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
22/**
23 * get_ext_path - Find an extent path for designated logical block number.
24 *
25 * @inode: an inode which is searched
26 * @lblock: logical block number to find an extent path
27 * @path: pointer to an extent path pointer (for output)
28 *
29 * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
30 * on failure.
31 */
32static inline int
33get_ext_path(struct inode *inode, ext4_lblk_t lblock,
34 struct ext4_ext_path **path)
35{
36 int ret = 0;
37
38 *path = ext4_ext_find_extent(inode, lblock, *path);
39 if (IS_ERR(*path)) {
40 ret = PTR_ERR(*path);
41 *path = NULL;
42 } else if ((*path)[ext_depth(inode)].p_ext == NULL)
43 ret = -ENODATA;
44
45 return ret;
46}
47
48/**
49 * copy_extent_status - Copy the extent's initialization status
50 *
51 * @src: an extent for getting initialize status
52 * @dest: an extent to be set the status
53 */
54static void
55copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
56{
57 if (ext4_ext_is_uninitialized(src))
58 ext4_ext_mark_uninitialized(dest);
59 else
60 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
61}
62
63/**
64 * mext_next_extent - Search for the next extent and set it to "extent"
65 *
66 * @inode: inode which is searched
67 * @path: this will obtain data for the next extent
68 * @extent: pointer to the next extent we have just gotten
69 *
70 * Search the next extent in the array of ext4_ext_path structure (@path)
71 * and set it to ext4_extent structure (@extent). In addition, the member of
72 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
73 * ext4_ext_path structure refers to the last extent, or a negative error
74 * value on failure.
75 */
76static int
77mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
78 struct ext4_extent **extent)
79{
80 struct ext4_extent_header *eh;
81 int ppos, leaf_ppos = path->p_depth;
82
83 ppos = leaf_ppos;
84 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
85 /* leaf block */
86 *extent = ++path[ppos].p_ext;
87 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
88 return 0;
89 }
90
91 while (--ppos >= 0) {
92 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
93 path[ppos].p_idx) {
94 int cur_ppos = ppos;
95
96 /* index block */
97 path[ppos].p_idx++;
98 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
99 if (path[ppos+1].p_bh)
100 brelse(path[ppos+1].p_bh);
101 path[ppos+1].p_bh =
102 sb_bread(inode->i_sb, path[ppos].p_block);
103 if (!path[ppos+1].p_bh)
104 return -EIO;
105 path[ppos+1].p_hdr =
106 ext_block_hdr(path[ppos+1].p_bh);
107
108 /* Halfway index block */
109 while (++cur_ppos < leaf_ppos) {
110 path[cur_ppos].p_idx =
111 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
112 path[cur_ppos].p_block =
113 ext4_idx_pblock(path[cur_ppos].p_idx);
114 if (path[cur_ppos+1].p_bh)
115 brelse(path[cur_ppos+1].p_bh);
116 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
117 path[cur_ppos].p_block);
118 if (!path[cur_ppos+1].p_bh)
119 return -EIO;
120 path[cur_ppos+1].p_hdr =
121 ext_block_hdr(path[cur_ppos+1].p_bh);
122 }
123
124 path[leaf_ppos].p_ext = *extent = NULL;
125
126 eh = path[leaf_ppos].p_hdr;
127 if (le16_to_cpu(eh->eh_entries) == 0)
128 /* empty leaf is found */
129 return -ENODATA;
130
131 /* leaf block */
132 path[leaf_ppos].p_ext = *extent =
133 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
134 path[leaf_ppos].p_block =
135 ext4_ext_pblock(path[leaf_ppos].p_ext);
136 return 0;
137 }
138 }
139 /* We found the last extent */
140 return 1;
141}
142
143/**
144 * mext_check_null_inode - NULL check for two inodes
145 *
146 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
147 */
148static int
149mext_check_null_inode(struct inode *inode1, struct inode *inode2,
150 const char *function, unsigned int line)
151{
152 int ret = 0;
153
154 if (inode1 == NULL) {
155 __ext4_error(inode2->i_sb, function, line,
156 "Both inodes should not be NULL: "
157 "inode1 NULL inode2 %lu", inode2->i_ino);
158 ret = -EIO;
159 } else if (inode2 == NULL) {
160 __ext4_error(inode1->i_sb, function, line,
161 "Both inodes should not be NULL: "
162 "inode1 %lu inode2 NULL", inode1->i_ino);
163 ret = -EIO;
164 }
165 return ret;
166}
167
168/**
169 * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
170 *
171 * @orig_inode: original inode structure
172 * @donor_inode: donor inode structure
173 * Acquire write lock of i_data_sem of the two inodes (orig and donor) by
174 * i_ino order.
175 */
176static void
177double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
178{
179 struct inode *first = orig_inode, *second = donor_inode;
180
181 /*
182 * Use the inode number to provide the stable locking order instead
183 * of its address, because the C language doesn't guarantee you can
184 * compare pointers that don't come from the same array.
185 */
186 if (donor_inode->i_ino < orig_inode->i_ino) {
187 first = donor_inode;
188 second = orig_inode;
189 }
190
191 down_write(&EXT4_I(first)->i_data_sem);
192 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
193}
194
195/**
196 * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
197 *
198 * @orig_inode: original inode structure to be released its lock first
199 * @donor_inode: donor inode structure to be released its lock second
200 * Release write lock of i_data_sem of two inodes (orig and donor).
201 */
202static void
203double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
204{
205 up_write(&EXT4_I(orig_inode)->i_data_sem);
206 up_write(&EXT4_I(donor_inode)->i_data_sem);
207}
208
209/**
210 * mext_insert_across_blocks - Insert extents across leaf block
211 *
212 * @handle: journal handle
213 * @orig_inode: original inode
214 * @o_start: first original extent to be changed
215 * @o_end: last original extent to be changed
216 * @start_ext: first new extent to be inserted
217 * @new_ext: middle of new extent to be inserted
218 * @end_ext: last new extent to be inserted
219 *
220 * Allocate a new leaf block and insert extents into it. Return 0 on success,
221 * or a negative error value on failure.
222 */
223static int
224mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
225 struct ext4_extent *o_start, struct ext4_extent *o_end,
226 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
227 struct ext4_extent *end_ext)
228{
229 struct ext4_ext_path *orig_path = NULL;
230 ext4_lblk_t eblock = 0;
231 int new_flag = 0;
232 int end_flag = 0;
233 int err = 0;
234
235 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
236 if (o_start == o_end) {
237
238 /* start_ext new_ext end_ext
239 * donor |---------|-----------|--------|
240 * orig |------------------------------|
241 */
242 end_flag = 1;
243 } else {
244
245 /* start_ext new_ext end_ext
246 * donor |---------|----------|---------|
247 * orig |---------------|--------------|
248 */
249 o_end->ee_block = end_ext->ee_block;
250 o_end->ee_len = end_ext->ee_len;
251 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
252 }
253
254 o_start->ee_len = start_ext->ee_len;
255 eblock = le32_to_cpu(start_ext->ee_block);
256 new_flag = 1;
257
258 } else if (start_ext->ee_len && new_ext->ee_len &&
259 !end_ext->ee_len && o_start == o_end) {
260
261 /* start_ext new_ext
262 * donor |--------------|---------------|
263 * orig |------------------------------|
264 */
265 o_start->ee_len = start_ext->ee_len;
266 eblock = le32_to_cpu(start_ext->ee_block);
267 new_flag = 1;
268
269 } else if (!start_ext->ee_len && new_ext->ee_len &&
270 end_ext->ee_len && o_start == o_end) {
271
272 /* new_ext end_ext
273 * donor |--------------|---------------|
274 * orig |------------------------------|
275 */
276 o_end->ee_block = end_ext->ee_block;
277 o_end->ee_len = end_ext->ee_len;
278 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
279
280 /*
281 * Set 0 to the extent block if new_ext was
282 * the first block.
283 */
284 if (new_ext->ee_block)
285 eblock = le32_to_cpu(new_ext->ee_block);
286
287 new_flag = 1;
288 } else {
289 ext4_debug("ext4 move extent: Unexpected insert case\n");
290 return -EIO;
291 }
292
293 if (new_flag) {
294 err = get_ext_path(orig_inode, eblock, &orig_path);
295 if (err)
296 goto out;
297
298 if (ext4_ext_insert_extent(handle, orig_inode,
299 orig_path, new_ext, 0))
300 goto out;
301 }
302
303 if (end_flag) {
304 err = get_ext_path(orig_inode,
305 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
306 if (err)
307 goto out;
308
309 if (ext4_ext_insert_extent(handle, orig_inode,
310 orig_path, end_ext, 0))
311 goto out;
312 }
313out:
314 if (orig_path) {
315 ext4_ext_drop_refs(orig_path);
316 kfree(orig_path);
317 }
318
319 return err;
320
321}
322
323/**
324 * mext_insert_inside_block - Insert new extent to the extent block
325 *
326 * @o_start: first original extent to be moved
327 * @o_end: last original extent to be moved
328 * @start_ext: first new extent to be inserted
329 * @new_ext: middle of new extent to be inserted
330 * @end_ext: last new extent to be inserted
331 * @eh: extent header of target leaf block
332 * @range_to_move: used to decide how to insert extent
333 *
334 * Insert extents into the leaf block. The extent (@o_start) is overwritten
335 * by inserted extents.
