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 "ext4_jbd2.h"
12 #include "ext4.h"
13 #include "ext4_extents.h"
14
15 /**
16 * get_ext_path() - Find an extent path for designated logical block number.
17 * @inode: inode to be searched
18 * @lblock: logical block number to find an extent path
19 * @ppath: pointer to an extent path pointer (for output)
20 *
21 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
22 * on failure.
23 */
24 static inline int
get_ext_path(struct inode * inode,ext4_lblk_t lblock,struct ext4_ext_path ** ppath)25 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
26 struct ext4_ext_path **ppath)
27 {
28 struct ext4_ext_path *path;
29
30 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
31 if (IS_ERR(path))
32 return PTR_ERR(path);
33 if (path[ext_depth(inode)].p_ext == NULL) {
34 ext4_ext_drop_refs(path);
35 kfree(path);
36 *ppath = NULL;
37 return -ENODATA;
38 }
39 *ppath = path;
40 return 0;
41 }
42
43 /**
44 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
45 * @first: inode to be locked
46 * @second: inode to be locked
47 *
48 * Acquire write lock of i_data_sem of the two inodes
49 */
50 void
ext4_double_down_write_data_sem(struct inode * first,struct inode * second)51 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
52 {
53 if (first < second) {
54 down_write(&EXT4_I(first)->i_data_sem);
55 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
56 } else {
57 down_write(&EXT4_I(second)->i_data_sem);
58 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
59
60 }
61 }
62
63 /**
64 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
65 *
66 * @orig_inode: original inode structure to be released its lock first
67 * @donor_inode: donor inode structure to be released its lock second
68 * Release write lock of i_data_sem of two inodes (orig and donor).
69 */
70 void
ext4_double_up_write_data_sem(struct inode * orig_inode,struct inode * donor_inode)71 ext4_double_up_write_data_sem(struct inode *orig_inode,
72 struct inode *donor_inode)
73 {
74 up_write(&EXT4_I(orig_inode)->i_data_sem);
75 up_write(&EXT4_I(donor_inode)->i_data_sem);
76 }
77
78 /**
79 * mext_check_coverage - Check that all extents in range has the same type
80 *
81 * @inode: inode in question
82 * @from: block offset of inode
83 * @count: block count to be checked
84 * @unwritten: extents expected to be unwritten
85 * @err: pointer to save error value
86 *
87 * Return 1 if all extents in range has expected type, and zero otherwise.
88 */
89 static int
mext_check_coverage(struct inode * inode,ext4_lblk_t from,ext4_lblk_t count,int unwritten,int * err)90 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
91 int unwritten, int *err)
92 {
93 struct ext4_ext_path *path = NULL;
94 struct ext4_extent *ext;
95 int ret = 0;
96 ext4_lblk_t last = from + count;
97 while (from < last) {
98 *err = get_ext_path(inode, from, &path);
99 if (*err)
100 goto out;
101 ext = path[ext_depth(inode)].p_ext;
102 if (unwritten != ext4_ext_is_unwritten(ext))
103 goto out;
104 from += ext4_ext_get_actual_len(ext);
105 ext4_ext_drop_refs(path);
106 }
107 ret = 1;
108 out:
109 ext4_ext_drop_refs(path);
110 kfree(path);
111 return ret;
112 }
113
114 /**
115 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
116 *
117 * @inode1: the inode structure
118 * @inode2: the inode structure
119 * @index1: page index
120 * @index2: page index
121 * @page: result page vector
122 *
123 * Grab two locked pages for inode's by inode order
124 */
125 static int
mext_page_double_lock(struct inode * inode1,struct inode * inode2,pgoff_t index1,pgoff_t index2,struct page * page[2])126 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
127 pgoff_t index1, pgoff_t index2, struct page *page[2])
128 {
129 struct address_space *mapping[2];
130 unsigned fl = AOP_FLAG_NOFS;
131
132 BUG_ON(!inode1 || !inode2);
133 if (inode1 < inode2) {
134 mapping[0] = inode1->i_mapping;
135 mapping[1] = inode2->i_mapping;
136 } else {
137 swap(index1, index2);
138 mapping[0] = inode2->i_mapping;
139 mapping[1] = inode1->i_mapping;
