_blk - o_start;
649 o_start = next_blk;
650 continue;
651 /* Check hole after the start pos */
652 } else if (cur_blk > o_start) {
653 /* Skip hole */
654 d_start += cur_blk - o_start;
655 o_start = cur_blk;
656 /* Extent inside requested range ?*/
657 if (cur_blk >= o_end)
658 goto out;
659 } else { /* in_range(o_start, o_blk, o_len) */
660 cur_len += cur_blk - o_start;
661 }
662 unwritten = ext4_ext_is_unwritten(ex);
663 if (o_end - o_start < cur_len)
664 cur_len = o_end - o_start;
665
666 orig_page_index = o_start >> (PAGE_SHIFT -
667 orig_inode->i_blkbits);
668 donor_page_index = d_start >> (PAGE_SHIFT -
669 donor_inode->i_blkbits);
670 offset_in_page = o_start % blocks_per_page;
671 if (cur_len > blocks_per_page - offset_in_page)
672 cur_len = blocks_per_page - offset_in_page;
673 /*
674 * Up semaphore to avoid following problems:
675 * a. transaction deadlock among ext4_journal_start,
676 * ->write_begin via pagefault, and jbd2_journal_commit
677 * b. racing with ->read_folio, ->write_begin, and
678 * ext4_get_block in move_extent_per_page
679 */
680 ext4_double_up_write_data_sem(orig_inode, donor_inode);
681 /* Swap original branches with new branches */
682 move_extent_per_page(o_filp, donor_inode,
683 orig_page_index, donor_page_index,
684 offset_in_page, cur_len,
685 unwritten, &ret);
686 ext4_double_down_write_data_sem(orig_inode, donor_inode);
687 if (ret < 0)
688 break;
689 o_start += cur_len;
690 d_start += cur_len;
691 }
692 *moved_len = o_start - orig_blk;
693 if (*moved_len > len)
694 *moved_len = len;
695
696out:
697 if (*moved_len) {
698 ext4_discard_preallocations(orig_inode, 0);
699 ext4_discard_preallocations(donor_inode, 0);
700 }
701
702 ext4_free_ext_path(path);
703 ext4_double_up_write_data_sem(orig_inode, donor_inode);
704 unlock_two_nondirectories(orig_inode, donor_inode);
705
706 return ret;
707}
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 */
25static inline int
26get_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 */
50void
51ext4_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 */
70void
71ext4_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 */
89static int
90mext_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;
107out:
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 */
123static int
124mext_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 */
170static int
171mext_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 folio_set_error(folio);
204 return err;
205 }
206 if (!buffer_mapped(bh)) {
207 folio_zero_range(folio, block_start, blocksize);
208 set_buffer_uptodate(bh);
209 continue;
210 }
211 }
212 BUG_ON(nr >= MAX_BUF_PER_PAGE);
213 arr[nr++] = bh;
214 }
215 /* No io required */
216 if (!nr)
217 goto out;
218
219 for (i = 0; i < nr; i++) {
220 bh = arr[i];
221 if (!bh_uptodate_or_lock(bh)) {
222 err = ext4_read_bh(bh, 0, NULL);
223 if (err)
224 return err;
225 }
226 }
227out:
228 if (!partial)
229 folio_mark_uptodate(folio);
230 return 0;
231}
232
233/**
234 * move_extent_per_page - Move extent data per page
235 *
236 * @o_filp: file structure of original file
237 * @donor_inode: donor inode
238 * @orig_page_offset: page index on original file
239 * @donor_page_offset: page index on donor file
240 * @data_offset_in_page: block index where data swapping starts
241 * @block_len_in_page: the number of blocks to be swapped
242 * @unwritten: orig extent is unwritten or not
243 * @err: pointer to save return value
244 *
245 * Save the data in original inode blocks and replace original inode extents
246 * with donor inode extents by calling ext4_swap_extents().
247 * Finally, write out the saved data in new original inode blocks. Return
248 * replaced block count.
