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/*
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_ext_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 **orig_path)
36{
37 int ret = 0;
38 struct ext4_ext_path *path;
39
40 path = ext4_ext_find_extent(inode, lblock, *orig_path, EXT4_EX_NOCACHE);
41 if (IS_ERR(path))
42 ret = PTR_ERR(path);
43 else if (path[ext_depth(inode)].p_ext == NULL)
44 ret = -ENODATA;
45 else
46 *orig_path = path;
47
48 return ret;
49}
50
51/**
52 * copy_extent_status - Copy the extent's initialization status
53 *
54 * @src: an extent for getting initialize status
55 * @dest: an extent to be set the status
56 */
57static void
58copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
59{
60 if (ext4_ext_is_uninitialized(src))
61 ext4_ext_mark_uninitialized(dest);
62 else
63 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
64}
65
66/**
67 * mext_next_extent - Search for the next extent and set it to "extent"
68 *
69 * @inode: inode which is searched
70 * @path: this will obtain data for the next extent
71 * @extent: pointer to the next extent we have just gotten
72 *
73 * Search the next extent in the array of ext4_ext_path structure (@path)
74 * and set it to ext4_extent structure (@extent). In addition, the member of
75 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
76 * ext4_ext_path structure refers to the last extent, or a negative error
77 * value on failure.
78 */
79int
80mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
81 struct ext4_extent **extent)
82{
83 struct ext4_extent_header *eh;
84 int ppos, leaf_ppos = path->p_depth;
85
86 ppos = leaf_ppos;
87 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
88 /* leaf block */
89 *extent = ++path[ppos].p_ext;
90 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
91 return 0;
92 }
93
94 while (--ppos >= 0) {
95 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
96 path[ppos].p_idx) {
97 int cur_ppos = ppos;
98
99 /* index block */
100 path[ppos].p_idx++;
101 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
102 if (path[ppos+1].p_bh)
103 brelse(path[ppos+1].p_bh);
104 path[ppos+1].p_bh =
105 sb_bread(inode->i_sb, path[ppos].p_block);
106 if (!path[ppos+1].p_bh)
107 return -EIO;
108 path[ppos+1].p_hdr =
109 ext_block_hdr(path[ppos+1].p_bh);
110
111 /* Halfway index block */
112 while (++cur_ppos < leaf_ppos) {
113 path[cur_ppos].p_idx =
114 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
115 path[cur_ppos].p_block =
116 ext4_idx_pblock(path[cur_ppos].p_idx);
117 if (path[cur_ppos+1].p_bh)
118 brelse(path[cur_ppos+1].p_bh);
119 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
120 path[cur_ppos].p_block);
121 if (!path[cur_ppos+1].p_bh)
122 return -EIO;
123 path[cur_ppos+1].p_hdr =
124 ext_block_hdr(path[cur_ppos+1].p_bh);
125 }
126
127 path[leaf_ppos].p_ext = *extent = NULL;
128
129 eh = path[leaf_ppos].p_hdr;
130 if (le16_to_cpu(eh->eh_entries) == 0)
131 /* empty leaf is found */
132 return -ENODATA;
133
134 /* leaf block */
135 path[leaf_ppos].p_ext = *extent =
136 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
137 path[leaf_ppos].p_block =
138 ext4_ext_pblock(path[leaf_ppos].p_ext);
139 return 0;
140 }
141 }
142 /* We found the last extent */
143 return 1;
144}
145
146/**
147 * ext4_double_down_write_data_sem - Acquire two inodes' write lock
148 * of i_data_sem
149 *
150 * Acquire write lock of i_data_sem of the two inodes
151 */
152void
153ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
154{
155 if (first < second) {
156 down_write(&EXT4_I(first)->i_data_sem);
157 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
158 } else {
159 down_write(&EXT4_I(second)->i_data_sem);
160 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
161
162 }
163}
164
165/**
166 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
167 *
168 * @orig_inode: original inode structure to be released its lock first
169 * @donor_inode: donor inode structure to be released its lock second
170 * Release write lock of i_data_sem of two inodes (orig and donor).
171 */
172void
173ext4_double_up_write_data_sem(struct inode *orig_inode,
174 struct inode *donor_inode)
175{
176 up_write(&EXT4_I(orig_inode)->i_data_sem);
177 up_write(&EXT4_I(donor_inode)->i_data_sem);
178}
179
180/**
181 * mext_insert_across_blocks - Insert extents across leaf block
182 *
183 * @handle: journal handle
184 * @orig_inode: original inode
185 * @o_start: first original extent to be changed
186 * @o_end: last original extent to be changed
187 * @start_ext: first new extent to be inserted
188 * @new_ext: middle of new extent to be inserted
189 * @end_ext: last new extent to be inserted
190 *
191 * Allocate a new leaf block and insert extents into it. Return 0 on success,
192 * or a negative error value on failure.
193 */
194static int
195mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
196 struct ext4_extent *o_start, struct ext4_extent *o_end,
197 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
198 struct ext4_extent *end_ext)
199{
200 struct ext4_ext_path *orig_path = NULL;
201 ext4_lblk_t eblock = 0;
202 int new_flag = 0;
203 int end_flag = 0;
204 int err = 0;
205
206 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
207 if (o_start == o_end) {
208
209 /* start_ext new_ext end_ext
210 * donor |---------|-----------|--------|
211 * orig |------------------------------|
212 */
213 end_flag = 1;
214 } else {
215
216 /* start_ext new_ext end_ext
217 * donor |---------|----------|---------|
218 * orig |---------------|--------------|
219 */
220 o_end->ee_block = end_ext->ee_block;
221 o_end->ee_len = end_ext->ee_len;
222 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
223 }
224
225 o_start->ee_len = start_ext->ee_len;
226 eblock = le32_to_cpu(start_ext->ee_block);
227 new_flag = 1;
228
229 } else if (start_ext->ee_len && new_ext->ee_len &&
230 !end_ext->ee_len && o_start == o_end) {
231
232 /* start_ext new_ext
233 * donor |--------------|---------------|
234 * orig |------------------------------|
235 */
236 o_start->ee_len = start_ext->ee_len;
237 eblock = le32_to_cpu(start_ext->ee_block);
238 new_flag = 1;
239
240 } else if (!start_ext->ee_len && new_ext->ee_len &&
241 end_ext->ee_len && o_start == o_end) {
242
243 /* new_ext end_ext
244 * donor |--------------|---------------|
245 * orig |------------------------------|
246 */
247 o_end->ee_block = end_ext->ee_block;
248 o_end->ee_len = end_ext->ee_len;
249 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
250
251 /*
252 * Set 0 to the extent block if new_ext was
253 * the first block.
