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_extents.h"
21#include "ext4.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 **path)
36{
37 int ret = 0;
38
39 *path = ext4_ext_find_extent(inode, lblock, *path);
40 if (IS_ERR(*path)) {
41 ret = PTR_ERR(*path);
42 *path = NULL;
43 } else if ((*path)[ext_depth(inode)].p_ext == NULL)
44 ret = -ENODATA;
45
46 return ret;
47}
48
49/**
50 * copy_extent_status - Copy the extent's initialization status
51 *
52 * @src: an extent for getting initialize status
53 * @dest: an extent to be set the status
54 */
55static void
56copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
57{
58 if (ext4_ext_is_uninitialized(src))
59 ext4_ext_mark_uninitialized(dest);
60 else
61 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
62}
63
64/**
65 * mext_next_extent - Search for the next extent and set it to "extent"
66 *
67 * @inode: inode which is searched
68 * @path: this will obtain data for the next extent
69 * @extent: pointer to the next extent we have just gotten
70 *
71 * Search the next extent in the array of ext4_ext_path structure (@path)
72 * and set it to ext4_extent structure (@extent). In addition, the member of
73 * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
74 * ext4_ext_path structure refers to the last extent, or a negative error
75 * value on failure.
76 */
77static int
78mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
79 struct ext4_extent **extent)
80{
81 struct ext4_extent_header *eh;
82 int ppos, leaf_ppos = path->p_depth;
83
84 ppos = leaf_ppos;
85 if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
86 /* leaf block */
87 *extent = ++path[ppos].p_ext;
88 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
89 return 0;
90 }
91
92 while (--ppos >= 0) {
93 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
94 path[ppos].p_idx) {
95 int cur_ppos = ppos;
96
97 /* index block */
98 path[ppos].p_idx++;
99 path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
100 if (path[ppos+1].p_bh)
101 brelse(path[ppos+1].p_bh);
102 path[ppos+1].p_bh =
103 sb_bread(inode->i_sb, path[ppos].p_block);
104 if (!path[ppos+1].p_bh)
105 return -EIO;
106 path[ppos+1].p_hdr =
107 ext_block_hdr(path[ppos+1].p_bh);
108
109 /* Halfway index block */
110 while (++cur_ppos < leaf_ppos) {
111 path[cur_ppos].p_idx =
112 EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
113 path[cur_ppos].p_block =
114 ext4_idx_pblock(path[cur_ppos].p_idx);
115 if (path[cur_ppos+1].p_bh)
116 brelse(path[cur_ppos+1].p_bh);
117 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
118 path[cur_ppos].p_block);
119 if (!path[cur_ppos+1].p_bh)
120 return -EIO;
121 path[cur_ppos+1].p_hdr =
122 ext_block_hdr(path[cur_ppos+1].p_bh);
123 }
124
125 path[leaf_ppos].p_ext = *extent = NULL;
126
127 eh = path[leaf_ppos].p_hdr;
128 if (le16_to_cpu(eh->eh_entries) == 0)
129 /* empty leaf is found */
130 return -ENODATA;
131
132 /* leaf block */
133 path[leaf_ppos].p_ext = *extent =
134 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
135 path[leaf_ppos].p_block =
136 ext4_ext_pblock(path[leaf_ppos].p_ext);
137 return 0;
138 }
139 }
140 /* We found the last extent */
141 return 1;
142}
143
144/**
145 * mext_check_null_inode - NULL check for two inodes
146 *
147 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
148 */
149static int
150mext_check_null_inode(struct inode *inode1, struct inode *inode2,
151 const char *function, unsigned int line)
152{
153 int ret = 0;
154
155 if (inode1 == NULL) {
156 __ext4_error(inode2->i_sb, function, line,
157 "Both inodes should not be NULL: "
158 "inode1 NULL inode2 %lu", inode2->i_ino);
159 ret = -EIO;
160 } else if (inode2 == NULL) {
161 __ext4_error(inode1->i_sb, function, line,
162 "Both inodes should not be NULL: "
163 "inode1 %lu inode2 NULL", inode1->i_ino);
164 ret = -EIO;
165 }
166 return ret;
167}
168
169/**
170 * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
171 *
172 * @orig_inode: original inode structure
173 * @donor_inode: donor inode structure
174 * Acquire write lock of i_data_sem of the two inodes (orig and donor) by
175 * i_ino order.