249 */
250static int
251move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
252 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
253 int data_offset_in_page,
254 int block_len_in_page, int unwritten, int *err)
255{
256 struct inode *orig_inode = file_inode(o_filp);
257 struct folio *folio[2] = {NULL, NULL};
258 handle_t *handle;
259 ext4_lblk_t orig_blk_offset, donor_blk_offset;
260 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
261 unsigned int tmp_data_size, data_size, replaced_size;
262 int i, err2, jblocks, retries = 0;
263 int replaced_count = 0;
264 int from = data_offset_in_page << orig_inode->i_blkbits;
265 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
266 struct super_block *sb = orig_inode->i_sb;
267 struct buffer_head *bh = NULL;
268
269 /*
270 * It needs twice the amount of ordinary journal buffers because
271 * inode and donor_inode may change each different metadata blocks.
272 */
273again:
274 *err = 0;
275 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
276 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
277 if (IS_ERR(handle)) {
278 *err = PTR_ERR(handle);
279 return 0;
280 }
281
282 orig_blk_offset = orig_page_offset * blocks_per_page +
283 data_offset_in_page;
284
285 donor_blk_offset = donor_page_offset * blocks_per_page +
286 data_offset_in_page;
287
288 /* Calculate data_size */
289 if ((orig_blk_offset + block_len_in_page - 1) ==
290 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
291 /* Replace the last block */
292 tmp_data_size = orig_inode->i_size & (blocksize - 1);
293 /*
294 * If data_size equal zero, it shows data_size is multiples of
295 * blocksize. So we set appropriate value.
296 */
297 if (tmp_data_size == 0)
298 tmp_data_size = blocksize;
299
300 data_size = tmp_data_size +
301 ((block_len_in_page - 1) << orig_inode->i_blkbits);
302 } else
303 data_size = block_len_in_page << orig_inode->i_blkbits;
304
305 replaced_size = data_size;
306
307 *err = mext_folio_double_lock(orig_inode, donor_inode, orig_page_offset,
308 donor_page_offset, folio);
309 if (unlikely(*err < 0))
310 goto stop_journal;
311 /*
312 * If orig extent was unwritten it can become initialized
313 * at any time after i_data_sem was dropped, in order to
314 * serialize with delalloc we have recheck extent while we
315 * hold page's lock, if it is still the case data copy is not
316 * necessary, just swap data blocks between orig and donor.
317 */
318
319 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
320 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
321 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
322
323 if (unwritten) {
324 ext4_double_down_write_data_sem(orig_inode, donor_inode);
325 /* If any of extents in range became initialized we have to
326 * fallback to data copying */
327 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
328 block_len_in_page, 1, err);
329 if (*err)
330 goto drop_data_sem;
331
332 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
333 block_len_in_page, 1, err);
334 if (*err)
335 goto drop_data_sem;
336
337 if (!unwritten) {
338 ext4_double_up_write_data_sem(orig_inode, donor_inode);
339 goto data_copy;
340 }
341 if (!filemap_release_folio(folio[0], 0) ||
342 !filemap_release_folio(folio[1], 0)) {
343 *err = -EBUSY;
344 goto drop_data_sem;
345 }
346 replaced_count = ext4_swap_extents(handle, orig_inode,
347 donor_inode, orig_blk_offset,
348 donor_blk_offset,
349 block_len_in_page, 1, err);
350 drop_data_sem:
351 ext4_double_up_write_data_sem(orig_inode, donor_inode);
352 goto unlock_folios;
353 }
354data_copy:
355 *err = mext_page_mkuptodate(folio[0], from, from + replaced_size);
356 if (*err)
357 goto unlock_folios;
358
359 /* At this point all buffers in range are uptodate, old mapping layout
360 * is no longer required, try to drop it now. */
361 if (!filemap_release_folio(folio[0], 0) ||
362 !filemap_release_folio(folio[1], 0)) {
363 *err = -EBUSY;
364 goto unlock_folios;
365 }
366 ext4_double_down_write_data_sem(orig_inode, donor_inode);
367 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
368 orig_blk_offset, donor_blk_offset,
369 block_len_in_page, 1, err);
370 ext4_double_up_write_data_sem(orig_inode, donor_inode);
371 if (*err) {
372 if (replaced_count) {
373 block_len_in_page = replaced_count;
374 replaced_size =
375 block_len_in_page << orig_inode->i_blkbits;
376 } else
377 goto unlock_folios;
378 }
379 /* Perform all necessary steps similar write_begin()/write_end()
380 * but keeping in mind that i_size will not change */
381 bh = folio_buffers(folio[0]);
382 if (!bh)
383 bh = create_empty_buffers(folio[0],
384 1 << orig_inode->i_blkbits, 0);
385 for (i = 0; i < data_offset_in_page; i++)
386 bh = bh->b_this_page;
387 for (i = 0; i < block_len_in_page; i++) {
388 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
389 if (*err < 0)
390 goto repair_branches;
391 bh = bh->b_this_page;
392 }
393
394 block_commit_write(&folio[0]->page, from, from + replaced_size);
395
396 /* Even in case of data=writeback it is reasonable to pin
397 * inode to transaction, to prevent unexpected data loss */
398 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
399 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
400
401unlock_folios:
402 folio_unlock(folio[0]);
403 folio_put(folio[0]);
404 folio_unlock(folio[1]);
405 folio_put(folio[1]);
406stop_journal:
407 ext4_journal_stop(handle);
408 if (*err == -ENOSPC &&
409 ext4_should_retry_alloc(sb, &retries))
410 goto again;
411 /* Buffer was busy because probably is pinned to journal transaction,
412 * force transaction commit may help to free it. */
413 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
414 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
415 goto again;
416 return replaced_count;
417
418repair_branches:
419 /*
420 * This should never ever happen!
421 * Extents are swapped already, but we are not able to copy data.
422 * Try to swap extents to it's original places
423 */
424 ext4_double_down_write_data_sem(orig_inode, donor_inode);
425 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
426 orig_blk_offset, donor_blk_offset,
427 block_len_in_page, 0, &err2);
428 ext4_double_up_write_data_sem(orig_inode, donor_inode);
429 if (replaced_count != block_len_in_page) {
430 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
431 EIO, "Unable to copy data block,"
432 " data will be lost.");
433 *err = -EIO;
434 }
435 replaced_count = 0;
436 goto unlock_folios;
437}
438
439/**
440 * mext_check_arguments - Check whether move extent can be done
441 *
442 * @orig_inode: original inode
443 * @donor_inode: donor inode
444 * @orig_start: logical start offset in block for orig
445 * @donor_start: logical start offset in block for donor
446 * @len: the number of blocks to be moved
447 *
448 * Check the arguments of ext4_move_extents() whether the files can be
449 * exchanged with each other.
450 * Return 0 on success, or a negative error value on failure.