254 */
255 if (new_ext->ee_block)
256 eblock = le32_to_cpu(new_ext->ee_block);
257
258 new_flag = 1;
259 } else {
260 ext4_debug("ext4 move extent: Unexpected insert case\n");
261 return -EIO;
262 }
263
264 if (new_flag) {
265 err = get_ext_path(orig_inode, eblock, &orig_path);
266 if (err)
267 goto out;
268
269 if (ext4_ext_insert_extent(handle, orig_inode,
270 orig_path, new_ext, 0))
271 goto out;
272 }
273
274 if (end_flag) {
275 err = get_ext_path(orig_inode,
276 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
277 if (err)
278 goto out;
279
280 if (ext4_ext_insert_extent(handle, orig_inode,
281 orig_path, end_ext, 0))
282 goto out;
283 }
284out:
285 if (orig_path) {
286 ext4_ext_drop_refs(orig_path);
287 kfree(orig_path);
288 }
289
290 return err;
291
292}
293
294/**
295 * mext_insert_inside_block - Insert new extent to the extent block
296 *
297 * @o_start: first original extent to be moved
298 * @o_end: last original extent to be moved
299 * @start_ext: first new extent to be inserted
300 * @new_ext: middle of new extent to be inserted
301 * @end_ext: last new extent to be inserted
302 * @eh: extent header of target leaf block
303 * @range_to_move: used to decide how to insert extent
304 *
305 * Insert extents into the leaf block. The extent (@o_start) is overwritten
306 * by inserted extents.
307 */
308static void
309mext_insert_inside_block(struct ext4_extent *o_start,
310 struct ext4_extent *o_end,
311 struct ext4_extent *start_ext,
312 struct ext4_extent *new_ext,
313 struct ext4_extent *end_ext,
314 struct ext4_extent_header *eh,
315 int range_to_move)
316{
317 int i = 0;
318 unsigned long len;
319
320 /* Move the existing extents */
321 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
322 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
323 (unsigned long)(o_end + 1);
324 memmove(o_end + 1 + range_to_move, o_end + 1, len);
325 }
326
327 /* Insert start entry */
328 if (start_ext->ee_len)
329 o_start[i++].ee_len = start_ext->ee_len;
330
331 /* Insert new entry */
332 if (new_ext->ee_len) {
333 o_start[i] = *new_ext;
334 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
335 }
336
337 /* Insert end entry */
338 if (end_ext->ee_len)
339 o_start[i] = *end_ext;
340
341 /* Increment the total entries counter on the extent block */
342 le16_add_cpu(&eh->eh_entries, range_to_move);
343}
344
345/**
346 * mext_insert_extents - Insert new extent
347 *
348 * @handle: journal handle
349 * @orig_inode: original inode
350 * @orig_path: path indicates first extent to be changed
351 * @o_start: first original extent to be changed
352 * @o_end: last original extent to be changed
353 * @start_ext: first new extent to be inserted
354 * @new_ext: middle of new extent to be inserted
355 * @end_ext: last new extent to be inserted
356 *
357 * Call the function to insert extents. If we cannot add more extents into
358 * the leaf block, we call mext_insert_across_blocks() to create a
359 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
360 * on success, or a negative error value on failure.
361 */
362static int
363mext_insert_extents(handle_t *handle, struct inode *orig_inode,
364 struct ext4_ext_path *orig_path,
365 struct ext4_extent *o_start,
366 struct ext4_extent *o_end,
367 struct ext4_extent *start_ext,
368 struct ext4_extent *new_ext,
369 struct ext4_extent *end_ext)
370{
371 struct ext4_extent_header *eh;
372 unsigned long need_slots, slots_range;
373 int range_to_move, depth, ret;
374
375 /*
376 * The extents need to be inserted
377 * start_extent + new_extent + end_extent.
378 */
379 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
380 (new_ext->ee_len ? 1 : 0);
381
382 /* The number of slots between start and end */
383 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
384 / sizeof(struct ext4_extent);
385
386 /* Range to move the end of extent */
387 range_to_move = need_slots - slots_range;
388 depth = orig_path->p_depth;
389 orig_path += depth;
390 eh = orig_path->p_hdr;
391
392 if (depth) {
393 /* Register to journal */
394 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
395 if (ret)
396 return ret;
397 }
398
399 /* Expansion */
400 if (range_to_move > 0 &&
401 (range_to_move > le16_to_cpu(eh->eh_max)
402 - le16_to_cpu(eh->eh_entries))) {
403
404 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
405 o_end, start_ext, new_ext, end_ext);
406 if (ret < 0)
407 return ret;
408 } else
409 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
410 end_ext, eh, range_to_move);
411
412 return ext4_ext_dirty(handle, orig_inode, orig_path);
413}
414
415/**
416 * mext_leaf_block - Move one leaf extent block into the inode.
417 *
418 * @handle: journal handle
419 * @orig_inode: original inode
420 * @orig_path: path indicates first extent to be changed
421 * @dext: donor extent
422 * @from: start offset on the target file
423 *
424 * In order to insert extents into the leaf block, we must divide the extent
425 * in the leaf block into three extents. The one is located to be inserted
426 * extents, and the others are located around it.