336 */
337static void
338mext_insert_inside_block(struct ext4_extent *o_start,
339 struct ext4_extent *o_end,
340 struct ext4_extent *start_ext,
341 struct ext4_extent *new_ext,
342 struct ext4_extent *end_ext,
343 struct ext4_extent_header *eh,
344 int range_to_move)
345{
346 int i = 0;
347 unsigned long len;
348
349 /* Move the existing extents */
350 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
351 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
352 (unsigned long)(o_end + 1);
353 memmove(o_end + 1 + range_to_move, o_end + 1, len);
354 }
355
356 /* Insert start entry */
357 if (start_ext->ee_len)
358 o_start[i++].ee_len = start_ext->ee_len;
359
360 /* Insert new entry */
361 if (new_ext->ee_len) {
362 o_start[i] = *new_ext;
363 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
364 }
365
366 /* Insert end entry */
367 if (end_ext->ee_len)
368 o_start[i] = *end_ext;
369
370 /* Increment the total entries counter on the extent block */
371 le16_add_cpu(&eh->eh_entries, range_to_move);
372}
373
374/**
375 * mext_insert_extents - Insert new extent
376 *
377 * @handle: journal handle
378 * @orig_inode: original inode
379 * @orig_path: path indicates first extent to be changed
380 * @o_start: first original extent to be changed
381 * @o_end: last original extent to be changed
382 * @start_ext: first new extent to be inserted
383 * @new_ext: middle of new extent to be inserted
384 * @end_ext: last new extent to be inserted
385 *
386 * Call the function to insert extents. If we cannot add more extents into
387 * the leaf block, we call mext_insert_across_blocks() to create a
388 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
389 * on success, or a negative error value on failure.
390 */
391static int
392mext_insert_extents(handle_t *handle, struct inode *orig_inode,
393 struct ext4_ext_path *orig_path,
394 struct ext4_extent *o_start,
395 struct ext4_extent *o_end,
396 struct ext4_extent *start_ext,
397 struct ext4_extent *new_ext,
398 struct ext4_extent *end_ext)
399{
400 struct ext4_extent_header *eh;
401 unsigned long need_slots, slots_range;
402 int range_to_move, depth, ret;
403
404 /*
405 * The extents need to be inserted
406 * start_extent + new_extent + end_extent.
407 */
408 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
409 (new_ext->ee_len ? 1 : 0);
410
411 /* The number of slots between start and end */
412 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
413 / sizeof(struct ext4_extent);
414
415 /* Range to move the end of extent */
416 range_to_move = need_slots - slots_range;
417 depth = orig_path->p_depth;
418 orig_path += depth;
419 eh = orig_path->p_hdr;
420
421 if (depth) {
422 /* Register to journal */
423 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
424 if (ret)
425 return ret;
426 }
427
428 /* Expansion */
429 if (range_to_move > 0 &&
430 (range_to_move > le16_to_cpu(eh->eh_max)
431 - le16_to_cpu(eh->eh_entries))) {
432
433 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
434 o_end, start_ext, new_ext, end_ext);
435 if (ret < 0)
436 return ret;
437 } else
438 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
439 end_ext, eh, range_to_move);
440
441 if (depth) {
442 ret = ext4_handle_dirty_metadata(handle, orig_inode,
443 orig_path->p_bh);
444 if (ret)
445 return ret;
446 } else {
447 ret = ext4_mark_inode_dirty(handle, orig_inode);
448 if (ret < 0)
449 return ret;
450 }
451
452 return 0;
453}
454
455/**
456 * mext_leaf_block - Move one leaf extent block into the inode.
457 *
458 * @handle: journal handle
459 * @orig_inode: original inode
460 * @orig_path: path indicates first extent to be changed
461 * @dext: donor extent
462 * @from: start offset on the target file
463 *
464 * In order to insert extents into the leaf block, we must divide the extent
465 * in the leaf block into three extents. The one is located to be inserted
466 * extents, and the others are located around it.
467 *
468 * Therefore, this function creates structures to save extents of the leaf
469 * block, and inserts extents by calling mext_insert_extents() with
470 * created extents. Return 0 on success, or a negative error value on failure.
471 */
472static int
473mext_leaf_block(handle_t *handle, struct inode *orig_inode,
474 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
475 ext4_lblk_t *from)
476{
477 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
478 struct ext4_extent new_ext, start_ext, end_ext;
479 ext4_lblk_t new_ext_end;
480 int oext_alen, new_ext_alen, end_ext_alen;
481 int depth = ext_depth(orig_inode);
482 int ret;
483
484 start_ext.ee_block = end_ext.ee_block = 0;
485 o_start = o_end = oext = orig_path[depth].p_ext;
486 oext_alen = ext4_ext_get_actual_len(oext);
487 start_ext.ee_len = end_ext.ee_len = 0;
488
489 new_ext.ee_block = cpu_to_le32(*from);
490 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
491 new_ext.ee_len = dext->ee_len;
492 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
493 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
494
495 /*
496 * Case: original extent is first
497 * oext |--------|
498 * new_ext |--|
499 * start_ext |--|
500 */
501 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
502 le32_to_cpu(new_ext.ee_block) <
503 le32_to_cpu(oext->ee_block) + oext_alen) {
504 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
505 le32_to_cpu(oext->ee_block));
506 start_ext.ee_block = oext->ee_block;
507 copy_extent_status(oext, &start_ext);
508 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
509 prev_ext = oext - 1;
510 /*
511 * We can merge new_ext into previous extent,
512 * if these are contiguous and same extent type.
513 */
514 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
515 &new_ext)) {
516 o_start = prev_ext;
517 start_ext.ee_len = cpu_to_le16(
518 ext4_ext_get_actual_len(prev_ext) +
519 new_ext_alen);
520 start_ext.ee_block = oext->ee_block;
521 copy_extent_status(prev_ext, &start_ext);
522 new_ext.ee_len = 0;
523 }
524 }
525
526 /*
527 * Case: new_ext_end must be less than oext
528 * oext |-----------|
529 * new_ext |-------|
530 */
531 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
532 EXT4_ERROR_INODE(orig_inode,
533 "new_ext_end(%u) should be less than or equal to "
534 "oext->ee_block(%u) + oext_alen(%d) - 1",
535 new_ext_end, le32_to_cpu(oext->ee_block),
536 oext_alen);
537 ret = -EIO;
538 goto out;
539 }
540
541 /*
542 * Case: new_ext is smaller than original extent
543 * oext |---------------|
544 * new_ext |-----------|
545 * end_ext |---|
546 */
547 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
548 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
549 end_ext.ee_len =
550 cpu_to_le16(le32_to_cpu(oext->ee_block) +
551 oext_alen - 1 - new_ext_end);
552 copy_extent_status(oext, &end_ext);
553 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
554 ext4_ext_store_pblock(&end_ext,
555 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
556 end_ext.ee_block =
557 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
558 oext_alen - end_ext_alen);
559 }
560
561 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
562 o_end, &start_ext, &new_ext, &end_ext);
563out:
564 return ret;
565}
566
567/**
568 * mext_calc_swap_extents - Calculate extents for extent swapping.
569 *
570 * @tmp_dext: the extent that will belong to the original inode
571 * @tmp_oext: the extent that will belong to the donor inode
572 * @orig_off: block offset of original inode
573 * @donor_off: block offset of donor inode
574 * @max_count: the maximum length of extents
575 *
576 * Return 0 on success, or a negative error value on failure.