451 */
452static int
453mext_check_arguments(struct inode *orig_inode,
454 struct inode *donor_inode, __u64 orig_start,
455 __u64 donor_start, __u64 *len)
456{
457 __u64 orig_eof, donor_eof;
458 unsigned int blkbits = orig_inode->i_blkbits;
459 unsigned int blocksize = 1 << blkbits;
460
461 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
462 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
463
464
465 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
466 ext4_debug("ext4 move extent: suid or sgid is set"
467 " to donor file [ino:orig %lu, donor %lu]\n",
468 orig_inode->i_ino, donor_inode->i_ino);
469 return -EINVAL;
470 }
471
472 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
473 return -EPERM;
474
475 /* Ext4 move extent does not support swap files */
476 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
477 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
478 orig_inode->i_ino, donor_inode->i_ino);
479 return -ETXTBSY;
480 }
481
482 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
483 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
484 orig_inode->i_ino, donor_inode->i_ino);
485 return -EOPNOTSUPP;
486 }
487
488 /* Ext4 move extent supports only extent based file */
489 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
490 ext4_debug("ext4 move extent: orig file is not extents "
491 "based file [ino:orig %lu]\n", orig_inode->i_ino);
492 return -EOPNOTSUPP;
493 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
494 ext4_debug("ext4 move extent: donor file is not extents "
495 "based file [ino:donor %lu]\n", donor_inode->i_ino);
496 return -EOPNOTSUPP;
497 }
498
499 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
500 ext4_debug("ext4 move extent: File size is 0 byte\n");
501 return -EINVAL;
502 }
503
504 /* Start offset should be same */
505 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
506 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
507 ext4_debug("ext4 move extent: orig and donor's start "
508 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
509 orig_inode->i_ino, donor_inode->i_ino);
510 return -EINVAL;
511 }
512
513 if ((orig_start >= EXT_MAX_BLOCKS) ||
514 (donor_start >= EXT_MAX_BLOCKS) ||
515 (*len > EXT_MAX_BLOCKS) ||
516 (donor_start + *len >= EXT_MAX_BLOCKS) ||
517 (orig_start + *len >= EXT_MAX_BLOCKS)) {
518 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
519 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
520 orig_inode->i_ino, donor_inode->i_ino);
521 return -EINVAL;
522 }
523 if (orig_eof <= orig_start)
524 *len = 0;
525 else if (orig_eof < orig_start + *len - 1)
526 *len = orig_eof - orig_start;
527 if (donor_eof <= donor_start)
528 *len = 0;
529 else if (donor_eof < donor_start + *len - 1)
530 *len = donor_eof - donor_start;
531 if (!*len) {
532 ext4_debug("ext4 move extent: len should not be 0 "
533 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
534 donor_inode->i_ino);
535 return -EINVAL;
536 }
537
538 return 0;
539}
540
541/**
542 * ext4_move_extents - Exchange the specified range of a file
543 *
544 * @o_filp: file structure of the original file
545 * @d_filp: file structure of the donor file
546 * @orig_blk: start offset in block for orig
547 * @donor_blk: start offset in block for donor
548 * @len: the number of blocks to be moved
549 * @moved_len: moved block length
550 *
551 * This function returns 0 and moved block length is set in moved_len
552 * if succeed, otherwise returns error value.
553 *
554 */
555int
556ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
557 __u64 donor_blk, __u64 len, __u64 *moved_len)
558{
559 struct inode *orig_inode = file_inode(o_filp);
560 struct inode *donor_inode = file_inode(d_filp);
561 struct ext4_ext_path *path = NULL;
562 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
563 ext4_lblk_t o_end, o_start = orig_blk;
564 ext4_lblk_t d_start = donor_blk;
565 int ret;
566
567 if (orig_inode->i_sb != donor_inode->i_sb) {
568 ext4_debug("ext4 move extent: The argument files "
569 "should be in same FS [ino:orig %lu, donor %lu]\n",
570 orig_inode->i_ino, donor_inode->i_ino);