427 *
428 * Therefore, this function creates structures to save extents of the leaf
429 * block, and inserts extents by calling mext_insert_extents() with
430 * created extents. Return 0 on success, or a negative error value on failure.
431 */
432static int
433mext_leaf_block(handle_t *handle, struct inode *orig_inode,
434 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
435 ext4_lblk_t *from)
436{
437 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
438 struct ext4_extent new_ext, start_ext, end_ext;
439 ext4_lblk_t new_ext_end;
440 int oext_alen, new_ext_alen, end_ext_alen;
441 int depth = ext_depth(orig_inode);
442 int ret;
443
444 start_ext.ee_block = end_ext.ee_block = 0;
445 o_start = o_end = oext = orig_path[depth].p_ext;
446 oext_alen = ext4_ext_get_actual_len(oext);
447 start_ext.ee_len = end_ext.ee_len = 0;
448
449 new_ext.ee_block = cpu_to_le32(*from);
450 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
451 new_ext.ee_len = dext->ee_len;
452 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
453 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
454
455 /*
456 * Case: original extent is first
457 * oext |--------|
458 * new_ext |--|
459 * start_ext |--|
460 */
461 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
462 le32_to_cpu(new_ext.ee_block) <
463 le32_to_cpu(oext->ee_block) + oext_alen) {
464 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
465 le32_to_cpu(oext->ee_block));
466 start_ext.ee_block = oext->ee_block;
467 copy_extent_status(oext, &start_ext);
468 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
469 prev_ext = oext - 1;
470 /*
471 * We can merge new_ext into previous extent,
472 * if these are contiguous and same extent type.
473 */
474 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
475 &new_ext)) {
476 o_start = prev_ext;
477 start_ext.ee_len = cpu_to_le16(
478 ext4_ext_get_actual_len(prev_ext) +
479 new_ext_alen);
480 start_ext.ee_block = oext->ee_block;
481 copy_extent_status(prev_ext, &start_ext);
482 new_ext.ee_len = 0;
483 }
484 }
485
486 /*
487 * Case: new_ext_end must be less than oext
488 * oext |-----------|
489 * new_ext |-------|
490 */
491 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
492 EXT4_ERROR_INODE(orig_inode,
493 "new_ext_end(%u) should be less than or equal to "
494 "oext->ee_block(%u) + oext_alen(%d) - 1",
495 new_ext_end, le32_to_cpu(oext->ee_block),
496 oext_alen);
497 ret = -EIO;
498 goto out;
499 }
500
501 /*
502 * Case: new_ext is smaller than original extent
503 * oext |---------------|
504 * new_ext |-----------|
505 * end_ext |---|
506 */
507 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
508 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
509 end_ext.ee_len =
510 cpu_to_le16(le32_to_cpu(oext->ee_block) +
511 oext_alen - 1 - new_ext_end);
512 copy_extent_status(oext, &end_ext);
513 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
514 ext4_ext_store_pblock(&end_ext,
515 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
516 end_ext.ee_block =
517 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
518 oext_alen - end_ext_alen);
519 }
520
521 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
522 o_end, &start_ext, &new_ext, &end_ext);
523out:
524 return ret;
525}
526
527/**
528 * mext_calc_swap_extents - Calculate extents for extent swapping.
529 *
530 * @tmp_dext: the extent that will belong to the original inode
531 * @tmp_oext: the extent that will belong to the donor inode
532 * @orig_off: block offset of original inode
533 * @donor_off: block offset of donor inode
534 * @max_count: the maximum length of extents
535 *
536 * Return 0 on success, or a negative error value on failure.
537 */
538static int
539mext_calc_swap_extents(struct ext4_extent *tmp_dext,
540 struct ext4_extent *tmp_oext,
541 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
542 ext4_lblk_t max_count)
543{
544 ext4_lblk_t diff, orig_diff;
545 struct ext4_extent dext_old, oext_old;
546
547 BUG_ON(orig_off != donor_off);
548
549 /* original and donor extents have to cover the same block offset */
550 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
551 le32_to_cpu(tmp_oext->ee_block) +
552 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
553 return -ENODATA;
554
555 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
556 le32_to_cpu(tmp_dext->ee_block) +
557 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
558 return -ENODATA;
559
560 dext_old = *tmp_dext;
561 oext_old = *tmp_oext;
562
563 /* When tmp_dext is too large, pick up the target range. */
564 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
565
566 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
567 le32_add_cpu(&tmp_dext->ee_block, diff);
568 le16_add_cpu(&tmp_dext->ee_len, -diff);
569
570 if (max_count < ext4_ext_get_actual_len(tmp_dext))
571 tmp_dext->ee_len = cpu_to_le16(max_count);
572
573 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
574 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
575
576 /* Adjust extent length if donor extent is larger than orig */
577 if (ext4_ext_get_actual_len(tmp_dext) >
578 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
579 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
580 orig_diff);
581
582 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
583
584 copy_extent_status(&oext_old, tmp_dext);
585 copy_extent_status(&dext_old, tmp_oext);
586
587 return 0;
588}
589
590/**
591 * mext_check_coverage - Check that all extents in range has the same type
592 *
593 * @inode: inode in question
594 * @from: block offset of inode
595 * @count: block count to be checked
596 * @uninit: extents expected to be uninitialized
597 * @err: pointer to save error value
598 *
599 * Return 1 if all extents in range has expected type, and zero otherwise.
600 */
601static int
602mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
603 int uninit, int *err)
604{
605 struct ext4_ext_path *path = NULL;
606 struct ext4_extent *ext;
607 int ret = 0;
608 ext4_lblk_t last = from + count;
609 while (from < last) {
610 *err = get_ext_path(inode, from, &path);
611 if (*err)
612 goto out;
613 ext = path[ext_depth(inode)].p_ext;
614 if (uninit != ext4_ext_is_uninitialized(ext))
615 goto out;
616 from += ext4_ext_get_actual_len(ext);
617 ext4_ext_drop_refs(path);
618 }
619 ret = 1;
620out:
621 if (path) {
622 ext4_ext_drop_refs(path);
623 kfree(path);
624 }
625 return ret;
626}
627
628/**
629 * mext_replace_branches - Replace original extents with new extents
630 *
631 * @handle: journal handle
632 * @orig_inode: original inode
633 * @donor_inode: donor inode
634 * @from: block offset of orig_inode
635 * @count: block count to be replaced
636 * @err: pointer to save return value
637 *
638 * Replace original inode extents and donor inode extents page by page.