577 */
578static int
579mext_calc_swap_extents(struct ext4_extent *tmp_dext,
580 struct ext4_extent *tmp_oext,
581 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
582 ext4_lblk_t max_count)
583{
584 ext4_lblk_t diff, orig_diff;
585 struct ext4_extent dext_old, oext_old;
586
587 BUG_ON(orig_off != donor_off);
588
589 /* original and donor extents have to cover the same block offset */
590 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
591 le32_to_cpu(tmp_oext->ee_block) +
592 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
593 return -ENODATA;
594
595 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
596 le32_to_cpu(tmp_dext->ee_block) +
597 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
598 return -ENODATA;
599
600 dext_old = *tmp_dext;
601 oext_old = *tmp_oext;
602
603 /* When tmp_dext is too large, pick up the target range. */
604 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
605
606 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
607 tmp_dext->ee_block =
608 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
609 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
610
611 if (max_count < ext4_ext_get_actual_len(tmp_dext))
612 tmp_dext->ee_len = cpu_to_le16(max_count);
613
614 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
615 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
616
617 /* Adjust extent length if donor extent is larger than orig */
618 if (ext4_ext_get_actual_len(tmp_dext) >
619 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
620 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
621 orig_diff);
622
623 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
624
625 copy_extent_status(&oext_old, tmp_dext);
626 copy_extent_status(&dext_old, tmp_oext);
627
628 return 0;
629}
630
631/**
632 * mext_replace_branches - Replace original extents with new extents
633 *
634 * @handle: journal handle
635 * @orig_inode: original inode
636 * @donor_inode: donor inode
637 * @from: block offset of orig_inode
638 * @count: block count to be replaced
639 * @err: pointer to save return value
640 *
641 * Replace original inode extents and donor inode extents page by page.
642 * We implement this replacement in the following three steps:
643 * 1. Save the block information of original and donor inodes into
644 * dummy extents.
645 * 2. Change the block information of original inode to point at the
646 * donor inode blocks.
647 * 3. Change the block information of donor inode to point at the saved
648 * original inode blocks in the dummy extents.
649 *
650 * Return replaced block count.
651 */
652static int
653mext_replace_branches(handle_t *handle, struct inode *orig_inode,
654 struct inode *donor_inode, ext4_lblk_t from,
655 ext4_lblk_t count, int *err)
656{
657 struct ext4_ext_path *orig_path = NULL;
658 struct ext4_ext_path *donor_path = NULL;
659 struct ext4_extent *oext, *dext;
660 struct ext4_extent tmp_dext, tmp_oext;
661 ext4_lblk_t orig_off = from, donor_off = from;
662 int depth;
663 int replaced_count = 0;
664 int dext_alen;
665
666 /* Protect extent trees against block allocations via delalloc */
667 double_down_write_data_sem(orig_inode, donor_inode);
668
669 /* Get the original extent for the block "orig_off" */
670 *err = get_ext_path(orig_inode, orig_off, &orig_path);
671 if (*err)
672 goto out;
673
674 /* Get the donor extent for the head */
675 *err = get_ext_path(donor_inode, donor_off, &donor_path);
676 if (*err)
677 goto out;
678 depth = ext_depth(orig_inode);
679 oext = orig_path[depth].p_ext;
680 tmp_oext = *oext;
681
682 depth = ext_depth(donor_inode);
683 dext = donor_path[depth].p_ext;
684 tmp_dext = *dext;
685
686 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
687 donor_off, count);
688 if (*err)
689 goto out;
690
691 /* Loop for the donor extents */
692 while (1) {
693 /* The extent for donor must be found. */
694 if (!dext) {
695 EXT4_ERROR_INODE(donor_inode,
696 "The extent for donor must be found");
697 *err = -EIO;
698 goto out;
699 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
700 EXT4_ERROR_INODE(donor_inode,
701 "Donor offset(%u) and the first block of donor "
702 "extent(%u) should be equal",
703 donor_off,
704 le32_to_cpu(tmp_dext.ee_block));
705 *err = -EIO;
706 goto out;
707 }
708
709 /* Set donor extent to orig extent */
710 *err = mext_leaf_block(handle, orig_inode,
711 orig_path, &tmp_dext, &orig_off);
712 if (*err)
713 goto out;
714
715 /* Set orig extent to donor extent */
716 *err = mext_leaf_block(handle, donor_inode,
717 donor_path, &tmp_oext, &donor_off);
718 if (*err)
719 goto out;
720
721 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
722 replaced_count += dext_alen;
723 donor_off += dext_alen;
724 orig_off += dext_alen;
725
726 /* Already moved the expected blocks */
727 if (replaced_count >= count)
728 break;
729
730 if (orig_path)
731 ext4_ext_drop_refs(orig_path);
732 *err = get_ext_path(orig_inode, orig_off, &orig_path);
733 if (*err)
734 goto out;
735 depth = ext_depth(orig_inode);
736 oext = orig_path[depth].p_ext;
737 tmp_oext = *oext;
738
739 if (donor_path)
740 ext4_ext_drop_refs(donor_path);
741 *err = get_ext_path(donor_inode, donor_off, &donor_path);
742 if (*err)
743 goto out;
744 depth = ext_depth(donor_inode);
745 dext = donor_path[depth].p_ext;
746 tmp_dext = *dext;
747
748 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
749 donor_off, count - replaced_count);
750 if (*err)
751 goto out;
752 }
753
754out:
755 if (orig_path) {
756 ext4_ext_drop_refs(orig_path);
757 kfree(orig_path);
758 }
759 if (donor_path) {
760 ext4_ext_drop_refs(donor_path);
761 kfree(donor_path);
762 }
763
764 ext4_ext_invalidate_cache(orig_inode);
765 ext4_ext_invalidate_cache(donor_inode);
766
767 double_up_write_data_sem(orig_inode, donor_inode);
768
769 return replaced_count;
770}
771
772/**
773 * move_extent_per_page - Move extent data per page
774 *
775 * @o_filp: file structure of original file
776 * @donor_inode: donor inode
777 * @orig_page_offset: page index on original file
778 * @data_offset_in_page: block index where data swapping starts
779 * @block_len_in_page: the number of blocks to be swapped
780 * @uninit: orig extent is uninitialized or not
781 * @err: pointer to save return value
782 *
783 * Save the data in original inode blocks and replace original inode extents
784 * with donor inode extents by calling mext_replace_branches().
785 * Finally, write out the saved data in new original inode blocks. Return
786 * replaced block count.
787 */
788static int
789move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
790 pgoff_t orig_page_offset, int data_offset_in_page,
791 int block_len_in_page, int uninit, int *err)
792{
793 struct inode *orig_inode = o_filp->f_dentry->d_inode;
794 struct address_space *mapping = orig_inode->i_mapping;
795 struct buffer_head *bh;
796 struct page *page = NULL;
797 const struct address_space_operations *a_ops = mapping->a_ops;
798 handle_t *handle;
799 ext4_lblk_t orig_blk_offset;
800 long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
801 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
802 unsigned int w_flags = 0;
803 unsigned int tmp_data_size, data_size, replaced_size;
804 void *fsdata;
805 int i, jblocks;
806 int err2 = 0;
807 int replaced_count = 0;
808 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
809
810 /*
811 * It needs twice the amount of ordinary journal buffers because
812 * inode and donor_inode may change each different metadata blocks.
813 */
814 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
815 handle = ext4_journal_start(orig_inode, jblocks);
816 if (IS_ERR(handle)) {
817 *err = PTR_ERR(handle);
818 return 0;
819 }
820
821 if (segment_eq(get_fs(), KERNEL_DS))
822 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
823
824 orig_blk_offset = orig_page_offset * blocks_per_page +
825 data_offset_in_page;
826
827 /*
828 * If orig extent is uninitialized one,
829 * it's not necessary force the page into memory
830 * and then force it to be written out again.
831 * Just swap data blocks between orig and donor.
832 */
833 if (uninit) {
834 replaced_count = mext_replace_branches(handle, orig_inode,
835 donor_inode, orig_blk_offset,
836 block_len_in_page, err);
837 goto out2;
838 }
839
840 offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
841
842 /* Calculate data_size */
843 if ((orig_blk_offset + block_len_in_page - 1) ==
844 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
845 /* Replace the last block */
846 tmp_data_size = orig_inode->i_size & (blocksize - 1);
847 /*
848 * If data_size equal zero, it shows data_size is multiples of
849 * blocksize. So we set appropriate value.
850 */
851 if (tmp_data_size == 0)
852 tmp_data_size = blocksize;
853
854 data_size = tmp_data_size +
855 ((block_len_in_page - 1) << orig_inode->i_blkbits);
856 } else
857 data_size = block_len_in_page << orig_inode->i_blkbits;
858
859 replaced_size = data_size;
860
861 *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags,
862 &page, &fsdata);
863 if (unlikely(*err < 0))
864 goto out;
865
866 if (!PageUptodate(page)) {
867 mapping->a_ops->readpage(o_filp, page);
868 lock_page(page);
869 }
870
871 /*
872 * try_to_release_page() doesn't call releasepage in writeback mode.
873 * We should care about the order of writing to the same file
874 * by multiple move extent processes.
875 * It needs to call wait_on_page_writeback() to wait for the
876 * writeback of the page.