140 }
141
142 page[0] = grab_cache_page_write_begin(mapping[0], index1, fl);
143 if (!page[0])
144 return -ENOMEM;
145
146 page[1] = grab_cache_page_write_begin(mapping[1], index2, fl);
147 if (!page[1]) {
148 unlock_page(page[0]);
149 put_page(page[0]);
150 return -ENOMEM;
151 }
152 /*
153 * grab_cache_page_write_begin() may not wait on page's writeback if
154 * BDI not demand that. But it is reasonable to be very conservative
155 * here and explicitly wait on page's writeback
156 */
157 wait_on_page_writeback(page[0]);
158 wait_on_page_writeback(page[1]);
159 if (inode1 > inode2)
160 swap(page[0], page[1]);
161
162 return 0;
163 }
164
165 /* Force page buffers uptodate w/o dropping page's lock */
166 static int
mext_page_mkuptodate(struct page * page,unsigned from,unsigned to)167 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
168 {
169 struct inode *inode = page->mapping->host;
170 sector_t block;
171 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
172 unsigned int blocksize, block_start, block_end;
173 int i, err, nr = 0, partial = 0;
174 BUG_ON(!PageLocked(page));
175 BUG_ON(PageWriteback(page));
176
177 if (PageUptodate(page))
178 return 0;
179
180 blocksize = i_blocksize(inode);
181 if (!page_has_buffers(page))
182 create_empty_buffers(page, blocksize, 0);
183
184 head = page_buffers(page);
185 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
186 for (bh = head, block_start = 0; bh != head || !block_start;
187 block++, block_start = block_end, bh = bh->b_this_page) {
188 block_end = block_start + blocksize;
189 if (block_end <= from || block_start >= to) {
190 if (!buffer_uptodate(bh))
191 partial = 1;
192 continue;
193 }
194 if (buffer_uptodate(bh))
195 continue;
196 if (!buffer_mapped(bh)) {
197 err = ext4_get_block(inode, block, bh, 0);
198 if (err) {
199 SetPageError(page);
200 return err;
201 }
202 if (!buffer_mapped(bh)) {
203 zero_user(page, block_start, blocksize);
204 set_buffer_uptodate(bh);
205 continue;
206 }
207 }
208 BUG_ON(nr >= MAX_BUF_PER_PAGE);
209 arr[nr++] = bh;
210 }
211 /* No io required */
212 if (!nr)
213 goto out;
214
215 for (i = 0; i < nr; i++) {
216 bh = arr[i];
217 if (!bh_uptodate_or_lock(bh)) {
218 err = ext4_read_bh(bh, 0, NULL);
219 if (err)
220 return err;
221 }
222 }
223 out:
224 if (!partial)
225 SetPageUptodate(page);
226 return 0;
227 }
228
229 /**
230 * move_extent_per_page - Move extent data per page
231 *
232 * @o_filp: file structure of original file
233 * @donor_inode: donor inode
234 * @orig_page_offset: page index on original file
235 * @donor_page_offset: page index on donor file
236 * @data_offset_in_page: block index where data swapping starts
237 * @block_len_in_page: the number of blocks to be swapped
238 * @unwritten: orig extent is unwritten or not
239 * @err: pointer to save return value
240 *
241 * Save the data in original inode blocks and replace original inode extents
242 * with donor inode extents by calling ext4_swap_extents().
243 * Finally, write out the saved data in new original inode blocks. Return
244 * replaced block count.
245 */
246 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)247 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
248 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
249 int data_offset_in_page,
250 int block_len_in_page, int unwritten, int *err)
251 {
252 struct inode *orig_inode = file_inode(o_filp);
253 struct page *pagep[2] = {NULL, NULL};
254 handle_t *handle;
255 ext4_lblk_t orig_blk_offset, donor_blk_offset;
256 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
257 unsigned int tmp_data_size, data_size, replaced_size;
258 int i, err2, jblocks, retries = 0;
259 int replaced_count = 0;
260 int from = data_offset_in_page << orig_inode->i_blkbits;
261 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
262 struct super_block *sb = orig_inode->i_sb;
263 struct buffer_head *bh = NULL;
264
265 /*
266 * It needs twice the amount of ordinary journal buffers because
267 * inode and donor_inode may change each different metadata blocks.