639 * We implement this replacement in the following three steps:
640 * 1. Save the block information of original and donor inodes into
641 * dummy extents.
642 * 2. Change the block information of original inode to point at the
643 * donor inode blocks.
644 * 3. Change the block information of donor inode to point at the saved
645 * original inode blocks in the dummy extents.
646 *
647 * Return replaced block count.
648 */
649static int
650mext_replace_branches(handle_t *handle, struct inode *orig_inode,
651 struct inode *donor_inode, ext4_lblk_t from,
652 ext4_lblk_t count, int *err)
653{
654 struct ext4_ext_path *orig_path = NULL;
655 struct ext4_ext_path *donor_path = NULL;
656 struct ext4_extent *oext, *dext;
657 struct ext4_extent tmp_dext, tmp_oext;
658 ext4_lblk_t orig_off = from, donor_off = from;
659 int depth;
660 int replaced_count = 0;
661 int dext_alen;
662
663 *err = ext4_es_remove_extent(orig_inode, from, count);
664 if (*err)
665 goto out;
666
667 *err = ext4_es_remove_extent(donor_inode, from, count);
668 if (*err)
669 goto out;
670
671 /* Get the original extent for the block "orig_off" */
672 *err = get_ext_path(orig_inode, orig_off, &orig_path);
673 if (*err)
674 goto out;
675
676 /* Get the donor extent for the head */
677 *err = get_ext_path(donor_inode, donor_off, &donor_path);
678 if (*err)
679 goto out;
680 depth = ext_depth(orig_inode);
681 oext = orig_path[depth].p_ext;
682 tmp_oext = *oext;
683
684 depth = ext_depth(donor_inode);
685 dext = donor_path[depth].p_ext;
686 if (unlikely(!dext))
687 goto missing_donor_extent;
688 tmp_dext = *dext;
689
690 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
691 donor_off, count);
692 if (*err)
693 goto out;
694
695 /* Loop for the donor extents */
696 while (1) {
697 /* The extent for donor must be found. */
698 if (unlikely(!dext)) {
699 missing_donor_extent:
700 EXT4_ERROR_INODE(donor_inode,
701 "The extent for donor must be found");
702 *err = -EIO;
703 goto out;
704 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
705 EXT4_ERROR_INODE(donor_inode,
706 "Donor offset(%u) and the first block of donor "
707 "extent(%u) should be equal",
708 donor_off,
709 le32_to_cpu(tmp_dext.ee_block));
710 *err = -EIO;
711 goto out;
712 }
713
714 /* Set donor extent to orig extent */
715 *err = mext_leaf_block(handle, orig_inode,
716 orig_path, &tmp_dext, &orig_off);
717 if (*err)
718 goto out;
719
720 /* Set orig extent to donor extent */
721 *err = mext_leaf_block(handle, donor_inode,
722 donor_path, &tmp_oext, &donor_off);
723 if (*err)
724 goto out;
725
726 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
727 replaced_count += dext_alen;
728 donor_off += dext_alen;
729 orig_off += dext_alen;
730
731 BUG_ON(replaced_count > count);
732 /* Already moved the expected blocks */
733 if (replaced_count >= count)
734 break;
735
736 if (orig_path)
737 ext4_ext_drop_refs(orig_path);
738 *err = get_ext_path(orig_inode, orig_off, &orig_path);
739 if (*err)
740 goto out;
741 depth = ext_depth(orig_inode);
742 oext = orig_path[depth].p_ext;
743 tmp_oext = *oext;
744
745 if (donor_path)
746 ext4_ext_drop_refs(donor_path);
747 *err = get_ext_path(donor_inode, donor_off, &donor_path);
748 if (*err)
749 goto out;
750 depth = ext_depth(donor_inode);
751 dext = donor_path[depth].p_ext;
752 tmp_dext = *dext;
753
754 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
755 donor_off, count - replaced_count);
756 if (*err)
757 goto out;
758 }
759
760out:
761 if (orig_path) {
762 ext4_ext_drop_refs(orig_path);
763 kfree(orig_path);
764 }
765 if (donor_path) {
766 ext4_ext_drop_refs(donor_path);
767 kfree(donor_path);
768 }
769
770 return replaced_count;
771}
772
773/**
774 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
775 *
776 * @inode1: the inode structure
777 * @inode2: the inode structure
778 * @index: page index
779 * @page: result page vector
780 *
781 * Grab two locked pages for inode's by inode order
782 */
783static int
784mext_page_double_lock(struct inode *inode1, struct inode *inode2,
785 pgoff_t index, struct page *page[2])
786{
787 struct address_space *mapping[2];
788 unsigned fl = AOP_FLAG_NOFS;
789
790 BUG_ON(!inode1 || !inode2);
791 if (inode1 < inode2) {
792 mapping[0] = inode1->i_mapping;
793 mapping[1] = inode2->i_mapping;
794 } else {
795 mapping[0] = inode2->i_mapping;
796 mapping[1] = inode1->i_mapping;
797 }
798
799 page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
800 if (!page[0])
801 return -ENOMEM;
802
803 page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
804 if (!page[1]) {
805 unlock_page(page[0]);
806 page_cache_release(page[0]);
807 return -ENOMEM;
808 }
809 /*
810 * grab_cache_page_write_begin() may not wait on page's writeback if
811 * BDI not demand that. But it is reasonable to be very conservative
812 * here and explicitly wait on page's writeback