877 */
878 wait_on_page_writeback(page);
879
880 /* Release old bh and drop refs */
881 try_to_release_page(page, 0);
882
883 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
884 orig_blk_offset, block_len_in_page,
885 &err2);
886 if (err2) {
887 if (replaced_count) {
888 block_len_in_page = replaced_count;
889 replaced_size =
890 block_len_in_page << orig_inode->i_blkbits;
891 } else
892 goto out;
893 }
894
895 if (!page_has_buffers(page))
896 create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0);
897
898 bh = page_buffers(page);
899 for (i = 0; i < data_offset_in_page; i++)
900 bh = bh->b_this_page;
901
902 for (i = 0; i < block_len_in_page; i++) {
903 *err = ext4_get_block(orig_inode,
904 (sector_t)(orig_blk_offset + i), bh, 0);
905 if (*err < 0)
906 goto out;
907
908 if (bh->b_this_page != NULL)
909 bh = bh->b_this_page;
910 }
911
912 *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size,
913 page, fsdata);
914 page = NULL;
915
916out:
917 if (unlikely(page)) {
918 if (PageLocked(page))
919 unlock_page(page);
920 page_cache_release(page);
921 ext4_journal_stop(handle);
922 }
923out2:
924 ext4_journal_stop(handle);
925
926 if (err2)
927 *err = err2;
928
929 return replaced_count;
930}
931
932/**
933 * mext_check_arguments - Check whether move extent can be done
934 *
935 * @orig_inode: original inode
936 * @donor_inode: donor inode
937 * @orig_start: logical start offset in block for orig
938 * @donor_start: logical start offset in block for donor
939 * @len: the number of blocks to be moved
940 *
941 * Check the arguments of ext4_move_extents() whether the files can be
942 * exchanged with each other.
943 * Return 0 on success, or a negative error value on failure.
944 */
945static int
946mext_check_arguments(struct inode *orig_inode,
947 struct inode *donor_inode, __u64 orig_start,
948 __u64 donor_start, __u64 *len)
949{
950 ext4_lblk_t orig_blocks, donor_blocks;
951 unsigned int blkbits = orig_inode->i_blkbits;
952 unsigned int blocksize = 1 << blkbits;
953
954 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
955 ext4_debug("ext4 move extent: suid or sgid is set"
956 " to donor file [ino:orig %lu, donor %lu]\n",
957 orig_inode->i_ino, donor_inode->i_ino);
958 return -EINVAL;
959 }
960
961 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
962 return -EPERM;
963
964 /* Ext4 move extent does not support swapfile */
965 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
966 ext4_debug("ext4 move extent: The argument files should "
967 "not be swapfile [ino:orig %lu, donor %lu]\n",
968 orig_inode->i_ino, donor_inode->i_ino);
969 return -EINVAL;
970 }
971
972 /* Files should be in the same ext4 FS */
973 if (orig_inode->i_sb != donor_inode->i_sb) {
974 ext4_debug("ext4 move extent: The argument files "
975 "should be in same FS [ino:orig %lu, donor %lu]\n",
976 orig_inode->i_ino, donor_inode->i_ino);
977 return -EINVAL;
978 }
979
980 /* Ext4 move extent supports only extent based file */
981 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
982 ext4_debug("ext4 move extent: orig file is not extents "
983 "based file [ino:orig %lu]\n", orig_inode->i_ino);
984 return -EOPNOTSUPP;
985 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
986 ext4_debug("ext4 move extent: donor file is not extents "
987 "based file [ino:donor %lu]\n", donor_inode->i_ino);
988 return -EOPNOTSUPP;
989 }
990
991 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
992 ext4_debug("ext4 move extent: File size is 0 byte\n");
993 return -EINVAL;
994 }
995
996 /* Start offset should be same */
997 if (orig_start != donor_start) {
998 ext4_debug("ext4 move extent: orig and donor's start "
999 "offset are not same [ino:orig %lu, donor %lu]\n",
1000 orig_inode->i_ino, donor_inode->i_ino);
1001 return -EINVAL;
1002 }
1003
1004 if ((orig_start >= EXT_MAX_BLOCKS) ||
1005 (donor_start >= EXT_MAX_BLOCKS) ||
1006 (*len > EXT_MAX_BLOCKS) ||
1007 (orig_start + *len >= EXT_MAX_BLOCKS)) {
1008 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1009 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1010 orig_inode->i_ino, donor_inode->i_ino);
1011 return -EINVAL;
1012 }
1013
1014 if (orig_inode->i_size > donor_inode->i_size) {
1015 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1016 /* TODO: eliminate this artificial restriction */
1017 if (orig_start >= donor_blocks) {
1018 ext4_debug("ext4 move extent: orig start offset "
1019 "[%llu] should be less than donor file blocks "
1020 "[%u] [ino:orig %lu, donor %lu]\n",
1021 orig_start, donor_blocks,
1022 orig_inode->i_ino, donor_inode->i_ino);
1023 return -EINVAL;
1024 }
1025
1026 /* TODO: eliminate this artificial restriction */
1027 if (orig_start + *len > donor_blocks) {
1028 ext4_debug("ext4 move extent: End offset [%llu] should "
1029 "be less than donor file blocks [%u]."
1030 "So adjust length from %llu to %llu "
1031 "[ino:orig %lu, donor %lu]\n",
1032 orig_start + *len, donor_blocks,
1033 *len, donor_blocks - orig_start,
1034 orig_inode->i_ino, donor_inode->i_ino);
1035 *len = donor_blocks - orig_start;
1036 }
1037 } else {
1038 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1039 if (orig_start >= orig_blocks) {
1040 ext4_debug("ext4 move extent: start offset [%llu] "
1041 "should be less than original file blocks "
1042 "[%u] [ino:orig %lu, donor %lu]\n",
1043 orig_start, orig_blocks,
1044 orig_inode->i_ino, donor_inode->i_ino);
1045 return -EINVAL;
1046 }
1047
1048 if (orig_start + *len > orig_blocks) {
1049 ext4_debug("ext4 move extent: Adjust length "
1050 "from %llu to %llu. Because it should be "
1051 "less than original file blocks "
1052 "[ino:orig %lu, donor %lu]\n",
1053 *len, orig_blocks - orig_start,
1054 orig_inode->i_ino, donor_inode->i_ino);
1055 *len = orig_blocks - orig_start;
1056 }
1057 }
1058
1059 if (!*len) {
1060 ext4_debug("ext4 move extent: len should not be 0 "
1061 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1062 donor_inode->i_ino);
1063 return -EINVAL;
1064 }
1065
1066 return 0;
1067}
1068
1069/**
1070 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1071 *
1072 * @inode1: the inode structure
1073 * @inode2: the inode structure
1074 *
1075 * Lock two inodes' i_mutex by i_ino order.
1076 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1077 */
1078static int
1079mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1080{
1081 int ret = 0;
1082
1083 BUG_ON(inode1 == NULL && inode2 == NULL);
1084
1085 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1086 if (ret < 0)
1087 goto out;
1088
1089 if (inode1 == inode2) {
1090 mutex_lock(&inode1->i_mutex);
1091 goto out;
1092 }
1093
1094 if (inode1->i_ino < inode2->i_ino) {
1095 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1096 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1097 } else {
1098 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1099 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1100 }
1101
1102out:
1103 return ret;
1104}
1105
1106/**
1107 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1108 *
1109 * @inode1: the inode that is released first
1110 * @inode2: the inode that is released second
1111 *
1112 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1113 */
1114
1115static int
1116mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1117{
1118 int ret = 0;
1119
1120 BUG_ON(inode1 == NULL && inode2 == NULL);
1121
1122 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1123 if (ret < 0)
1124 goto out;
1125
1126 if (inode1)
1127 mutex_unlock(&inode1->i_mutex);
1128
1129 if (inode2 && inode2 != inode1)
1130 mutex_unlock(&inode2->i_mutex);
1131
1132out:
1133 return ret;
1134}
1135
1136/**
1137 * ext4_move_extents - Exchange the specified range of a file
1138 *
1139 * @o_filp: file structure of the original file
1140 * @d_filp: file structure of the donor file
1141 * @orig_start: start offset in block for orig
1142 * @donor_start: start offset in block for donor
1143 * @len: the number of blocks to be moved
1144 * @moved_len: moved block length
1145 *
1146 * This function returns 0 and moved block length is set in moved_len
1147 * if succeed, otherwise returns error value.
1148 *
1149 * Note: ext4_move_extents() proceeds the following order.
1150 * 1:ext4_move_extents() calculates the last block number of moving extent
1151 * function by the start block number (orig_start) and the number of blocks
1152 * to be moved (len) specified as arguments.
1153 * If the {orig, donor}_start points a hole, the extent's start offset
1154 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1155 * after hole behind.
1156 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1157 * or the ext_cur exceeds the block_end which is last logical block number.
1158 * 3:To get the length of continues area, call mext_next_extent()
1159 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1160 * until find un-continuous extent, the start logical block number exceeds
1161 * the block_end or the extent points to the last extent.