176 */
177static void
178double_down_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
179{
180 struct inode *first = orig_inode, *second = donor_inode;
181
182 /*
183 * Use the inode number to provide the stable locking order instead
184 * of its address, because the C language doesn't guarantee you can
185 * compare pointers that don't come from the same array.
186 */
187 if (donor_inode->i_ino < orig_inode->i_ino) {
188 first = donor_inode;
189 second = orig_inode;
190 }
191
192 down_write(&EXT4_I(first)->i_data_sem);
193 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
194}
195
196/**
197 * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
198 *
199 * @orig_inode: original inode structure to be released its lock first
200 * @donor_inode: donor inode structure to be released its lock second
201 * Release write lock of i_data_sem of two inodes (orig and donor).
202 */
203static void
204double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
205{
206 up_write(&EXT4_I(orig_inode)->i_data_sem);
207 up_write(&EXT4_I(donor_inode)->i_data_sem);
208}
209
210/**
211 * mext_insert_across_blocks - Insert extents across leaf block
212 *
213 * @handle: journal handle
214 * @orig_inode: original inode
215 * @o_start: first original extent to be changed
216 * @o_end: last original extent to be changed
217 * @start_ext: first new extent to be inserted
218 * @new_ext: middle of new extent to be inserted
219 * @end_ext: last new extent to be inserted
220 *
221 * Allocate a new leaf block and insert extents into it. Return 0 on success,
222 * or a negative error value on failure.
223 */
224static int
225mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
226 struct ext4_extent *o_start, struct ext4_extent *o_end,
227 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
228 struct ext4_extent *end_ext)
229{
230 struct ext4_ext_path *orig_path = NULL;
231 ext4_lblk_t eblock = 0;
232 int new_flag = 0;
233 int end_flag = 0;
234 int err = 0;
235
236 if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
237 if (o_start == o_end) {
238
239 /* start_ext new_ext end_ext
240 * donor |---------|-----------|--------|
241 * orig |------------------------------|
242 */
243 end_flag = 1;
244 } else {
245
246 /* start_ext new_ext end_ext
247 * donor |---------|----------|---------|
248 * orig |---------------|--------------|
249 */
250 o_end->ee_block = end_ext->ee_block;
251 o_end->ee_len = end_ext->ee_len;
252 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
253 }
254
255 o_start->ee_len = start_ext->ee_len;
256 eblock = le32_to_cpu(start_ext->ee_block);
257 new_flag = 1;
258
259 } else if (start_ext->ee_len && new_ext->ee_len &&
260 !end_ext->ee_len && o_start == o_end) {
261
262 /* start_ext new_ext
263 * donor |--------------|---------------|
264 * orig |------------------------------|
265 */
266 o_start->ee_len = start_ext->ee_len;
267 eblock = le32_to_cpu(start_ext->ee_block);
268 new_flag = 1;
269
270 } else if (!start_ext->ee_len && new_ext->ee_len &&
271 end_ext->ee_len && o_start == o_end) {
272
273 /* new_ext end_ext
274 * donor |--------------|---------------|
275 * orig |------------------------------|
276 */
277 o_end->ee_block = end_ext->ee_block;
278 o_end->ee_len = end_ext->ee_len;
279 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
280
281 /*
282 * Set 0 to the extent block if new_ext was
283 * the first block.