813 */
814 wait_on_page_writeback(page[0]);
815 wait_on_page_writeback(page[1]);
816 if (inode1 > inode2) {
817 struct page *tmp;
818 tmp = page[0];
819 page[0] = page[1];
820 page[1] = tmp;
821 }
822 return 0;
823}
824
825/* Force page buffers uptodate w/o dropping page's lock */
826static int
827mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
828{
829 struct inode *inode = page->mapping->host;
830 sector_t block;
831 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
832 unsigned int blocksize, block_start, block_end;
833 int i, err, nr = 0, partial = 0;
834 BUG_ON(!PageLocked(page));
835 BUG_ON(PageWriteback(page));
836
837 if (PageUptodate(page))
838 return 0;
839
840 blocksize = 1 << inode->i_blkbits;
841 if (!page_has_buffers(page))
842 create_empty_buffers(page, blocksize, 0);
843
844 head = page_buffers(page);
845 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
846 for (bh = head, block_start = 0; bh != head || !block_start;
847 block++, block_start = block_end, bh = bh->b_this_page) {
848 block_end = block_start + blocksize;
849 if (block_end <= from || block_start >= to) {
850 if (!buffer_uptodate(bh))
851 partial = 1;
852 continue;
853 }
854 if (buffer_uptodate(bh))
855 continue;
856 if (!buffer_mapped(bh)) {
857 err = ext4_get_block(inode, block, bh, 0);
858 if (err) {
859 SetPageError(page);
860 return err;
861 }
862 if (!buffer_mapped(bh)) {
863 zero_user(page, block_start, blocksize);
864 set_buffer_uptodate(bh);
865 continue;
866 }
867 }
868 BUG_ON(nr >= MAX_BUF_PER_PAGE);
869 arr[nr++] = bh;
870 }
871 /* No io required */
872 if (!nr)
873 goto out;
874
875 for (i = 0; i < nr; i++) {
876 bh = arr[i];
877 if (!bh_uptodate_or_lock(bh)) {
878 err = bh_submit_read(bh);
879 if (err)
880 return err;
881 }
882 }
883out:
884 if (!partial)
885 SetPageUptodate(page);
886 return 0;
887}
888
889/**
890 * move_extent_per_page - Move extent data per page
891 *
892 * @o_filp: file structure of original file
893 * @donor_inode: donor inode
894 * @orig_page_offset: page index on original file
895 * @data_offset_in_page: block index where data swapping starts
896 * @block_len_in_page: the number of blocks to be swapped
897 * @uninit: orig extent is uninitialized or not
898 * @err: pointer to save return value
899 *
900 * Save the data in original inode blocks and replace original inode extents
901 * with donor inode extents by calling mext_replace_branches().
902 * Finally, write out the saved data in new original inode blocks. Return
903 * replaced block count.
904 */
905static int
906move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
907 pgoff_t orig_page_offset, int data_offset_in_page,
908 int block_len_in_page, int uninit, int *err)
909{
910 struct inode *orig_inode = file_inode(o_filp);
911 struct page *pagep[2] = {NULL, NULL};
912 handle_t *handle;
913 ext4_lblk_t orig_blk_offset;
914 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
915 unsigned int w_flags = 0;
916 unsigned int tmp_data_size, data_size, replaced_size;
917 int err2, jblocks, retries = 0;
918 int replaced_count = 0;
919 int from = data_offset_in_page << orig_inode->i_blkbits;
920 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
921
922 /*
923 * It needs twice the amount of ordinary journal buffers because
924 * inode and donor_inode may change each different metadata blocks.
925 */
926again:
927 *err = 0;
928 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
929 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
930 if (IS_ERR(handle)) {
931 *err = PTR_ERR(handle);
932 return 0;
933 }
934
935 if (segment_eq(get_fs(), KERNEL_DS))
936 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
937
938 orig_blk_offset = orig_page_offset * blocks_per_page +
939 data_offset_in_page;
940
941 /* Calculate data_size */
942 if ((orig_blk_offset + block_len_in_page - 1) ==
943 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
944 /* Replace the last block */
945 tmp_data_size = orig_inode->i_size & (blocksize - 1);
946 /*
947 * If data_size equal zero, it shows data_size is multiples of
948 * blocksize. So we set appropriate value.
949 */
950 if (tmp_data_size == 0)
951 tmp_data_size = blocksize;
952
953 data_size = tmp_data_size +
954 ((block_len_in_page - 1) << orig_inode->i_blkbits);
955 } else
956 data_size = block_len_in_page << orig_inode->i_blkbits;
957
958 replaced_size = data_size;
959
960 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
961 pagep);
962 if (unlikely(*err < 0))
963 goto stop_journal;
964 /*
965 * If orig extent was uninitialized it can become initialized
966 * at any time after i_data_sem was dropped, in order to
967 * serialize with delalloc we have recheck extent while we
968 * hold page's lock, if it is still the case data copy is not
969 * necessary, just swap data blocks between orig and donor.