1162 * 4:Exchange the original inode data with donor inode data
1163 * from orig_page_offset to seq_end_page.
1164 * The start indexes of data are specified as arguments.
1165 * That of the original inode is orig_page_offset,
1166 * and the donor inode is also orig_page_offset
1167 * (To easily handle blocksize != pagesize case, the offset for the
1168 * donor inode is block unit).
1169 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1170 * then returns to step 2.
1171 * 6:Release holecheck_path, orig_path and set the len to moved_len
1172 * which shows the number of moved blocks.
1173 * The moved_len is useful for the command to calculate the file offset
1174 * for starting next move extent ioctl.
1175 * 7:Return 0 on success, or a negative error value on failure.
1176 */
1177int
1178ext4_move_extents(struct file *o_filp, struct file *d_filp,
1179 __u64 orig_start, __u64 donor_start, __u64 len,
1180 __u64 *moved_len)
1181{
1182 struct inode *orig_inode = o_filp->f_dentry->d_inode;
1183 struct inode *donor_inode = d_filp->f_dentry->d_inode;
1184 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1185 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1186 ext4_lblk_t block_start = orig_start;
1187 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1188 ext4_lblk_t rest_blocks;
1189 pgoff_t orig_page_offset = 0, seq_end_page;
1190 int ret1, ret2, depth, last_extent = 0;
1191 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1192 int data_offset_in_page;
1193 int block_len_in_page;
1194 int uninit;
1195
1196 /* orig and donor should be different file */
1197 if (orig_inode->i_ino == donor_inode->i_ino) {
1198 ext4_debug("ext4 move extent: The argument files should not "
1199 "be same file [ino:orig %lu, donor %lu]\n",
1200 orig_inode->i_ino, donor_inode->i_ino);
1201 return -EINVAL;
1202 }
1203
1204 /* Regular file check */
1205 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1206 ext4_debug("ext4 move extent: The argument files should be "
1207 "regular file [ino:orig %lu, donor %lu]\n",
1208 orig_inode->i_ino, donor_inode->i_ino);
1209 return -EINVAL;
1210 }
1211
1212 /* Protect orig and donor inodes against a truncate */
1213 ret1 = mext_inode_double_lock(orig_inode, donor_inode);
1214 if (ret1 < 0)
1215 return ret1;
1216
1217 /* Protect extent tree against block allocations via delalloc */
1218 double_down_write_data_sem(orig_inode, donor_inode);
1219 /* Check the filesystem environment whether move_extent can be done */
1220 ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start,
1221 donor_start, &len);
1222 if (ret1)
1223 goto out;
1224
1225 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1226 block_end = block_start + len - 1;
1227 if (file_end < block_end)
1228 len -= block_end - file_end;
1229
1230 ret1 = get_ext_path(orig_inode, block_start, &orig_path);
1231 if (ret1)
1232 goto out;
1233
1234 /* Get path structure to check the hole */
1235 ret1 = get_ext_path(orig_inode, block_start, &holecheck_path);
1236 if (ret1)
1237 goto out;
1238
1239 depth = ext_depth(orig_inode);
1240 ext_cur = holecheck_path[depth].p_ext;
1241
1242 /*
1243 * Get proper starting location of block replacement if block_start was
1244 * within the hole.
1245 */
1246 if (le32_to_cpu(ext_cur->ee_block) +
1247 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1248 /*
1249 * The hole exists between extents or the tail of
1250 * original file.
1251 */
1252 last_extent = mext_next_extent(orig_inode,
1253 holecheck_path, &ext_cur);
1254 if (last_extent < 0) {
1255 ret1 = last_extent;
1256 goto out;
1257 }
1258 last_extent = mext_next_extent(orig_inode, orig_path,
1259 &ext_dummy);
1260 if (last_extent < 0) {
1261 ret1 = last_extent;
1262 goto out;
1263 }
1264 seq_start = le32_to_cpu(ext_cur->ee_block);
1265 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1266 /* The hole exists at the beginning of original file. */
1267 seq_start = le32_to_cpu(ext_cur->ee_block);
1268 else
1269 seq_start = block_start;
1270
1271 /* No blocks within the specified range. */
1272 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1273 ext4_debug("ext4 move extent: The specified range of file "
1274 "may be the hole\n");
1275 ret1 = -EINVAL;
1276 goto out;
1277 }
1278
1279 /* Adjust start blocks */
1280 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1281 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1282 max(le32_to_cpu(ext_cur->ee_block), block_start);
1283
1284 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1285 seq_blocks += add_blocks;
1286
1287 /* Adjust tail blocks */
1288 if (seq_start + seq_blocks - 1 > block_end)
1289 seq_blocks = block_end - seq_start + 1;
1290
1291 ext_prev = ext_cur;
1292 last_extent = mext_next_extent(orig_inode, holecheck_path,
1293 &ext_cur);
1294 if (last_extent < 0) {
1295 ret1 = last_extent;
1296 break;
1297 }
1298 add_blocks = ext4_ext_get_actual_len(ext_cur);
1299
1300 /*
1301 * Extend the length of contiguous block (seq_blocks)
1302 * if extents are contiguous.
1303 */
1304 if (ext4_can_extents_be_merged(orig_inode,
1305 ext_prev, ext_cur) &&
1306 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1307 !last_extent)
1308 continue;
1309
1310 /* Is original extent is uninitialized */
1311 uninit = ext4_ext_is_uninitialized(ext_prev);
1312
1313 data_offset_in_page = seq_start % blocks_per_page;
1314
1315 /*
1316 * Calculate data blocks count that should be swapped
1317 * at the first page.
1318 */
1319 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1320 /* Swapped blocks are across pages */
1321 block_len_in_page =
1322 blocks_per_page - data_offset_in_page;
1323 } else {
1324 /* Swapped blocks are in a page */
1325 block_len_in_page = seq_blocks;
1326 }
1327
1328 orig_page_offset = seq_start >>
1329 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1330 seq_end_page = (seq_start + seq_blocks - 1) >>
1331 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1332 seq_start = le32_to_cpu(ext_cur->ee_block);
1333 rest_blocks = seq_blocks;
1334
1335 /*
1336 * Up semaphore to avoid following problems:
1337 * a. transaction deadlock among ext4_journal_start,
1338 * ->write_begin via pagefault, and jbd2_journal_commit
1339 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1340 * in move_extent_per_page
1341 */
1342 double_up_write_data_sem(orig_inode, donor_inode);
1343
1344 while (orig_page_offset <= seq_end_page) {
1345
1346 /* Swap original branches with new branches */
1347 block_len_in_page = move_extent_per_page(
1348 o_filp, donor_inode,
1349 orig_page_offset,
1350 data_offset_in_page,
1351 block_len_in_page, uninit,
1352 &ret1);
1353
1354 /* Count how many blocks we have exchanged */
1355 *moved_len += block_len_in_page;
1356 if (ret1 < 0)
1357 break;
1358 if (*moved_len > len) {
1359 EXT4_ERROR_INODE(orig_inode,
1360 "We replaced blocks too much! "
1361 "sum of replaced: %llu requested: %llu",
1362 *moved_len, len);
1363 ret1 = -EIO;
1364 break;
1365 }
1366
1367 orig_page_offset++;
1368 data_offset_in_page = 0;
1369 rest_blocks -= block_len_in_page;
1370 if (rest_blocks > blocks_per_page)
1371 block_len_in_page = blocks_per_page;
1372 else
1373 block_len_in_page = rest_blocks;
1374 }
1375
1376 double_down_write_data_sem(orig_inode, donor_inode);
1377 if (ret1 < 0)
1378 break;
1379
1380 /* Decrease buffer counter */
1381 if (holecheck_path)
1382 ext4_ext_drop_refs(holecheck_path);
1383 ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path);
1384 if (ret1)
1385 break;
1386 depth = holecheck_path->p_depth;
1387
1388 /* Decrease buffer counter */
1389 if (orig_path)
1390 ext4_ext_drop_refs(orig_path);
1391 ret1 = get_ext_path(orig_inode, seq_start, &orig_path);
1392 if (ret1)
1393 break;
1394
1395 ext_cur = holecheck_path[depth].p_ext;
1396 add_blocks = ext4_ext_get_actual_len(ext_cur);
1397 seq_blocks = 0;
1398
1399 }
1400out:
1401 if (*moved_len) {
1402 ext4_discard_preallocations(orig_inode);
1403 ext4_discard_preallocations(donor_inode);
1404 }
1405
1406 if (orig_path) {
1407 ext4_ext_drop_refs(orig_path);
1408 kfree(orig_path);
1409 }
1410 if (holecheck_path) {
1411 ext4_ext_drop_refs(holecheck_path);
1412 kfree(holecheck_path);
1413 }
1414 double_up_write_data_sem(orig_inode, donor_inode);
1415 ret2 = mext_inode_double_unlock(orig_inode, donor_inode);
1416
1417 if (ret1)
1418 return ret1;
1419 else if (ret2)
1420 return ret2;
1421
1422 return 0;
1423}
1// SPDX-License-Identifier: LGPL-2.1
2/*
3 * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
4 * Written by Takashi Sato <t-sato@yk.jp.nec.com>
5 * Akira Fujita <a-fujita@rs.jp.nec.com>
6 */
7
8#include <linux/fs.h>
9#include <linux/quotaops.h>
10#include <linux/slab.h>
11#include <linux/sched/mm.h>
12#include "ext4_jbd2.h"
13#include "ext4.h"
14#include "ext4_extents.h"
15
16/**
17 * get_ext_path() - Find an extent path for designated logical block number.