284 */
285 if (new_ext->ee_block)
286 eblock = le32_to_cpu(new_ext->ee_block);
287
288 new_flag = 1;
289 } else {
290 ext4_debug("ext4 move extent: Unexpected insert case\n");
291 return -EIO;
292 }
293
294 if (new_flag) {
295 err = get_ext_path(orig_inode, eblock, &orig_path);
296 if (err)
297 goto out;
298
299 if (ext4_ext_insert_extent(handle, orig_inode,
300 orig_path, new_ext, 0))
301 goto out;
302 }
303
304 if (end_flag) {
305 err = get_ext_path(orig_inode,
306 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
307 if (err)
308 goto out;
309
310 if (ext4_ext_insert_extent(handle, orig_inode,
311 orig_path, end_ext, 0))
312 goto out;
313 }
314out:
315 if (orig_path) {
316 ext4_ext_drop_refs(orig_path);
317 kfree(orig_path);
318 }
319
320 return err;
321
322}
323
324/**
325 * mext_insert_inside_block - Insert new extent to the extent block
326 *
327 * @o_start: first original extent to be moved
328 * @o_end: last original extent to be moved
329 * @start_ext: first new extent to be inserted
330 * @new_ext: middle of new extent to be inserted
331 * @end_ext: last new extent to be inserted
332 * @eh: extent header of target leaf block
333 * @range_to_move: used to decide how to insert extent
334 *
335 * Insert extents into the leaf block. The extent (@o_start) is overwritten
336 * by inserted extents.
337 */
338static void
339mext_insert_inside_block(struct ext4_extent *o_start,
340 struct ext4_extent *o_end,
341 struct ext4_extent *start_ext,
342 struct ext4_extent *new_ext,
343 struct ext4_extent *end_ext,
344 struct ext4_extent_header *eh,
345 int range_to_move)
346{
347 int i = 0;
348 unsigned long len;
349
350 /* Move the existing extents */
351 if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
352 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
353 (unsigned long)(o_end + 1);
354 memmove(o_end + 1 + range_to_move, o_end + 1, len);
355 }
356
357 /* Insert start entry */
358 if (start_ext->ee_len)
359 o_start[i++].ee_len = start_ext->ee_len;
360
361 /* Insert new entry */
362 if (new_ext->ee_len) {
363 o_start[i] = *new_ext;
364 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
365 }
366
367 /* Insert end entry */
368 if (end_ext->ee_len)
369 o_start[i] = *end_ext;
370
371 /* Increment the total entries counter on the extent block */
372 le16_add_cpu(&eh->eh_entries, range_to_move);
373}
374
375/**
376 * mext_insert_extents - Insert new extent
377 *
378 * @handle: journal handle
379 * @orig_inode: original inode
380 * @orig_path: path indicates first extent to be changed
381 * @o_start: first original extent to be changed
382 * @o_end: last original extent to be changed
383 * @start_ext: first new extent to be inserted
384 * @new_ext: middle of new extent to be inserted
385 * @end_ext: last new extent to be inserted
386 *
387 * Call the function to insert extents. If we cannot add more extents into
388 * the leaf block, we call mext_insert_across_blocks() to create a
389 * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
390 * on success, or a negative error value on failure.
391 */
392static int
393mext_insert_extents(handle_t *handle, struct inode *orig_inode,
394 struct ext4_ext_path *orig_path,
395 struct ext4_extent *o_start,
396 struct ext4_extent *o_end,
397 struct ext4_extent *start_ext,
398 struct ext4_extent *new_ext,
399 struct ext4_extent *end_ext)
400{
401 struct ext4_extent_header *eh;
402 unsigned long need_slots, slots_range;
403 int range_to_move, depth, ret;
404
405 /*
406 * The extents need to be inserted
407 * start_extent + new_extent + end_extent.
408 */
409 need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
410 (new_ext->ee_len ? 1 : 0);
411
412 /* The number of slots between start and end */
413 slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
414 / sizeof(struct ext4_extent);
415
416 /* Range to move the end of extent */
417 range_to_move = need_slots - slots_range;
418 depth = orig_path->p_depth;
419 orig_path += depth;
420 eh = orig_path->p_hdr;
421
422 if (depth) {
423 /* Register to journal */
424 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
425 if (ret)
426 return ret;
427 }
428
429 /* Expansion */
430 if (range_to_move > 0 &&
431 (range_to_move > le16_to_cpu(eh->eh_max)
432 - le16_to_cpu(eh->eh_entries))) {
433
434 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
435 o_end, start_ext, new_ext, end_ext);
436 if (ret < 0)
437 return ret;
438 } else
439 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
440 end_ext, eh, range_to_move);
441
442 if (depth) {
443 ret = ext4_handle_dirty_metadata(handle, orig_inode,
444 orig_path->p_bh);
445 if (ret)
446 return ret;
447 } else {
448 ret = ext4_mark_inode_dirty(handle, orig_inode);
449 if (ret < 0)
450 return ret;
451 }
452
453 return 0;
454}
455
456/**
457 * mext_leaf_block - Move one leaf extent block into the inode.