970 */
971 if (uninit) {
972 ext4_double_down_write_data_sem(orig_inode, donor_inode);
973 /* If any of extents in range became initialized we have to
974 * fallback to data copying */
975 uninit = mext_check_coverage(orig_inode, orig_blk_offset,
976 block_len_in_page, 1, err);
977 if (*err)
978 goto drop_data_sem;
979
980 uninit &= mext_check_coverage(donor_inode, orig_blk_offset,
981 block_len_in_page, 1, err);
982 if (*err)
983 goto drop_data_sem;
984
985 if (!uninit) {
986 ext4_double_up_write_data_sem(orig_inode, donor_inode);
987 goto data_copy;
988 }
989 if ((page_has_private(pagep[0]) &&
990 !try_to_release_page(pagep[0], 0)) ||
991 (page_has_private(pagep[1]) &&
992 !try_to_release_page(pagep[1], 0))) {
993 *err = -EBUSY;
994 goto drop_data_sem;
995 }
996 replaced_count = mext_replace_branches(handle, orig_inode,
997 donor_inode, orig_blk_offset,
998 block_len_in_page, err);
999 drop_data_sem:
1000 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1001 goto unlock_pages;
1002 }
1003data_copy:
1004 *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1005 if (*err)
1006 goto unlock_pages;
1007
1008 /* At this point all buffers in range are uptodate, old mapping layout
1009 * is no longer required, try to drop it now. */
1010 if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1011 (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1012 *err = -EBUSY;
1013 goto unlock_pages;
1014 }
1015
1016 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1017 orig_blk_offset,
1018 block_len_in_page, err);
1019 if (*err) {
1020 if (replaced_count) {
1021 block_len_in_page = replaced_count;
1022 replaced_size =
1023 block_len_in_page << orig_inode->i_blkbits;
1024 } else
1025 goto unlock_pages;
1026 }
1027 /* Perform all necessary steps similar write_begin()/write_end()
1028 * but keeping in mind that i_size will not change */
1029 *err = __block_write_begin(pagep[0], from, replaced_size,
1030 ext4_get_block);
1031 if (!*err)
1032 *err = block_commit_write(pagep[0], from, from + replaced_size);
1033
1034 if (unlikely(*err < 0))
1035 goto repair_branches;
1036
1037 /* Even in case of data=writeback it is reasonable to pin
1038 * inode to transaction, to prevent unexpected data loss */
1039 *err = ext4_jbd2_file_inode(handle, orig_inode);
1040
1041unlock_pages:
1042 unlock_page(pagep[0]);
1043 page_cache_release(pagep[0]);
1044 unlock_page(pagep[1]);
1045 page_cache_release(pagep[1]);
1046stop_journal:
1047 ext4_journal_stop(handle);
1048 /* Buffer was busy because probably is pinned to journal transaction,
1049 * force transaction commit may help to free it. */
1050 if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1051 &retries))
1052 goto again;
1053 return replaced_count;
1054
1055repair_branches:
1056 /*
1057 * This should never ever happen!
1058 * Extents are swapped already, but we are not able to copy data.
1059 * Try to swap extents to it's original places
1060 */
1061 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1062 replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1063 orig_blk_offset,
1064 block_len_in_page, &err2);
1065 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1066 if (replaced_count != block_len_in_page) {
1067 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1068 "Unable to copy data block,"
1069 " data will be lost.");
1070 *err = -EIO;
1071 }
1072 replaced_count = 0;
1073 goto unlock_pages;
1074}
1075
1076/**
1077 * mext_check_arguments - Check whether move extent can be done
1078 *
1079 * @orig_inode: original inode
1080 * @donor_inode: donor inode
1081 * @orig_start: logical start offset in block for orig
1082 * @donor_start: logical start offset in block for donor
1083 * @len: the number of blocks to be moved
1084 *
1085 * Check the arguments of ext4_move_extents() whether the files can be
1086 * exchanged with each other.
1087 * Return 0 on success, or a negative error value on failure.
1088 */
1089static int
1090mext_check_arguments(struct inode *orig_inode,
1091 struct inode *donor_inode, __u64 orig_start,
1092 __u64 donor_start, __u64 *len)
1093{
1094 ext4_lblk_t orig_blocks, donor_blocks;
1095 unsigned int blkbits = orig_inode->i_blkbits;
1096 unsigned int blocksize = 1 << blkbits;
1097
1098 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1099 ext4_debug("ext4 move extent: suid or sgid is set"
1100 " to donor file [ino:orig %lu, donor %lu]\n",
1101 orig_inode->i_ino, donor_inode->i_ino);
1102 return -EINVAL;
1103 }
1104
1105 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1106 return -EPERM;
1107
1108 /* Ext4 move extent does not support swapfile */
1109 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1110 ext4_debug("ext4 move extent: The argument files should "
1111 "not be swapfile [ino:orig %lu, donor %lu]\n",
1112 orig_inode->i_ino, donor_inode->i_ino);
1113 return -EINVAL;
1114 }
1115
1116 /* Ext4 move extent supports only extent based file */
1117 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1118 ext4_debug("ext4 move extent: orig file is not extents "
1119 "based file [ino:orig %lu]\n", orig_inode->i_ino);
1120 return -EOPNOTSUPP;
1121 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1122 ext4_debug("ext4 move extent: donor file is not extents "
1123 "based file [ino:donor %lu]\n", donor_inode->i_ino);
1124 return -EOPNOTSUPP;
1125 }
1126
1127 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1128 ext4_debug("ext4 move extent: File size is 0 byte\n");
1129 return -EINVAL;
1130 }
1131
1132 /* Start offset should be same */
1133 if (orig_start != donor_start) {
1134 ext4_debug("ext4 move extent: orig and donor's start "
1135 "offset are not same [ino:orig %lu, donor %lu]\n",
1136 orig_inode->i_ino, donor_inode->i_ino);
1137 return -EINVAL;
1138 }
1139
1140 if ((orig_start >= EXT_MAX_BLOCKS) ||
1141 (*len > EXT_MAX_BLOCKS) ||
1142 (orig_start + *len >= EXT_MAX_BLOCKS)) {
1143 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1144 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1145 orig_inode->i_ino, donor_inode->i_ino);
1146 return -EINVAL;
1147 }
1148
1149 if (orig_inode->i_size > donor_inode->i_size) {
1150 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1151 /* TODO: eliminate this artificial restriction */
1152 if (orig_start >= donor_blocks) {
1153 ext4_debug("ext4 move extent: orig start offset "
1154 "[%llu] should be less than donor file blocks "
1155 "[%u] [ino:orig %lu, donor %lu]\n",
1156 orig_start, donor_blocks,
1157 orig_inode->i_ino, donor_inode->i_ino);
1158 return -EINVAL;
1159 }
1160
1161 /* TODO: eliminate this artificial restriction */
1162 if (orig_start + *len > donor_blocks) {
1163 ext4_debug("ext4 move extent: End offset [%llu] should "
1164 "be less than donor file blocks [%u]."