18 * @inode: inode to be searched
19 * @lblock: logical block number to find an extent path
20 * @ppath: pointer to an extent path pointer (for output)
21 *
22 * ext4_find_extent wrapper. Return 0 on success, or a negative error value
23 * on failure.
24 */
25static inline int
26get_ext_path(struct inode *inode, ext4_lblk_t lblock,
27 struct ext4_ext_path **ppath)
28{
29 struct ext4_ext_path *path;
30
31 path = ext4_find_extent(inode, lblock, ppath, EXT4_EX_NOCACHE);
32 if (IS_ERR(path))
33 return PTR_ERR(path);
34 if (path[ext_depth(inode)].p_ext == NULL) {
35 ext4_free_ext_path(path);
36 *ppath = NULL;
37 return -ENODATA;
38 }
39 *ppath = path;
40 return 0;
41}
42
43/**
44 * ext4_double_down_write_data_sem() - write lock two inodes's i_data_sem
45 * @first: inode to be locked
46 * @second: inode to be locked
47 *
48 * Acquire write lock of i_data_sem of the two inodes
49 */
50void
51ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
52{
53 if (first < second) {
54 down_write(&EXT4_I(first)->i_data_sem);
55 down_write_nested(&EXT4_I(second)->i_data_sem, I_DATA_SEM_OTHER);
56 } else {
57 down_write(&EXT4_I(second)->i_data_sem);
58 down_write_nested(&EXT4_I(first)->i_data_sem, I_DATA_SEM_OTHER);
59
60 }
61}
62
63/**
64 * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
65 *
66 * @orig_inode: original inode structure to be released its lock first
67 * @donor_inode: donor inode structure to be released its lock second
68 * Release write lock of i_data_sem of two inodes (orig and donor).
69 */
70void
71ext4_double_up_write_data_sem(struct inode *orig_inode,
72 struct inode *donor_inode)
73{
74 up_write(&EXT4_I(orig_inode)->i_data_sem);
75 up_write(&EXT4_I(donor_inode)->i_data_sem);
76}
77
78/**
79 * mext_check_coverage - Check that all extents in range has the same type
80 *
81 * @inode: inode in question
82 * @from: block offset of inode
83 * @count: block count to be checked
84 * @unwritten: extents expected to be unwritten
85 * @err: pointer to save error value
86 *
87 * Return 1 if all extents in range has expected type, and zero otherwise.
88 */
89static int
90mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
91 int unwritten, int *err)
92{
93 struct ext4_ext_path *path = NULL;
94 struct ext4_extent *ext;
95 int ret = 0;
96 ext4_lblk_t last = from + count;
97 while (from < last) {
98 *err = get_ext_path(inode, from, &path);
99 if (*err)
100 goto out;
101 ext = path[ext_depth(inode)].p_ext;
102 if (unwritten != ext4_ext_is_unwritten(ext))
103 goto out;
104 from += ext4_ext_get_actual_len(ext);
105 }
106 ret = 1;
107out:
108 ext4_free_ext_path(path);
109 return ret;
110}
111
112/**
113 * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
114 *
115 * @inode1: the inode structure
116 * @inode2: the inode structure
117 * @index1: page index
118 * @index2: page index
119 * @page: result page vector
120 *
121 * Grab two locked pages for inode's by inode order
122 */
123static int
124mext_page_double_lock(struct inode *inode1, struct inode *inode2,
125 pgoff_t index1, pgoff_t index2, struct page *page[2])
126{
127 struct address_space *mapping[2];
128 unsigned int flags;
129
130 BUG_ON(!inode1 || !inode2);
131 if (inode1 < inode2) {
132 mapping[0] = inode1->i_mapping;
133 mapping[1] = inode2->i_mapping;
134 } else {
135 swap(index1, index2);
136 mapping[0] = inode2->i_mapping;
137 mapping[1] = inode1->i_mapping;
138 }
139
140 flags = memalloc_nofs_save();
141 page[0] = grab_cache_page_write_begin(mapping[0], index1);
142 if (!page[0]) {
143 memalloc_nofs_restore(flags);
144 return -ENOMEM;
145 }
146
147 page[1] = grab_cache_page_write_begin(mapping[1], index2);
148 memalloc_nofs_restore(flags);
149 if (!page[1]) {
150 unlock_page(page[0]);
151 put_page(page[0]);
152 return -ENOMEM;
153 }
154 /*
155 * grab_cache_page_write_begin() may not wait on page's writeback if
156 * BDI not demand that. But it is reasonable to be very conservative
157 * here and explicitly wait on page's writeback
158 */
159 wait_on_page_writeback(page[0]);
160 wait_on_page_writeback(page[1]);
161 if (inode1 > inode2)
162 swap(page[0], page[1]);
163
164 return 0;
165}
166
167/* Force page buffers uptodate w/o dropping page's lock */
168static int
169mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
170{
171 struct inode *inode = page->mapping->host;
172 sector_t block;
173 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
174 unsigned int blocksize, block_start, block_end;
175 int i, err, nr = 0, partial = 0;
176 BUG_ON(!PageLocked(page));
177 BUG_ON(PageWriteback(page));
178
179 if (PageUptodate(page))
180 return 0;
181
182 blocksize = i_blocksize(inode);
183 if (!page_has_buffers(page))
184 create_empty_buffers(page, blocksize, 0);
185
186 head = page_buffers(page);
187 block = (sector_t)page->index << (PAGE_SHIFT - inode->i_blkbits);
188 for (bh = head, block_start = 0; bh != head || !block_start;
189 block++, block_start = block_end, bh = bh->b_this_page) {
190 block_end = block_start + blocksize;
191 if (block_end <= from || block_start >= to) {
192 if (!buffer_uptodate(bh))
193 partial = 1;
194 continue;
195 }
196 if (buffer_uptodate(bh))
197 continue;
198 if (!buffer_mapped(bh)) {
199 err = ext4_get_block(inode, block, bh, 0);
200 if (err) {
201 SetPageError(page);
202 return err;
203 }
204 if (!buffer_mapped(bh)) {
205 zero_user(page, block_start, blocksize);
206 set_buffer_uptodate(bh);
207 continue;
208 }
209 }
210 BUG_ON(nr >= MAX_BUF_PER_PAGE);
211 arr[nr++] = bh;
212 }
213 /* No io required */
214 if (!nr)
215 goto out;
216
217 for (i = 0; i < nr; i++) {
218 bh = arr[i];
219 if (!bh_uptodate_or_lock(bh)) {
220 err = ext4_read_bh(bh, 0, NULL);
221 if (err)
222 return err;
223 }
224 }
225out:
226 if (!partial)
227 SetPageUptodate(page);
228 return 0;
229}
230
231/**
232 * move_extent_per_page - Move extent data per page
233 *
234 * @o_filp: file structure of original file
235 * @donor_inode: donor inode
236 * @orig_page_offset: page index on original file
237 * @donor_page_offset: page index on donor file
238 * @data_offset_in_page: block index where data swapping starts
239 * @block_len_in_page: the number of blocks to be swapped
240 * @unwritten: orig extent is unwritten or not
241 * @err: pointer to save return value
242 *
243 * Save the data in original inode blocks and replace original inode extents
244 * with donor inode extents by calling ext4_swap_extents().
245 * Finally, write out the saved data in new original inode blocks. Return
246 * replaced block count.
247 */
248static int
249move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
250 pgoff_t orig_page_offset, pgoff_t donor_page_offset,
251 int data_offset_in_page,
252 int block_len_in_page, int unwritten, int *err)
253{
254 struct inode *orig_inode = file_inode(o_filp);
255 struct page *pagep[2] = {NULL, NULL};
256 struct folio *folio[2] = {NULL, NULL};
257 handle_t *handle;
258 ext4_lblk_t orig_blk_offset, donor_blk_offset;
259 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
260 unsigned int tmp_data_size, data_size, replaced_size;
261 int i, err2, jblocks, retries = 0;
262 int replaced_count = 0;
263 int from = data_offset_in_page << orig_inode->i_blkbits;
264 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
265 struct super_block *sb = orig_inode->i_sb;
266 struct buffer_head *bh = NULL;
267
268 /*
269 * It needs twice the amount of ordinary journal buffers because
270 * inode and donor_inode may change each different metadata blocks.