458 *
459 * @handle: journal handle
460 * @orig_inode: original inode
461 * @orig_path: path indicates first extent to be changed
462 * @dext: donor extent
463 * @from: start offset on the target file
464 *
465 * In order to insert extents into the leaf block, we must divide the extent
466 * in the leaf block into three extents. The one is located to be inserted
467 * extents, and the others are located around it.
468 *
469 * Therefore, this function creates structures to save extents of the leaf
470 * block, and inserts extents by calling mext_insert_extents() with
471 * created extents. Return 0 on success, or a negative error value on failure.
472 */
473static int
474mext_leaf_block(handle_t *handle, struct inode *orig_inode,
475 struct ext4_ext_path *orig_path, struct ext4_extent *dext,
476 ext4_lblk_t *from)
477{
478 struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
479 struct ext4_extent new_ext, start_ext, end_ext;
480 ext4_lblk_t new_ext_end;
481 int oext_alen, new_ext_alen, end_ext_alen;
482 int depth = ext_depth(orig_inode);
483 int ret;
484
485 start_ext.ee_block = end_ext.ee_block = 0;
486 o_start = o_end = oext = orig_path[depth].p_ext;
487 oext_alen = ext4_ext_get_actual_len(oext);
488 start_ext.ee_len = end_ext.ee_len = 0;
489
490 new_ext.ee_block = cpu_to_le32(*from);
491 ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
492 new_ext.ee_len = dext->ee_len;
493 new_ext_alen = ext4_ext_get_actual_len(&new_ext);
494 new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
495
496 /*
497 * Case: original extent is first
498 * oext |--------|
499 * new_ext |--|
500 * start_ext |--|
501 */
502 if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
503 le32_to_cpu(new_ext.ee_block) <
504 le32_to_cpu(oext->ee_block) + oext_alen) {
505 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
506 le32_to_cpu(oext->ee_block));
507 start_ext.ee_block = oext->ee_block;
508 copy_extent_status(oext, &start_ext);
509 } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
510 prev_ext = oext - 1;
511 /*
512 * We can merge new_ext into previous extent,
513 * if these are contiguous and same extent type.
514 */
515 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
516 &new_ext)) {
517 o_start = prev_ext;
518 start_ext.ee_len = cpu_to_le16(
519 ext4_ext_get_actual_len(prev_ext) +
520 new_ext_alen);
521 start_ext.ee_block = oext->ee_block;
522 copy_extent_status(prev_ext, &start_ext);
523 new_ext.ee_len = 0;
524 }
525 }
526
527 /*
528 * Case: new_ext_end must be less than oext
529 * oext |-----------|
530 * new_ext |-------|
531 */
532 if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
533 EXT4_ERROR_INODE(orig_inode,
534 "new_ext_end(%u) should be less than or equal to "
535 "oext->ee_block(%u) + oext_alen(%d) - 1",
536 new_ext_end, le32_to_cpu(oext->ee_block),
537 oext_alen);
538 ret = -EIO;
539 goto out;
540 }
541
542 /*
543 * Case: new_ext is smaller than original extent
544 * oext |---------------|
545 * new_ext |-----------|
546 * end_ext |---|
547 */
548 if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
549 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
550 end_ext.ee_len =
551 cpu_to_le16(le32_to_cpu(oext->ee_block) +
552 oext_alen - 1 - new_ext_end);
553 copy_extent_status(oext, &end_ext);
554 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
555 ext4_ext_store_pblock(&end_ext,
556 (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
557 end_ext.ee_block =
558 cpu_to_le32(le32_to_cpu(o_end->ee_block) +
559 oext_alen - end_ext_alen);
560 }
561
562 ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
563 o_end, &start_ext, &new_ext, &end_ext);
564out:
565 return ret;
566}
567
568/**
569 * mext_calc_swap_extents - Calculate extents for extent swapping.