268 */
269 again:
270 *err = 0;
271 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
272 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
273 if (IS_ERR(handle)) {
274 *err = PTR_ERR(handle);
275 return 0;
276 }
277
278 orig_blk_offset = orig_page_offset * blocks_per_page +
279 data_offset_in_page;
280
281 donor_blk_offset = donor_page_offset * blocks_per_page +
282 data_offset_in_page;
283
284 /* Calculate data_size */
285 if ((orig_blk_offset + block_len_in_page - 1) ==
286 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
287 /* Replace the last block */
288 tmp_data_size = orig_inode->i_size & (blocksize - 1);
289 /*
290 * If data_size equal zero, it shows data_size is multiples of
291 * blocksize. So we set appropriate value.
292 */
293 if (tmp_data_size == 0)
294 tmp_data_size = blocksize;
295
296 data_size = tmp_data_size +
297 ((block_len_in_page - 1) << orig_inode->i_blkbits);
298 } else
299 data_size = block_len_in_page << orig_inode->i_blkbits;
300
301 replaced_size = data_size;
302
303 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
304 donor_page_offset, pagep);
305 if (unlikely(*err < 0))
306 goto stop_journal;
307 /*
308 * If orig extent was unwritten it can become initialized
309 * at any time after i_data_sem was dropped, in order to
310 * serialize with delalloc we have recheck extent while we
311 * hold page's lock, if it is still the case data copy is not
312 * necessary, just swap data blocks between orig and donor.
313 */
314 if (unwritten) {
315 ext4_double_down_write_data_sem(orig_inode, donor_inode);
316 /* If any of extents in range became initialized we have to
317 * fallback to data copying */
318 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
319 block_len_in_page, 1, err);
320 if (*err)
321 goto drop_data_sem;
322
323 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
324 block_len_in_page, 1, err);
325 if (*err)
326 goto drop_data_sem;
327
328 if (!unwritten) {
329 ext4_double_up_write_data_sem(orig_inode, donor_inode);
330 goto data_copy;
331 }
332 if ((page_has_private(pagep[0]) &&
333 !try_to_release_page(pagep[0], 0)) ||
334 (page_has_private(pagep[1]) &&
335 !try_to_release_page(pagep[1], 0))) {
336 *err = -EBUSY;
337 goto drop_data_sem;
338 }
339 replaced_count = ext4_swap_extents(handle, orig_inode,
340 donor_inode, orig_blk_offset,
341 donor_blk_offset,
342 block_len_in_page, 1, err);
343 drop_data_sem:
344 ext4_double_up_write_data_sem(orig_inode, donor_inode);
345 goto unlock_pages;
346 }
347 data_copy:
348 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
349 if (*err)
350 goto unlock_pages;
351
352 /* At this point all buffers in range are uptodate, old mapping layout
353 * is no longer required, try to drop it now. */
354 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
355 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
356 *err = -EBUSY;
357 goto unlock_pages;
358 }
359 ext4_double_down_write_data_sem(orig_inode, donor_inode);
360 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
361 orig_blk_offset, donor_blk_offset,
362 block_len_in_page, 1, err);
363 ext4_double_up_write_data_sem(orig_inode, donor_inode);
364 if (*err) {
365 if (replaced_count) {
366 block_len_in_page = replaced_count;
367 replaced_size =
368 block_len_in_page << orig_inode->i_blkbits;
369 } else
370 goto unlock_pages;
371 }
372 /* Perform all necessary steps similar write_begin()/write_end()
373 * but keeping in mind that i_size will not change */
374 if (!page_has_buffers(pagep[0]))
375 create_empty_buffers(pagep[0], 1 << orig_inode->i_blkbits, 0);
376 bh = page_buffers(pagep[0]);
377 for (i = 0; i < data_offset_in_page; i++)
378 bh = bh->b_this_page;
379 for (i = 0; i < block_len_in_page; i++) {
380 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
381 if (*err < 0)
382 break;
383 bh = bh->b_this_page;
384 }
385 if (!*err)
386 *err = block_commit_write(pagep[0], from, from + replaced_size);
387
388 if (unlikely(*err < 0))
389 goto repair_branches;
390
391 /* Even in case of data=writeback it is reasonable to pin
392 * inode to transaction, to prevent unexpected data loss */
393 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
394 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
395
396 unlock_pages:
397 unlock_page(pagep[0]);
398 put_page(pagep[0]);
399 unlock_page(pagep[1]);
400 put_page(pagep[1]);
401 stop_journal:
402 ext4_journal_stop(handle);
403 if (*err == -ENOSPC &&
404 ext4_should_retry_alloc(sb, &retries))
405 goto again;
406 /* Buffer was busy because probably is pinned to journal transaction,
407 * force transaction commit may help to free it. */
408 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
409 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
410 goto again;
411 return replaced_count;
412
413 repair_branches:
414 /*
415 * This should never ever happen!
