Loading...
1/*
2 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
3 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
4 * Akira Fujita <a-fujita@rs.jp.nec.com>
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2.1 of the GNU Lesser General Public License
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 */
15
16#include <linux/fs.h>
17#include <linux/quotaops.h>
18#include <linux/slab.h>
19#include "ext4_jbd2.h"
20#include "ext4.h"
21#include "ext4_extents.h"
22
23/**
24 * get_ext_path - Find an extent path for designated logical block number.
25 *
26 * @inode: an inode which is searched
27 * @lblock: logical block number to find an extent path
28 * @path: pointer to an extent path pointer (for output)
29 *
30 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
31 * on failure.
32 */
33static inline int
34get_ext_path(struct inode *inode, ext4_lblk_t lblock,
35 struct ext4_ext_path **ppath)
36{
37 struct ext4_ext_path *path;
38
39 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
40 if (IS_ERR(path))
41 return PTR_ERR(path);
42 if (path[ext_depth(inode)].p_ext == NULL) {
43 ext4_ext_drop_refs(path);
44 kfree(path);
45 *ppath = NULL;
46 return -ENODATA;
47 }
48 *ppath = path;
49 return 0;
50}
51
52/**
53 * ext4_double_down_write_data_sem - Acquire two inodes' write lock
54 * of i_data_sem
55 *
56 * Acquire write lock of i_data_sem of the two inodes
57 */
58void
59ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
60{
61 if (first < second) {
62 down_write(&EXT4_I(first)->i_data_sem);
63 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
64 } else {
65 down_write(&EXT4_I(second)->i_data_sem);
66 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
67
68 }
69}
70
71/**
72 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
73 *
74 * @orig_inode: original inode structure to be released its lock first
75 * @donor_inode: donor inode structure to be released its lock second
76 * Release write lock of i_data_sem of two inodes (orig and donor).
77 */
78void
79ext4_double_up_write_data_sem(struct inode *orig_inode,
80 struct inode *donor_inode)
81{
82 up_write(&EXT4_I(orig_inode)->i_data_sem);
83 up_write(&EXT4_I(donor_inode)->i_data_sem);
84}
85
86/**
87 * mext_check_coverage - Check that all extents in range has the same type
88 *
89 * @inode: inode in question
90 * @from: block offset of inode
91 * @count: block count to be checked
92 * @unwritten: extents expected to be unwritten
93 * @err: pointer to save error value
94 *
95 * Return 1 if all extents in range has expected type, and zero otherwise.
96 */
97static int
98mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
99 int unwritten, int *err)
100{
101 struct ext4_ext_path *path = NULL;
102 struct ext4_extent *ext;
103 int ret = 0;
104 ext4_lblk_t last = from + count;
105 while (from < last) {
106 *err = get_ext_path(inode, from, &path);
107 if (*err)
108 goto out;
109 ext = path[ext_depth(inode)].p_ext;
110 if (unwritten != ext4_ext_is_unwritten(ext))
111 goto out;
112 from += ext4_ext_get_actual_len(ext);
113 ext4_ext_drop_refs(path);
114 }
115 ret = 1;
116out:
117 ext4_ext_drop_refs(path);
118 kfree(path);
119 return ret;
120}
121
122/**
123 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
124 *
125 * @inode1: the inode structure
126 * @inode2: the inode structure
127 * @index1: page index
128 * @index2: page index
129 * @page: result page vector
130 *
131 * Grab two locked pages for inode's by inode order
132 */
133static int
134mext_page_double_lock(struct inode *inode1, struct inode *inode2,
135 pgoff_t index1, pgoff_t index2, struct page *page[2])
136{
137 struct address_space *mapping[2];
138 unsigned fl = AOP_FLAG_NOFS;
139
140 BUG_ON(!inode1 || !inode2);
141 if (inode1 < inode2) {
142 mapping[0] = inode1->i_mapping;
143 mapping[1] = inode2->i_mapping;
144 } else {
145 pgoff_t tmp = index1;
146 index1 = index2;
147 index2 = tmp;
148 mapping[0] = inode2->i_mapping;
149 mapping[1] = inode1->i_mapping;
150 }
151
152 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
153 if (!page[0])
154 return -ENOMEM;
155
156 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
157 if (!page[1]) {
158 unlock_page(page[0]);
159 put_page(page[0]);
160 return -ENOMEM;
161 }
162 /*
163 * grab_cache_page_write_begin() may not wait on page's writeback if
164 * BDI not demand that. But it is reasonable to be very conservative
165 * here and explicitly wait on page's writeback
166 */
167 wait_on_page_writeback(page[0]);
168 wait_on_page_writeback(page[1]);
169 if (inode1 > inode2)
170 swap(page[0], page[1]);
171
172 return 0;
173}
174
175/* Force page buffers uptodate w/o dropping page's lock */
176static int
177mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
178{
179 struct inode *inode = page->mapping->host;
180 sector_t block;
181 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
182 unsigned int blocksize, block_start, block_end;
183 int i, err, nr = 0, partial = 0;
184 BUG_ON(!PageLocked(page));
185 BUG_ON(PageWriteback(page));
186
187 if (PageUptodate(page))
188 return 0;
189
190 blocksize = 1 << inode->i_blkbits;
191 if (!page_has_buffers(page))
192 create_empty_buffers(page, blocksize, 0);
193
194 head = page_buffers(page);
195 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
196 for (bh = head, block_start = 0; bh != head || !block_start;
197 block++, block_start = block_end, bh = bh->b_this_page) {
198 block_end = block_start + blocksize;
199 if (block_end <= from || block_start >= to) {
200 if (!buffer_uptodate(bh))
201 partial = 1;
202 continue;
203 }
204 if (buffer_uptodate(bh))
205 continue;
206 if (!buffer_mapped(bh)) {
207 err = ext4_get_block(inode, block, bh, 0);
208 if (err) {
209 SetPageError(page);
210 return err;
211 }
212 if (!buffer_mapped(bh)) {
213 zero_user(page, block_start, blocksize);
214 set_buffer_uptodate(bh);
215 continue;
216 }
217 }
218 BUG_ON(nr >= MAX_BUF_PER_PAGE);
219 arr[nr++] = bh;
220 }
221 /* No io required */
222 if (!nr)
223 goto out;
224
225 for (i = 0; i < nr; i++) {
226 bh = arr[i];
227 if (!bh_uptodate_or_lock(bh)) {
228 err = bh_submit_read(bh);
229 if (err)
230 return err;
231 }
232 }
233out:
234 if (!partial)
235 SetPageUptodate(page);
236 return 0;
237}
238
239/**
240 * move_extent_per_page - Move extent data per page
241 *
242 * @o_filp: file structure of original file
243 * @donor_inode: donor inode
244 * @orig_page_offset: page index on original file
245 * @donor_page_offset: page index on donor file
246 * @data_offset_in_page: block index where data swapping starts
247 * @block_len_in_page: the number of blocks to be swapped
248 * @unwritten: orig extent is unwritten or not
249 * @err: pointer to save return value
250 *
251 * Save the data in original inode blocks and replace original inode extents
252 * with donor inode extents by calling ext4_swap_extents().
253 * Finally, write out the saved data in new original inode blocks. Return
254 * replaced block count.
255 */
256static int
257move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
258 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
259 int data_offset_in_page,
260 int block_len_in_page, int unwritten, int *err)
261{
262 struct inode *orig_inode = file_inode(o_filp);
263 struct page *pagep[2] = {NULL, NULL};
264 handle_t *handle;
265 ext4_lblk_t orig_blk_offset, donor_blk_offset;
266 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
267 unsigned int tmp_data_size, data_size, replaced_size;
268 int i, err2, jblocks, retries = 0;
269 int replaced_count = 0;
270 int from = data_offset_in_page << orig_inode->i_blkbits;
271 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
272 struct super_block *sb = orig_inode->i_sb;
273 struct buffer_head *bh = NULL;
274
275 /*
276 * It needs twice the amount of ordinary journal buffers because
277 * inode and donor_inode may change each different metadata blocks.
278 */
279again:
280 *err = 0;
281 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
282 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
283 if (IS_ERR(handle)) {
284 *err = PTR_ERR(handle);
285 return 0;
286 }
287
288 orig_blk_offset = orig_page_offset * blocks_per_page +
289 data_offset_in_page;
290
291 donor_blk_offset = donor_page_offset * blocks_per_page +
292 data_offset_in_page;
293
294 /* Calculate data_size */
295 if ((orig_blk_offset + block_len_in_page - 1) ==
296 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
297 /* Replace the last block */
298 tmp_data_size = orig_inode->i_size & (blocksize - 1);
299 /*
300 * If data_size equal zero, it shows data_size is multiples of
301 * blocksize. So we set appropriate value.
302 */
303 if (tmp_data_size == 0)
304 tmp_data_size = blocksize;
305
306 data_size = tmp_data_size +
307 ((block_len_in_page - 1) << orig_inode->i_blkbits);
308 } else
309 data_size = block_len_in_page << orig_inode->i_blkbits;
310
311 replaced_size = data_size;
312
313 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
314 donor_page_offset, pagep);
315 if (unlikely(*err < 0))
316 goto stop_journal;
317 /*
318 * If orig extent was unwritten it can become initialized
319 * at any time after i_data_sem was dropped, in order to
320 * serialize with delalloc we have recheck extent while we
321 * hold page's lock, if it is still the case data copy is not
322 * necessary, just swap data blocks between orig and donor.