570 *
571 * @tmp_dext: the extent that will belong to the original inode
572 * @tmp_oext: the extent that will belong to the donor inode
573 * @orig_off: block offset of original inode
574 * @donor_off: block offset of donor inode
575 * @max_count: the maximum length of extents
576 *
577 * Return 0 on success, or a negative error value on failure.
578 */
579static int
580mext_calc_swap_extents(struct ext4_extent *tmp_dext,
581 struct ext4_extent *tmp_oext,
582 ext4_lblk_t orig_off, ext4_lblk_t donor_off,
583 ext4_lblk_t max_count)
584{
585 ext4_lblk_t diff, orig_diff;
586 struct ext4_extent dext_old, oext_old;
587
588 BUG_ON(orig_off != donor_off);
589
590 /* original and donor extents have to cover the same block offset */
591 if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
592 le32_to_cpu(tmp_oext->ee_block) +
593 ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
594 return -ENODATA;
595
596 if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
597 le32_to_cpu(tmp_dext->ee_block) +
598 ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
599 return -ENODATA;
600
601 dext_old = *tmp_dext;
602 oext_old = *tmp_oext;
603
604 /* When tmp_dext is too large, pick up the target range. */
605 diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
606
607 ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
608 tmp_dext->ee_block =
609 cpu_to_le32(le32_to_cpu(tmp_dext->ee_block) + diff);
610 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_dext->ee_len) - diff);
611
612 if (max_count < ext4_ext_get_actual_len(tmp_dext))
613 tmp_dext->ee_len = cpu_to_le16(max_count);
614
615 orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
616 ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
617
618 /* Adjust extent length if donor extent is larger than orig */
619 if (ext4_ext_get_actual_len(tmp_dext) >
620 ext4_ext_get_actual_len(tmp_oext) - orig_diff)
621 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
622 orig_diff);
623
624 tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
625
626 copy_extent_status(&oext_old, tmp_dext);
627 copy_extent_status(&dext_old, tmp_oext);
628
629 return 0;
630}
631
632/**
633 * mext_replace_branches - Replace original extents with new extents
634 *
635 * @handle: journal handle
636 * @orig_inode: original inode
637 * @donor_inode: donor inode
638 * @from: block offset of orig_inode
639 * @count: block count to be replaced
640 * @err: pointer to save return value
641 *
642 * Replace original inode extents and donor inode extents page by page.
643 * We implement this replacement in the following three steps:
644 * 1. Save the block information of original and donor inodes into
645 * dummy extents.
646 * 2. Change the block information of original inode to point at the
647 * donor inode blocks.
648 * 3. Change the block information of donor inode to point at the saved
649 * original inode blocks in the dummy extents.
650 *
651 * Return replaced block count.