416 * Extents are swapped already, but we are not able to copy data.
417 * Try to swap extents to it's original places
418 */
419 ext4_double_down_write_data_sem(orig_inode, donor_inode);
420 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
421 orig_blk_offset, donor_blk_offset,
422 block_len_in_page, 0, &err2);
423 ext4_double_up_write_data_sem(orig_inode, donor_inode);
424 if (replaced_count != block_len_in_page) {
425 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
426 EIO, "Unable to copy data block,"
427 " data will be lost.");
428 *err = -EIO;
429 }
430 replaced_count = 0;
431 goto unlock_pages;
432 }
433
434 /**
435 * mext_check_arguments - Check whether move extent can be done
436 *
437 * @orig_inode: original inode
438 * @donor_inode: donor inode
439 * @orig_start: logical start offset in block for orig
440 * @donor_start: logical start offset in block for donor
441 * @len: the number of blocks to be moved
442 *
443 * Check the arguments of ext4_move_extents() whether the files can be
444 * exchanged with each other.
445 * Return 0 on success, or a negative error value on failure.
446 */
447 static int
mext_check_arguments(struct inode * orig_inode,struct inode * donor_inode,__u64 orig_start,__u64 donor_start,__u64 * len)448 mext_check_arguments(struct inode *orig_inode,
449 struct inode *donor_inode, __u64 orig_start,
450 __u64 donor_start, __u64 *len)
451 {
452 __u64 orig_eof, donor_eof;
453 unsigned int blkbits = orig_inode->i_blkbits;
454 unsigned int blocksize = 1 << blkbits;
455
456 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
457 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
458
459
460 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
461 ext4_debug("ext4 move extent: suid or sgid is set"
462 " to donor file [ino:orig %lu, donor %lu]\n",
463 orig_inode->i_ino, donor_inode->i_ino);
464 return -EINVAL;
465 }
466
467 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
468 return -EPERM;
469
470 /* Ext4 move extent does not support swapfile */
471 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
472 ext4_debug("ext4 move extent: The argument files should "
473 "not be swapfile [ino:orig %lu, donor %lu]\n",
474 orig_inode->i_ino, donor_inode->i_ino);
475 return -EBUSY;
476 }
477
478 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
479 ext4_debug("ext4 move extent: The argument files should "
480 "not be quota files [ino:orig %lu, donor %lu]\n",
481 orig_inode->i_ino, donor_inode->i_ino);
482 return -EBUSY;
483 }
484
485 /* Ext4 move extent supports only extent based file */
486 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
487 ext4_debug("ext4 move extent: orig file is not extents "
488 "based file [ino:orig %lu]\n", orig_inode->i_ino);
489 return -EOPNOTSUPP;
490 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
491 ext4_debug("ext4 move extent: donor file is not extents "
492 "based file [ino:donor %lu]\n", donor_inode->i_ino);
493 return -EOPNOTSUPP;
494 }
495
496 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
497 ext4_debug("ext4 move extent: File size is 0 byte\n");
498 return -EINVAL;
499 }
500
501 /* Start offset should be same */
502 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
503 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
504 ext4_debug("ext4 move extent: orig and donor's start "
505 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
506 orig_inode->i_ino, donor_inode->i_ino);
507 return -EINVAL;
508 }
509
510 if ((orig_start >= EXT_MAX_BLOCKS) ||
511 (donor_start >= EXT_MAX_BLOCKS) ||
512 (*len > EXT_MAX_BLOCKS) ||
513 (donor_start + *len >= EXT_MAX_BLOCKS) ||
514 (orig_start + *len >= EXT_MAX_BLOCKS)) {
515 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
516 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
517 orig_inode->i_ino, donor_inode->i_ino);
518 return -EINVAL;
519 }
520 if (orig_eof <= orig_start)
521 *len = 0;
522 else if (orig_eof < orig_start + *len - 1)
523 *len = orig_eof - orig_start;
524 if (donor_eof <= donor_start)
525 *len = 0;
526 else if (donor_eof < donor_start + *len - 1)
527 *len = donor_eof - donor_start;
528 if (!*len) {
529 ext4_debug("ext4 move extent: len should not be 0 "
530 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
531 donor_inode->i_ino);
532 return -EINVAL;
533 }
534
535 return 0;
536 }
537
538 /**
539 * ext4_move_extents - Exchange the specified range of a file
540 *
541 * @o_filp: file structure of the original file
542 * @d_filp: file structure of the donor file
543 * @orig_blk: start offset in block for orig
544 * @donor_blk: start offset in block for donor
545 * @len: the number of blocks to be moved
546 * @moved_len: moved block length
547 *
548 * This function returns 0 and moved block length is set in moved_len
549 * if succeed, otherwise returns error value.