1165 "So adjust length from %llu to %llu "
1166 "[ino:orig %lu, donor %lu]\n",
1167 orig_start + *len, donor_blocks,
1168 *len, donor_blocks - orig_start,
1169 orig_inode->i_ino, donor_inode->i_ino);
1170 *len = donor_blocks - orig_start;
1171 }
1172 } else {
1173 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1174 if (orig_start >= orig_blocks) {
1175 ext4_debug("ext4 move extent: start offset [%llu] "
1176 "should be less than original file blocks "
1177 "[%u] [ino:orig %lu, donor %lu]\n",
1178 orig_start, orig_blocks,
1179 orig_inode->i_ino, donor_inode->i_ino);
1180 return -EINVAL;
1181 }
1182
1183 if (orig_start + *len > orig_blocks) {
1184 ext4_debug("ext4 move extent: Adjust length "
1185 "from %llu to %llu. Because it should be "
1186 "less than original file blocks "
1187 "[ino:orig %lu, donor %lu]\n",
1188 *len, orig_blocks - orig_start,
1189 orig_inode->i_ino, donor_inode->i_ino);
1190 *len = orig_blocks - orig_start;
1191 }
1192 }
1193
1194 if (!*len) {
1195 ext4_debug("ext4 move extent: len should not be 0 "
1196 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1197 donor_inode->i_ino);
1198 return -EINVAL;
1199 }
1200
1201 return 0;
1202}
1203
1204/**
1205 * ext4_move_extents - Exchange the specified range of a file
1206 *
1207 * @o_filp: file structure of the original file
1208 * @d_filp: file structure of the donor file
1209 * @orig_start: start offset in block for orig
1210 * @donor_start: start offset in block for donor
1211 * @len: the number of blocks to be moved
1212 * @moved_len: moved block length
1213 *
1214 * This function returns 0 and moved block length is set in moved_len
1215 * if succeed, otherwise returns error value.
1216 *
1217 * Note: ext4_move_extents() proceeds the following order.
1218 * 1:ext4_move_extents() calculates the last block number of moving extent
1219 * function by the start block number (orig_start) and the number of blocks
1220 * to be moved (len) specified as arguments.
1221 * If the {orig, donor}_start points a hole, the extent's start offset
1222 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1223 * after hole behind.
1224 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1225 * or the ext_cur exceeds the block_end which is last logical block number.
1226 * 3:To get the length of continues area, call mext_next_extent()
1227 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1228 * until find un-continuous extent, the start logical block number exceeds
1229 * the block_end or the extent points to the last extent.
1230 * 4:Exchange the original inode data with donor inode data
1231 * from orig_page_offset to seq_end_page.
1232 * The start indexes of data are specified as arguments.
1233 * That of the original inode is orig_page_offset,
1234 * and the donor inode is also orig_page_offset
1235 * (To easily handle blocksize != pagesize case, the offset for the
1236 * donor inode is block unit).
1237 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1238 * then returns to step 2.
1239 * 6:Release holecheck_path, orig_path and set the len to moved_len
1240 * which shows the number of moved blocks.
1241 * The moved_len is useful for the command to calculate the file offset
1242 * for starting next move extent ioctl.
1243 * 7:Return 0 on success, or a negative error value on failure.
1244 */
1245int
1246ext4_move_extents(struct file *o_filp, struct file *d_filp,
1247 __u64 orig_start, __u64 donor_start, __u64 len,
1248 __u64 *moved_len)
1249{
1250 struct inode *orig_inode = file_inode(o_filp);
1251 struct inode *donor_inode = file_inode(d_filp);
1252 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1253 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1254 ext4_lblk_t block_start = orig_start;
1255 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1256 ext4_lblk_t rest_blocks;
1257 pgoff_t orig_page_offset = 0, seq_end_page;
1258 int ret, depth, last_extent = 0;
1259 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1260 int data_offset_in_page;
1261 int block_len_in_page;
1262 int uninit;
1263
1264 if (orig_inode->i_sb != donor_inode->i_sb) {
1265 ext4_debug("ext4 move extent: The argument files "
1266 "should be in same FS [ino:orig %lu, donor %lu]\n",
1267 orig_inode->i_ino, donor_inode->i_ino);
1268 return -EINVAL;
1269 }
1270
1271 /* orig and donor should be different inodes */
1272 if (orig_inode == donor_inode) {
1273 ext4_debug("ext4 move extent: The argument files should not "
1274 "be same inode [ino:orig %lu, donor %lu]\n",
1275 orig_inode->i_ino, donor_inode->i_ino);
1276 return -EINVAL;
1277 }
1278
1279 /* Regular file check */
1280 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1281 ext4_debug("ext4 move extent: The argument files should be "
1282 "regular file [ino:orig %lu, donor %lu]\n",
1283 orig_inode->i_ino, donor_inode->i_ino);
1284 return -EINVAL;
1285 }
1286 /* TODO: This is non obvious task to swap blocks for inodes with full
1287 jornaling enabled */
1288 if (ext4_should_journal_data(orig_inode) ||
1289 ext4_should_journal_data(donor_inode)) {
1290 return -EINVAL;
1291 }
1292 /* Protect orig and donor inodes against a truncate */
1293 lock_two_nondirectories(orig_inode, donor_inode);
1294
1295 /* Wait for all existing dio workers */
1296 ext4_inode_block_unlocked_dio(orig_inode);
1297 ext4_inode_block_unlocked_dio(donor_inode);
1298 inode_dio_wait(orig_inode);
1299 inode_dio_wait(donor_inode);
1300
1301 /* Protect extent tree against block allocations via delalloc */
1302 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1303 /* Check the filesystem environment whether move_extent can be done */
1304 ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1305 donor_start, &len);
1306 if (ret)
1307 goto out;
1308
1309 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1310 block_end = block_start + len - 1;
1311 if (file_end < block_end)
1312 len -= block_end - file_end;
1313
1314 ret = get_ext_path(orig_inode, block_start, &orig_path);
1315 if (ret)
1316 goto out;
1317
1318 /* Get path structure to check the hole */
1319 ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1320 if (ret)
1321 goto out;
1322
1323 depth = ext_depth(orig_inode);
1324 ext_cur = holecheck_path[depth].p_ext;
1325
1326 /*
1327 * Get proper starting location of block replacement if block_start was
1328 * within the hole.