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 ((page_has_private(pagep[0]) &&
343 !try_to_release_page(pagep[0], 0)) ||
344 (page_has_private(pagep[1]) &&
345 !try_to_release_page(pagep[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_pages;
356 }
357data_copy:
358 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
359 if (*err)
360 goto unlock_pages;
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 ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
365 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
366 *err = -EBUSY;
367 goto unlock_pages;
368 }
369 ext4_double_down_write_data_sem(orig_inode, donor_inode);
370 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
371 orig_blk_offset, donor_blk_offset,
372 block_len_in_page, 1, err);
373 ext4_double_up_write_data_sem(orig_inode, donor_inode);
374 if (*err) {
375 if (replaced_count) {
376 block_len_in_page = replaced_count;
377 replaced_size =
378 block_len_in_page << orig_inode->i_blkbits;
379 } else
380 goto unlock_pages;
381 }
382 /* Perform all necessary steps similar write_begin()/write_end()
383 * but keeping in mind that i_size will not change */
384 if (!page_has_buffers(pagep[0]))
385 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
386 bh = page_buffers(pagep[0]);
387 for (i = 0; i < data_offset_in_page; i++)
388 bh = bh->b_this_page;
389 for (i = 0; i < block_len_in_page; i++) {
390 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
391 if (*err < 0)
392 break;
393 bh = bh->b_this_page;
394 }
395 if (!*err)
396 *err = block_commit_write(pagep[0], from, from + replaced_size);
397
398 if (unlikely(*err < 0))
399 goto repair_branches;
400
401 /* Even in case of data=writeback it is reasonable to pin
402 * inode to transaction, to prevent unexpected data loss */
403 *err = ext4_jbd2_file_inode(handle, orig_inode);
404
405unlock_pages:
406 unlock_page(pagep[0]);
407 put_page(pagep[0]);
408 unlock_page(pagep[1]);
409 put_page(pagep[1]);
410stop_journal:
411 ext4_journal_stop(handle);
412 if (*err == -ENOSPC &&
413 ext4_should_retry_alloc(sb, &retries))
414 goto again;
415 /* Buffer was busy because probably is pinned to journal transaction,
416 * force transaction commit may help to free it. */
417 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
418 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
419 goto again;
420 return replaced_count;
421
422repair_branches:
423 /*
424 * This should never ever happen!
425 * Extents are swapped already, but we are not able to copy data.
426 * Try to swap extents to it's original places
427 */
428 ext4_double_down_write_data_sem(orig_inode, donor_inode);
429 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
430 orig_blk_offset, donor_blk_offset,
431 block_len_in_page, 0, &err2);
432 ext4_double_up_write_data_sem(orig_inode, donor_inode);
433 if (replaced_count != block_len_in_page) {
434 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
435 "Unable to copy data block,"
436 " data will be lost.");
437 *err = -EIO;
438 }
439 replaced_count = 0;
440 goto unlock_pages;
441}
442
443/**
444 * mext_check_arguments - Check whether move extent can be done
445 *
446 * @orig_inode: original inode
447 * @donor_inode: donor inode
448 * @orig_start: logical start offset in block for orig
449 * @donor_start: logical start offset in block for donor
450 * @len: the number of blocks to be moved
451 *
452 * Check the arguments of ext4_move_extents() whether the files can be
453 * exchanged with each other.
454 * Return 0 on success, or a negative error value on failure.