550 *
551 */
552 int
ext4_move_extents(struct file * o_filp,struct file * d_filp,__u64 orig_blk,__u64 donor_blk,__u64 len,__u64 * moved_len)553 ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
554 __u64 donor_blk, __u64 len, __u64 *moved_len)
555 {
556 struct inode *orig_inode = file_inode(o_filp);
557 struct inode *donor_inode = file_inode(d_filp);
558 struct ext4_ext_path *path = NULL;
559 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
560 ext4_lblk_t o_end, o_start = orig_blk;
561 ext4_lblk_t d_start = donor_blk;
562 int ret;
563
564 if (orig_inode->i_sb != donor_inode->i_sb) {
565 ext4_debug("ext4 move extent: The argument files "
566 "should be in same FS [ino:orig %lu, donor %lu]\n",
567 orig_inode->i_ino, donor_inode->i_ino);
568 return -EINVAL;
569 }
570
571 /* orig and donor should be different inodes */
572 if (orig_inode == donor_inode) {
573 ext4_debug("ext4 move extent: The argument files should not "
574 "be same inode [ino:orig %lu, donor %lu]\n",
575 orig_inode->i_ino, donor_inode->i_ino);
576 return -EINVAL;
577 }
578
579 /* Regular file check */
580 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
581 ext4_debug("ext4 move extent: The argument files should be "
582 "regular file [ino:orig %lu, donor %lu]\n",
583 orig_inode->i_ino, donor_inode->i_ino);
584 return -EINVAL;
585 }
586
587 /* TODO: it's not obvious how to swap blocks for inodes with full
588 journaling enabled */
589 if (ext4_should_journal_data(orig_inode) ||
590 ext4_should_journal_data(donor_inode)) {
591 ext4_msg(orig_inode->i_sb, KERN_ERR,
592 "Online defrag not supported with data journaling");
593 return -EOPNOTSUPP;
594 }
595
596 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
597 ext4_msg(orig_inode->i_sb, KERN_ERR,
598 "Online defrag not supported for encrypted files");
599 return -EOPNOTSUPP;
600 }
601
602 /* Protect orig and donor inodes against a truncate */
603 lock_two_nondirectories(orig_inode, donor_inode);
604
605 /* Wait for all existing dio workers */
606 inode_dio_wait(orig_inode);
607 inode_dio_wait(donor_inode);
608
609 /* Protect extent tree against block allocations via delalloc */
610 ext4_double_down_write_data_sem(orig_inode, donor_inode);
611 /* Check the filesystem environment whether move_extent can be done */
612 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
613 donor_blk, &len);
614 if (ret)
615 goto out;
616 o_end = o_start + len;
617
618 *moved_len = 0;
619 while (o_start < o_end) {
620 struct ext4_extent *ex;
621 ext4_lblk_t cur_blk, next_blk;
622 pgoff_t orig_page_index, donor_page_index;
623 int offset_in_page;
624 int unwritten, cur_len;
625
626 ret = get_ext_path(orig_inode, o_start, &path);
627 if (ret)
628 goto out;
629 ex = path[path->p_depth].p_ext;
630 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 *moved_len += move_extent_per_page(o_filp, donor_inode,
675 orig_page_index, donor_page_index,
676 offset_in_page, cur_len,
677 unwritten, &ret);
678 ext4_double_down_write_data_sem(orig_inode, donor_inode);
679 if (ret < 0)
680 break;
681 o_start += cur_len;
682 d_start += cur_len;
683 }
684
685 out:
686 if (*moved_len) {
687 ext4_discard_preallocations(orig_inode, 0);
688 ext4_discard_preallocations(donor_inode, 0);
689 }
690
691 ext4_ext_drop_refs(path);
692 kfree(path);
693 ext4_double_up_write_data_sem(orig_inode, donor_inode);
694 unlock_two_nondirectories(orig_inode, donor_inode);
695
696 return ret;
697 }
698