571 return -EINVAL;
572 }
573
574 /* orig and donor should be different inodes */
575 if (orig_inode == donor_inode) {
576 ext4_debug("ext4 move extent: The argument files should not "
577 "be same inode [ino:orig %lu, donor %lu]\n",
578 orig_inode->i_ino, donor_inode->i_ino);
579 return -EINVAL;
580 }
581
582 /* Regular file check */
583 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
584 ext4_debug("ext4 move extent: The argument files should be "
585 "regular file [ino:orig %lu, donor %lu]\n",
586 orig_inode->i_ino, donor_inode->i_ino);
587 return -EINVAL;
588 }
589
590 /* TODO: it's not obvious how to swap blocks for inodes with full
591 journaling enabled */
592 if (ext4_should_journal_data(orig_inode) ||
593 ext4_should_journal_data(donor_inode)) {
594 ext4_msg(orig_inode->i_sb, KERN_ERR,
595 "Online defrag not supported with data journaling");
596 return -EOPNOTSUPP;
597 }
598
599 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
600 ext4_msg(orig_inode->i_sb, KERN_ERR,
601 "Online defrag not supported for encrypted files");
602 return -EOPNOTSUPP;
603 }
604
605 /* Protect orig and donor inodes against a truncate */
606 lock_two_nondirectories(orig_inode, donor_inode);
607
608 /* Wait for all existing dio workers */
609 inode_dio_wait(orig_inode);
610 inode_dio_wait(donor_inode);
611
612 /* Protect extent tree against block allocations via delalloc */
613 ext4_double_down_write_data_sem(orig_inode, donor_inode);
614 /* Check the filesystem environment whether move_extent can be done */
615 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
616 donor_blk, &len);
617 if (ret)
618 goto out;
619 o_end = o_start + len;
620
621 *moved_len = 0;
622 while (o_start < o_end) {
623 struct ext4_extent *ex;
624 ext4_lblk_t cur_blk, next_blk;
625 pgoff_t orig_page_index, donor_page_index;
626 int offset_in_page;
627 int unwritten, cur_len;
628
629 ret = get_ext_path(orig_inode, o_start, &path);
630 if (ret)
631 goto out;
632 ex = path[path->p_depth].p_ext;
633 cur_blk = le32_to_cpu(ex->ee_block);
634 cur_len = ext4_ext_get_actual_len(ex);
635 /* Check hole before the start pos */
636 if (cur_blk + cur_len - 1 < o_start) {
637 next_blk = ext4_ext_next_allocated_block(path);
638 if (next_blk == EXT_MAX_BLOCKS) {
639 ret = -ENODATA;
640 goto out;
641 }
642 d_start += next_blk - o_start;
643 o_start = next_blk;
644 continue;
645 /* Check hole after the start pos */
646 } else if (cur_blk > o_start) {
647 /* Skip hole */
648 d_start += cur_blk - o_start;
649 o_start = cur_blk;
650 /* Extent inside requested range ?*/
651 if (cur_blk >= o_end)
652 goto out;
653 } else { /* in_range(o_start, o_blk, o_len) */
654 cur_len += cur_blk - o_start;
655 }
656 unwritten = ext4_ext_is_unwritten(ex);
657 if (o_end - o_start < cur_len)
658 cur_len = o_end - o_start;
659
660 orig_page_index = o_start >> (PAGE_SHIFT -
661 orig_inode->i_blkbits);
662 donor_page_index = d_start >> (PAGE_SHIFT -
663 donor_inode->i_blkbits);
664 offset_in_page = o_start % blocks_per_page;
665 if (cur_len > blocks_per_page - offset_in_page)
666 cur_len = blocks_per_page - offset_in_page;
667 /*
668 * Up semaphore to avoid following problems:
669 * a. transaction deadlock among ext4_journal_start,
670 * ->write_begin via pagefault, and jbd2_journal_commit
671 * b. racing with ->read_folio, ->write_begin, and
672 * ext4_get_block in move_extent_per_page
673 */
674 ext4_double_up_write_data_sem(orig_inode, donor_inode);
675 /* Swap original branches with new branches */
676 *moved_len += move_extent_per_page(o_filp, donor_inode,
677 orig_page_index, donor_page_index,
678 offset_in_page, cur_len,
679 unwritten, &ret);
680 ext4_double_down_write_data_sem(orig_inode, donor_inode);
681 if (ret < 0)
682 break;
683 o_start += cur_len;
684 d_start += cur_len;
685 }
686
687out:
688 if (*moved_len) {
689 ext4_discard_preallocations(orig_inode);
690 ext4_discard_preallocations(donor_inode);
691 }
692
693 ext4_free_ext_path(path);
694 ext4_double_up_write_data_sem(orig_inode, donor_inode);
695 unlock_two_nondirectories(orig_inode, donor_inode);
696
697 return ret;
698}