652 */
653static int
654mext_replace_branches(handle_t *handle, struct inode *orig_inode,
655 struct inode *donor_inode, ext4_lblk_t from,
656 ext4_lblk_t count, int *err)
657{
658 struct ext4_ext_path *orig_path = NULL;
659 struct ext4_ext_path *donor_path = NULL;
660 struct ext4_extent *oext, *dext;
661 struct ext4_extent tmp_dext, tmp_oext;
662 ext4_lblk_t orig_off = from, donor_off = from;
663 int depth;
664 int replaced_count = 0;
665 int dext_alen;
666
667 /* Protect extent trees against block allocations via delalloc */
668 double_down_write_data_sem(orig_inode, donor_inode);
669
670 /* Get the original extent for the block "orig_off" */
671 *err = get_ext_path(orig_inode, orig_off, &orig_path);
672 if (*err)
673 goto out;
674
675 /* Get the donor extent for the head */
676 *err = get_ext_path(donor_inode, donor_off, &donor_path);
677 if (*err)
678 goto out;
679 depth = ext_depth(orig_inode);
680 oext = orig_path[depth].p_ext;
681 tmp_oext = *oext;
682
683 depth = ext_depth(donor_inode);
684 dext = donor_path[depth].p_ext;
685 tmp_dext = *dext;
686
687 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
688 donor_off, count);
689 if (*err)
690 goto out;
691
692 /* Loop for the donor extents */
693 while (1) {
694 /* The extent for donor must be found. */
695 if (!dext) {
696 EXT4_ERROR_INODE(donor_inode,
697 "The extent for donor must be found");
698 *err = -EIO;
699 goto out;
700 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
701 EXT4_ERROR_INODE(donor_inode,
702 "Donor offset(%u) and the first block of donor "
703 "extent(%u) should be equal",
704 donor_off,
705 le32_to_cpu(tmp_dext.ee_block));
706 *err = -EIO;
707 goto out;
708 }
709
710 /* Set donor extent to orig extent */
711 *err = mext_leaf_block(handle, orig_inode,
712 orig_path, &tmp_dext, &orig_off);
713 if (*err)
714 goto out;
715
716 /* Set orig extent to donor extent */
717 *err = mext_leaf_block(handle, donor_inode,
718 donor_path, &tmp_oext, &donor_off);
719 if (*err)
720 goto out;
721
722 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
723 replaced_count += dext_alen;
724 donor_off += dext_alen;
725 orig_off += dext_alen;
726
727 /* Already moved the expected blocks */
728 if (replaced_count >= count)
729 break;
730
731 if (orig_path)
732 ext4_ext_drop_refs(orig_path);
733 *err = get_ext_path(orig_inode, orig_off, &orig_path);
734 if (*err)
735 goto out;
736 depth = ext_depth(orig_inode);
737 oext = orig_path[depth].p_ext;
738 tmp_oext = *oext;
739
740 if (donor_path)
741 ext4_ext_drop_refs(donor_path);
742 *err = get_ext_path(donor_inode, donor_off, &donor_path);
743 if (*err)
744 goto out;
745 depth = ext_depth(donor_inode);
746 dext = donor_path[depth].p_ext;
747 tmp_dext = *dext;
748
749 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
750 donor_off, count - replaced_count);
751 if (*err)
752 goto out;
753 }
754
755out:
756 if (orig_path) {
757 ext4_ext_drop_refs(orig_path);
758 kfree(orig_path);
759 }
760 if (donor_path) {
761 ext4_ext_drop_refs(donor_path);
762 kfree(donor_path);
763 }
764
765 ext4_ext_invalidate_cache(orig_inode);
766 ext4_ext_invalidate_cache(donor_inode);
767
768 double_up_write_data_sem(orig_inode, donor_inode);
769
770 return replaced_count;
771}
772
773/**
774 * move_extent_per_page - Move extent data per page
775 *
776 * @o_filp: file structure of original file
777 * @donor_inode: donor inode
778 * @orig_page_offset: page index on original file
779 * @data_offset_in_page: block index where data swapping starts
780 * @block_len_in_page: the number of blocks to be swapped
781 * @uninit: orig extent is uninitialized or not
782 * @err: pointer to save return value
783 *
784 * Save the data in original inode blocks and replace original inode extents
785 * with donor inode extents by calling mext_replace_branches().
786 * Finally, write out the saved data in new original inode blocks. Return
787 * replaced block count.
788 */
789static int
790move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
791 pgoff_t orig_page_offset, int data_offset_in_page,
792 int block_len_in_page, int uninit, int *err)
793{
794 struct inode *orig_inode = o_filp->f_dentry->d_inode;
795 struct address_space *mapping = orig_inode->i_mapping;
796 struct buffer_head *bh;
797 struct page *page = NULL;
798 const struct address_space_operations *a_ops = mapping->a_ops;
799 handle_t *handle;
800 ext4_lblk_t orig_blk_offset;
801 long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
802 unsigned long blocksize = orig_inode->i_sb->s_blocksize;
803 unsigned int w_flags = 0;
804 unsigned int tmp_data_size, data_size, replaced_size;
805 void *fsdata;
806 int i, jblocks;
807 int err2 = 0;
808 int replaced_count = 0;
809 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
810
811 /*
812 * It needs twice the amount of ordinary journal buffers because
813 * inode and donor_inode may change each different metadata blocks.