271 */
272again:
273 *err = 0;
274 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
275 handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
276 if (IS_ERR(handle)) {
277 *err = PTR_ERR(handle);
278 return 0;
279 }
280
281 orig_blk_offset = orig_page_offset * blocks_per_page +
282 data_offset_in_page;
283
284 donor_blk_offset = donor_page_offset * blocks_per_page +
285 data_offset_in_page;
286
287 /* Calculate data_size */
288 if ((orig_blk_offset + block_len_in_page - 1) ==
289 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
290 /* Replace the last block */
291 tmp_data_size = orig_inode->i_size & (blocksize - 1);
292 /*
293 * If data_size equal zero, it shows data_size is multiples of
294 * blocksize. So we set appropriate value.
295 */
296 if (tmp_data_size == 0)
297 tmp_data_size = blocksize;
298
299 data_size = tmp_data_size +
300 ((block_len_in_page - 1) << orig_inode->i_blkbits);
301 } else
302 data_size = block_len_in_page << orig_inode->i_blkbits;
303
304 replaced_size = data_size;
305
306 *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
307 donor_page_offset, pagep);
308 if (unlikely(*err < 0))
309 goto stop_journal;
310 /*
311 * If orig extent was unwritten it can become initialized
312 * at any time after i_data_sem was dropped, in order to
313 * serialize with delalloc we have recheck extent while we
314 * hold page's lock, if it is still the case data copy is not
315 * necessary, just swap data blocks between orig and donor.
316 */
317 folio[0] = page_folio(pagep[0]);
318 folio[1] = page_folio(pagep[1]);
319
320 VM_BUG_ON_FOLIO(folio_test_large(folio[0]), folio[0]);
321 VM_BUG_ON_FOLIO(folio_test_large(folio[1]), folio[1]);
322 VM_BUG_ON_FOLIO(folio_nr_pages(folio[0]) != folio_nr_pages(folio[1]), folio[1]);
323
324 if (unwritten) {
325 ext4_double_down_write_data_sem(orig_inode, donor_inode);
326 /* If any of extents in range became initialized we have to
327 * fallback to data copying */
328 unwritten = mext_check_coverage(orig_inode, orig_blk_offset,
329 block_len_in_page, 1, err);
330 if (*err)
331 goto drop_data_sem;
332
333 unwritten &= mext_check_coverage(donor_inode, donor_blk_offset,
334 block_len_in_page, 1, err);
335 if (*err)
336 goto drop_data_sem;
337
338 if (!unwritten) {
339 ext4_double_up_write_data_sem(orig_inode, donor_inode);
340 goto data_copy;
341 }
342 if ((folio_has_private(folio[0]) &&
343 !filemap_release_folio(folio[0], 0)) ||
344 (folio_has_private(folio[1]) &&
345 !filemap_release_folio(folio[1], 0))) {
346 *err = -EBUSY;
347 goto drop_data_sem;
348 }
349 replaced_count = ext4_swap_extents(handle, orig_inode,
350 donor_inode, orig_blk_offset,
351 donor_blk_offset,
352 block_len_in_page, 1, err);
353 drop_data_sem:
354 ext4_double_up_write_data_sem(orig_inode, donor_inode);
355 goto unlock_folios;
356 }
357data_copy:
358 *err = mext_page_mkuptodate(&folio[0]->page, from, from + replaced_size);
359 if (*err)
360 goto unlock_folios;
361
362 /* At this point all buffers in range are uptodate, old mapping layout
363 * is no longer required, try to drop it now. */
364 if ((folio_has_private(folio[0]) &&
365 !filemap_release_folio(folio[0], 0)) ||
366 (folio_has_private(folio[1]) &&
367 !filemap_release_folio(folio[1], 0))) {
368 *err = -EBUSY;
369 goto unlock_folios;
370 }
371 ext4_double_down_write_data_sem(orig_inode, donor_inode);
372 replaced_count = ext4_swap_extents(handle, orig_inode, donor_inode,
373 orig_blk_offset, donor_blk_offset,
374 block_len_in_page, 1, err);
375 ext4_double_up_write_data_sem(orig_inode, donor_inode);
376 if (*err) {
377 if (replaced_count) {
378 block_len_in_page = replaced_count;
379 replaced_size =
380 block_len_in_page << orig_inode->i_blkbits;
381 } else
382 goto unlock_folios;
383 }
384 /* Perform all necessary steps similar write_begin()/write_end()
385 * but keeping in mind that i_size will not change */
386 if (!folio_buffers(folio[0]))
387 create_empty_buffers(&folio[0]->page, 1 << orig_inode->i_blkbits, 0);
388 bh = folio_buffers(folio[0]);
389 for (i = 0; i < data_offset_in_page; i++)
390 bh = bh->b_this_page;
391 for (i = 0; i < block_len_in_page; i++) {
392 *err = ext4_get_block(orig_inode, orig_blk_offset + i, bh, 0);
393 if (*err < 0)
394 break;
395 bh = bh->b_this_page;
396 }
397 if (!*err)
398 *err = block_commit_write(&folio[0]->page, from, from + replaced_size);
399
400 if (unlikely(*err < 0))
401 goto repair_branches;
402
403 /* Even in case of data=writeback it is reasonable to pin
404 * inode to transaction, to prevent unexpected data loss */
405 *err = ext4_jbd2_inode_add_write(handle, orig_inode,
406 (loff_t)orig_page_offset << PAGE_SHIFT, replaced_size);
407
408unlock_folios:
409 folio_unlock(folio[0]);
410 folio_put(folio[0]);
411 folio_unlock(folio[1]);
412 folio_put(folio[1]);
413stop_journal:
414 ext4_journal_stop(handle);
415 if (*err == -ENOSPC &&
416 ext4_should_retry_alloc(sb, &retries))
417 goto again;
418 /* Buffer was busy because probably is pinned to journal transaction,
419 * force transaction commit may help to free it. */
420 if (*err == -EBUSY && retries++ < 4 && EXT4_SB(sb)->s_journal &&
421 jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal))
422 goto again;
423 return replaced_count;
424
425repair_branches:
426 /*
427 * This should never ever happen!
428 * Extents are swapped already, but we are not able to copy data.
429 * Try to swap extents to it's original places
430 */
431 ext4_double_down_write_data_sem(orig_inode, donor_inode);
432 replaced_count = ext4_swap_extents(handle, donor_inode, orig_inode,
433 orig_blk_offset, donor_blk_offset,
434 block_len_in_page, 0, &err2);
435 ext4_double_up_write_data_sem(orig_inode, donor_inode);
436 if (replaced_count != block_len_in_page) {
437 ext4_error_inode_block(orig_inode, (sector_t)(orig_blk_offset),
438 EIO, "Unable to copy data block,"
439 " data will be lost.");
440 *err = -EIO;
441 }
442 replaced_count = 0;
443 goto unlock_folios;
444}
445
446/**
447 * mext_check_arguments - Check whether move extent can be done
448 *
449 * @orig_inode: original inode
450 * @donor_inode: donor inode
451 * @orig_start: logical start offset in block for orig
452 * @donor_start: logical start offset in block for donor
453 * @len: the number of blocks to be moved
454 *
455 * Check the arguments of ext4_move_extents() whether the files can be
456 * exchanged with each other.
457 * Return 0 on success, or a negative error value on failure.