455 */
456static int
457mext_check_arguments(struct inode *orig_inode,
458 struct inode *donor_inode, __u64 orig_start,
459 __u64 donor_start, __u64 *len)
460{
461 __u64 orig_eof, donor_eof;
462 unsigned int blkbits = orig_inode->i_blkbits;
463 unsigned int blocksize = 1 << blkbits;
464
465 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
466 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
467
468
469 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
470 ext4_debug("ext4 move extent: suid or sgid is set"
471 " to donor file [ino:orig %lu, donor %lu]\n",
472 orig_inode->i_ino, donor_inode->i_ino);
473 return -EINVAL;
474 }
475
476 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
477 return -EPERM;
478
479 /* Ext4 move extent does not support swapfile */
480 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
481 ext4_debug("ext4 move extent: The argument files should "
482 "not be swapfile [ino:orig %lu, donor %lu]\n",
483 orig_inode->i_ino, donor_inode->i_ino);
484 return -EBUSY;
485 }
486
487 if (IS_NOQUOTA(orig_inode) || IS_NOQUOTA(donor_inode)) {
488 ext4_debug("ext4 move extent: The argument files should "
489 "not be quota files [ino:orig %lu, donor %lu]\n",
490 orig_inode->i_ino, donor_inode->i_ino);
491 return -EBUSY;
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 "offset are not alligned [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 + *len - 1)
530 *len = orig_eof - orig_start;
531 if (donor_eof < donor_start + *len - 1)
532 *len = donor_eof - donor_start;
533 if (!*len) {
534 ext4_debug("ext4 move extent: len should not be 0 "
535 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
536 donor_inode->i_ino);
537 return -EINVAL;
538 }
539
540 return 0;
541}
542
543/**
544 * ext4_move_extents - Exchange the specified range of a file
545 *
546 * @o_filp: file structure of the original file
547 * @d_filp: file structure of the donor file
548 * @orig_blk: start offset in block for orig
549 * @donor_blk: start offset in block for donor
550 * @len: the number of blocks to be moved
551 * @moved_len: moved block length
552 *
553 * This function returns 0 and moved block length is set in moved_len
554 * if succeed, otherwise returns error value.
555 *
556 */
557int
558ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
559 __u64 donor_blk, __u64 len, __u64 *moved_len)
560{
561 struct inode *orig_inode = file_inode(o_filp);
562 struct inode *donor_inode = file_inode(d_filp);
563 struct ext4_ext_path *path = NULL;
564 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
565 ext4_lblk_t o_end, o_start = orig_blk;
566 ext4_lblk_t d_start = donor_blk;
567 int ret;
568
569 if (orig_inode->i_sb != donor_inode->i_sb) {
570 ext4_debug("ext4 move extent: The argument files "
571 "should be in same FS [ino:orig %lu, donor %lu]\n",
572 orig_inode->i_ino, donor_inode->i_ino);
573 return -EINVAL;
574 }
575
576 /* orig and donor should be different inodes */
577 if (orig_inode == donor_inode) {
578 ext4_debug("ext4 move extent: The argument files should not "
579 "be same inode [ino:orig %lu, donor %lu]\n",
580 orig_inode->i_ino, donor_inode->i_ino);
581 return -EINVAL;
582 }
583
584 /* Regular file check */
585 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
586 ext4_debug("ext4 move extent: The argument files should be "
587 "regular file [ino:orig %lu, donor %lu]\n",
588 orig_inode->i_ino, donor_inode->i_ino);
589 return -EINVAL;
590 }
591
592 /* TODO: it's not obvious how to swap blocks for inodes with full
593 journaling enabled */
594 if (ext4_should_journal_data(orig_inode) ||
595 ext4_should_journal_data(donor_inode)) {
596 ext4_msg(orig_inode->i_sb, KERN_ERR,
597 "Online defrag not supported with data journaling");
598 return -EOPNOTSUPP;
599 }
600
601 /* Protect orig and donor inodes against a truncate */
602 lock_two_nondirectories(orig_inode, donor_inode);
603
604 /* Wait for all existing dio workers */
605 ext4_inode_block_unlocked_dio(orig_inode);
606 ext4_inode_block_unlocked_dio(donor_inode);
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 while (o_start < o_end) {
620 struct ext4_extent *ex;
621 ext4_lblk_t cur_blk, next_blk;
622 pgoff_t orig_page_index, donor_page_index;
623 int offset_in_page;
624 int unwritten, cur_len;
625
626 ret = get_ext_path(orig_inode, o_start, &path);
627 if (ret)
628 goto out;
629 ex = path[path->p_depth].p_ext;
630 next_blk = ext4_ext_next_allocated_block(path);
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 if (next_blk == EXT_MAX_BLOCKS) {
636 o_start = o_end;
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 ->readpage, ->write_begin, and ext4_get_block
670 * 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 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 *moved_len = o_start - orig_blk;
685 if (*moved_len > len)
686 *moved_len = len;
687
688out:
689 if (*moved_len) {
690 ext4_discard_preallocations(orig_inode);
691 ext4_discard_preallocations(donor_inode);
692 }
693
694 ext4_ext_drop_refs(path);
695 kfree(path);
696 ext4_double_up_write_data_sem(orig_inode, donor_inode);
697 ext4_inode_resume_unlocked_dio(orig_inode);
698 ext4_inode_resume_unlocked_dio(donor_inode);
699 unlock_two_nondirectories(orig_inode, donor_inode);
700
701 return ret;
702}
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}