1329 */
1330 if (le32_to_cpu(ext_cur->ee_block) +
1331 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1332 /*
1333 * The hole exists between extents or the tail of
1334 * original file.
1335 */
1336 last_extent = mext_next_extent(orig_inode,
1337 holecheck_path, &ext_cur);
1338 if (last_extent < 0) {
1339 ret = last_extent;
1340 goto out;
1341 }
1342 last_extent = mext_next_extent(orig_inode, orig_path,
1343 &ext_dummy);
1344 if (last_extent < 0) {
1345 ret = last_extent;
1346 goto out;
1347 }
1348 seq_start = le32_to_cpu(ext_cur->ee_block);
1349 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1350 /* The hole exists at the beginning of original file. */
1351 seq_start = le32_to_cpu(ext_cur->ee_block);
1352 else
1353 seq_start = block_start;
1354
1355 /* No blocks within the specified range. */
1356 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1357 ext4_debug("ext4 move extent: The specified range of file "
1358 "may be the hole\n");
1359 ret = -EINVAL;
1360 goto out;
1361 }
1362
1363 /* Adjust start blocks */
1364 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1365 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1366 max(le32_to_cpu(ext_cur->ee_block), block_start);
1367
1368 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1369 seq_blocks += add_blocks;
1370
1371 /* Adjust tail blocks */
1372 if (seq_start + seq_blocks - 1 > block_end)
1373 seq_blocks = block_end - seq_start + 1;
1374
1375 ext_prev = ext_cur;
1376 last_extent = mext_next_extent(orig_inode, holecheck_path,
1377 &ext_cur);
1378 if (last_extent < 0) {
1379 ret = last_extent;
1380 break;
1381 }
1382 add_blocks = ext4_ext_get_actual_len(ext_cur);
1383
1384 /*
1385 * Extend the length of contiguous block (seq_blocks)
1386 * if extents are contiguous.
1387 */
1388 if (ext4_can_extents_be_merged(orig_inode,
1389 ext_prev, ext_cur) &&
1390 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1391 !last_extent)
1392 continue;
1393
1394 /* Is original extent is uninitialized */
1395 uninit = ext4_ext_is_uninitialized(ext_prev);
1396
1397 data_offset_in_page = seq_start % blocks_per_page;
1398
1399 /*
1400 * Calculate data blocks count that should be swapped
1401 * at the first page.
1402 */
1403 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1404 /* Swapped blocks are across pages */
1405 block_len_in_page =
1406 blocks_per_page - data_offset_in_page;
1407 } else {
1408 /* Swapped blocks are in a page */
1409 block_len_in_page = seq_blocks;
1410 }
1411
1412 orig_page_offset = seq_start >>
1413 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1414 seq_end_page = (seq_start + seq_blocks - 1) >>
1415 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1416 seq_start = le32_to_cpu(ext_cur->ee_block);
1417 rest_blocks = seq_blocks;
1418
1419 /*
1420 * Up semaphore to avoid following problems:
1421 * a. transaction deadlock among ext4_journal_start,
1422 * ->write_begin via pagefault, and jbd2_journal_commit
1423 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1424 * in move_extent_per_page
1425 */
1426 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1427
1428 while (orig_page_offset <= seq_end_page) {
1429
1430 /* Swap original branches with new branches */
1431 block_len_in_page = move_extent_per_page(
1432 o_filp, donor_inode,
1433 orig_page_offset,
1434 data_offset_in_page,
1435 block_len_in_page, uninit,
1436 &ret);
1437
1438 /* Count how many blocks we have exchanged */
1439 *moved_len += block_len_in_page;
1440 if (ret < 0)
1441 break;
1442 if (*moved_len > len) {
1443 EXT4_ERROR_INODE(orig_inode,
1444 "We replaced blocks too much! "
1445 "sum of replaced: %llu requested: %llu",
1446 *moved_len, len);
1447 ret = -EIO;
1448 break;
1449 }
1450
1451 orig_page_offset++;
1452 data_offset_in_page = 0;
1453 rest_blocks -= block_len_in_page;
1454 if (rest_blocks > blocks_per_page)
1455 block_len_in_page = blocks_per_page;
1456 else
1457 block_len_in_page = rest_blocks;
1458 }
1459
1460 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1461 if (ret < 0)
1462 break;
1463
1464 /* Decrease buffer counter */
1465 if (holecheck_path)
1466 ext4_ext_drop_refs(holecheck_path);
1467 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1468 if (ret)
1469 break;
1470 depth = holecheck_path->p_depth;
1471
1472 /* Decrease buffer counter */
1473 if (orig_path)
1474 ext4_ext_drop_refs(orig_path);
1475 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1476 if (ret)
1477 break;
1478
1479 ext_cur = holecheck_path[depth].p_ext;
1480 add_blocks = ext4_ext_get_actual_len(ext_cur);
1481 seq_blocks = 0;
1482
1483 }
1484out:
1485 if (*moved_len) {
1486 ext4_discard_preallocations(orig_inode);
1487 ext4_discard_preallocations(donor_inode);
1488 }
1489
1490 if (orig_path) {
1491 ext4_ext_drop_refs(orig_path);
1492 kfree(orig_path);
1493 }
1494 if (holecheck_path) {
1495 ext4_ext_drop_refs(holecheck_path);
1496 kfree(holecheck_path);
1497 }
1498 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1499 ext4_inode_resume_unlocked_dio(orig_inode);
1500 ext4_inode_resume_unlocked_dio(donor_inode);
1501 unlock_two_nondirectories(orig_inode, donor_inode);
1502
1503 return ret;
1504}