458 */
459static int
460mext_check_arguments(struct inode *orig_inode,
461 struct inode *donor_inode, __u64 orig_start,
462 __u64 donor_start, __u64 *len)
463{
464 __u64 orig_eof, donor_eof;
465 unsigned int blkbits = orig_inode->i_blkbits;
466 unsigned int blocksize = 1 << blkbits;
467
468 orig_eof = (i_size_read(orig_inode) + blocksize - 1) >> blkbits;
469 donor_eof = (i_size_read(donor_inode) + blocksize - 1) >> blkbits;
470
471
472 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
473 ext4_debug("ext4 move extent: suid or sgid is set"
474 " to donor file [ino:orig %lu, donor %lu]\n",
475 orig_inode->i_ino, donor_inode->i_ino);
476 return -EINVAL;
477 }
478
479 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
480 return -EPERM;
481
482 /* Ext4 move extent does not support swap files */
483 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
484 ext4_debug("ext4 move extent: The argument files should not be swap files [ino:orig %lu, donor %lu]\n",
485 orig_inode->i_ino, donor_inode->i_ino);
486 return -ETXTBSY;
487 }
488
489 if (ext4_is_quota_file(orig_inode) && ext4_is_quota_file(donor_inode)) {
490 ext4_debug("ext4 move extent: The argument files should not be quota files [ino:orig %lu, donor %lu]\n",
491 orig_inode->i_ino, donor_inode->i_ino);
492 return -EOPNOTSUPP;
493 }
494
495 /* Ext4 move extent supports only extent based file */
496 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
497 ext4_debug("ext4 move extent: orig file is not extents "
498 "based file [ino:orig %lu]\n", orig_inode->i_ino);
499 return -EOPNOTSUPP;
500 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
501 ext4_debug("ext4 move extent: donor file is not extents "
502 "based file [ino:donor %lu]\n", donor_inode->i_ino);
503 return -EOPNOTSUPP;
504 }
505
506 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
507 ext4_debug("ext4 move extent: File size is 0 byte\n");
508 return -EINVAL;
509 }
510
511 /* Start offset should be same */
512 if ((orig_start & ~(PAGE_MASK >> orig_inode->i_blkbits)) !=
513 (donor_start & ~(PAGE_MASK >> orig_inode->i_blkbits))) {
514 ext4_debug("ext4 move extent: orig and donor's start "
515 "offsets are not aligned [ino:orig %lu, donor %lu]\n",
516 orig_inode->i_ino, donor_inode->i_ino);
517 return -EINVAL;
518 }
519
520 if ((orig_start >= EXT_MAX_BLOCKS) ||
521 (donor_start >= EXT_MAX_BLOCKS) ||
522 (*len > EXT_MAX_BLOCKS) ||
523 (donor_start + *len >= EXT_MAX_BLOCKS) ||
524 (orig_start + *len >= EXT_MAX_BLOCKS)) {
525 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
526 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
527 orig_inode->i_ino, donor_inode->i_ino);
528 return -EINVAL;
529 }
530 if (orig_eof <= orig_start)
531 *len = 0;
532 else if (orig_eof < orig_start + *len - 1)
533 *len = orig_eof - orig_start;
534 if (donor_eof <= donor_start)
535 *len = 0;
536 else if (donor_eof < donor_start + *len - 1)
537 *len = donor_eof - donor_start;
538 if (!*len) {
539 ext4_debug("ext4 move extent: len should not be 0 "
540 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
541 donor_inode->i_ino);
542 return -EINVAL;
543 }
544
545 return 0;
546}
547
548/**
549 * ext4_move_extents - Exchange the specified range of a file
550 *
551 * @o_filp: file structure of the original file
552 * @d_filp: file structure of the donor file
553 * @orig_blk: start offset in block for orig
554 * @donor_blk: start offset in block for donor
555 * @len: the number of blocks to be moved
556 * @moved_len: moved block length
557 *
558 * This function returns 0 and moved block length is set in moved_len
559 * if succeed, otherwise returns error value.
560 *
561 */
562int
563ext4_move_extents(struct file *o_filp, struct file *d_filp, __u64 orig_blk,
564 __u64 donor_blk, __u64 len, __u64 *moved_len)
565{
566 struct inode *orig_inode = file_inode(o_filp);
567 struct inode *donor_inode = file_inode(d_filp);
568 struct ext4_ext_path *path = NULL;
569 int blocks_per_page = PAGE_SIZE >> orig_inode->i_blkbits;
570 ext4_lblk_t o_end, o_start = orig_blk;
571 ext4_lblk_t d_start = donor_blk;
572 int ret;
573
574 if (orig_inode->i_sb != donor_inode->i_sb) {
575 ext4_debug("ext4 move extent: The argument files "
576 "should be in same FS [ino:orig %lu, donor %lu]\n",
577 orig_inode->i_ino, donor_inode->i_ino);
578 return -EINVAL;
579 }
580
581 /* orig and donor should be different inodes */
582 if (orig_inode == donor_inode) {
583 ext4_debug("ext4 move extent: The argument files should not "
584 "be same inode [ino:orig %lu, donor %lu]\n",
585 orig_inode->i_ino, donor_inode->i_ino);
586 return -EINVAL;
587 }
588
589 /* Regular file check */
590 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
591 ext4_debug("ext4 move extent: The argument files should be "
592 "regular file [ino:orig %lu, donor %lu]\n",
593 orig_inode->i_ino, donor_inode->i_ino);
594 return -EINVAL;
595 }
596
597 /* TODO: it's not obvious how to swap blocks for inodes with full
598 journaling enabled */
599 if (ext4_should_journal_data(orig_inode) ||
600 ext4_should_journal_data(donor_inode)) {
601 ext4_msg(orig_inode->i_sb, KERN_ERR,
602 "Online defrag not supported with data journaling");
603 return -EOPNOTSUPP;
604 }
605
606 if (IS_ENCRYPTED(orig_inode) || IS_ENCRYPTED(donor_inode)) {
607 ext4_msg(orig_inode->i_sb, KERN_ERR,
608 "Online defrag not supported for encrypted files");
609 return -EOPNOTSUPP;
610 }
611
612 /* Protect orig and donor inodes against a truncate */
613 lock_two_nondirectories(orig_inode, donor_inode);
614
615 /* Wait for all existing dio workers */
616 inode_dio_wait(orig_inode);
617 inode_dio_wait(donor_inode);
618
619 /* Protect extent tree against block allocations via delalloc */
620 ext4_double_down_write_data_sem(orig_inode, donor_inode);
621 /* Check the filesystem environment whether move_extent can be done */
622 ret = mext_check_arguments(orig_inode, donor_inode, orig_blk,
623 donor_blk, &len);
624 if (ret)
625 goto out;
626 o_end = o_start + len;
627
628 while (o_start < o_end) {
629 struct ext4_extent *ex;
630 ext4_lblk_t cur_blk, next_blk;
631 pgoff_t orig_page_index, donor_page_index;
632 int offset_in_page;
633 int unwritten, cur_len;
634
635 ret = get_ext_path(orig_inode, o_start, &path);
636 if (ret)
637 goto out;
638 ex = path[path->p_depth].p_ext;
639 cur_blk = le32_to_cpu(ex->ee_block);
640 cur_len = ext4_ext_get_actual_len(ex);
641 /* Check hole before the start pos */
642 if (cur_blk + cur_len - 1 < o_start) {
643 next_blk = ext4_ext_next_allocated_block(path);
644 if (next_blk == EXT_MAX_BLOCKS) {
645 ret = -ENODATA;
646 goto out;
647 }
648 d_start += next_blk - o_start;
649 o_start = next_blk;
650 continue;
651 /* Check hole after the start pos */
652 } else if (cur_blk > o_start) {
653 /* Skip hole */
654 d_start += cur_blk - o_start;
655 o_start = cur_blk;
656 /* Extent inside requested range ?*/
657 if (cur_blk >= o_end)
658 goto out;
659 } else { /* in_range(o_start, o_blk, o_len) */
660 cur_len += cur_blk - o_start;
661 }
662 unwritten = ext4_ext_is_unwritten(ex);
663 if (o_end - o_start < cur_len)
664 cur_len = o_end - o_start;
665
666 orig_page_index = o_start >> (PAGE_SHIFT -
667 orig_inode->i_blkbits);
668 donor_page_index = d_start >> (PAGE_SHIFT -
669 donor_inode->i_blkbits);
670 offset_in_page = o_start % blocks_per_page;
671 if (cur_len > blocks_per_page - offset_in_page)
672 cur_len = blocks_per_page - offset_in_page;
673 /*
674 * Up semaphore to avoid following problems:
675 * a. transaction deadlock among ext4_journal_start,
676 * ->write_begin via pagefault, and jbd2_journal_commit
677 * b. racing with ->read_folio, ->write_begin, and
678 * ext4_get_block in move_extent_per_page
679 */
680 ext4_double_up_write_data_sem(orig_inode, donor_inode);
681 /* Swap original branches with new branches */
682 move_extent_per_page(o_filp, donor_inode,
683 orig_page_index, donor_page_index,
684 offset_in_page, cur_len,
685 unwritten, &ret);
686 ext4_double_down_write_data_sem(orig_inode, donor_inode);
687 if (ret < 0)
688 break;
689 o_start += cur_len;
690 d_start += cur_len;
691 }
692 *moved_len = o_start - orig_blk;
693 if (*moved_len > len)
694 *moved_len = len;
695
696out:
697 if (*moved_len) {
698 ext4_discard_preallocations(orig_inode, 0);
699 ext4_discard_preallocations(donor_inode, 0);
700 }
701
702 ext4_free_ext_path(path);
703 ext4_double_up_write_data_sem(orig_inode, donor_inode);
704 unlock_two_nondirectories(orig_inode, donor_inode);
705
706 return ret;
707}