814 */
815 jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
816 handle = ext4_journal_start(orig_inode, jblocks);
817 if (IS_ERR(handle)) {
818 *err = PTR_ERR(handle);
819 return 0;
820 }
821
822 if (segment_eq(get_fs(), KERNEL_DS))
823 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
824
825 orig_blk_offset = orig_page_offset * blocks_per_page +
826 data_offset_in_page;
827
828 /*
829 * If orig extent is uninitialized one,
830 * it's not necessary force the page into memory
831 * and then force it to be written out again.
832 * Just swap data blocks between orig and donor.
833 */
834 if (uninit) {
835 replaced_count = mext_replace_branches(handle, orig_inode,
836 donor_inode, orig_blk_offset,
837 block_len_in_page, err);
838 goto out2;
839 }
840
841 offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
842
843 /* Calculate data_size */
844 if ((orig_blk_offset + block_len_in_page - 1) ==
845 ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
846 /* Replace the last block */
847 tmp_data_size = orig_inode->i_size & (blocksize - 1);
848 /*
849 * If data_size equal zero, it shows data_size is multiples of
850 * blocksize. So we set appropriate value.
851 */
852 if (tmp_data_size == 0)
853 tmp_data_size = blocksize;
854
855 data_size = tmp_data_size +
856 ((block_len_in_page - 1) << orig_inode->i_blkbits);
857 } else
858 data_size = block_len_in_page << orig_inode->i_blkbits;
859
860 replaced_size = data_size;
861
862 *err = a_ops->write_begin(o_filp, mapping, offs, data_size, w_flags,
863 &page, &fsdata);
864 if (unlikely(*err < 0))
865 goto out;
866
867 if (!PageUptodate(page)) {
868 mapping->a_ops->readpage(o_filp, page);
869 lock_page(page);
870 }
871
872 /*
873 * try_to_release_page() doesn't call releasepage in writeback mode.
874 * We should care about the order of writing to the same file
875 * by multiple move extent processes.
876 * It needs to call wait_on_page_writeback() to wait for the
877 * writeback of the page.
878 */
879 wait_on_page_writeback(page);
880
881 /* Release old bh and drop refs */
882 try_to_release_page(page, 0);
883
884 replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
885 orig_blk_offset, block_len_in_page,
886 &err2);
887 if (err2) {
888 if (replaced_count) {
889 block_len_in_page = replaced_count;
890 replaced_size =
891 block_len_in_page << orig_inode->i_blkbits;
892 } else
893 goto out;
894 }
895
896 if (!page_has_buffers(page))
897 create_empty_buffers(page, 1 << orig_inode->i_blkbits, 0);
898
899 bh = page_buffers(page);
900 for (i = 0; i < data_offset_in_page; i++)
901 bh = bh->b_this_page;
902
903 for (i = 0; i < block_len_in_page; i++) {
904 *err = ext4_get_block(orig_inode,
905 (sector_t)(orig_blk_offset + i), bh, 0);
906 if (*err < 0)
907 goto out;
908
909 if (bh->b_this_page != NULL)
910 bh = bh->b_this_page;
911 }
912
913 *err = a_ops->write_end(o_filp, mapping, offs, data_size, replaced_size,
914 page, fsdata);
915 page = NULL;
916
917out:
918 if (unlikely(page)) {
919 if (PageLocked(page))
920 unlock_page(page);
921 page_cache_release(page);
922 ext4_journal_stop(handle);
923 }
924out2:
925 ext4_journal_stop(handle);
926
927 if (err2)
928 *err = err2;
929
930 return replaced_count;
931}
932
933/**
934 * mext_check_arguments - Check whether move extent can be done
935 *
936 * @orig_inode: original inode
937 * @donor_inode: donor inode
938 * @orig_start: logical start offset in block for orig
939 * @donor_start: logical start offset in block for donor
940 * @len: the number of blocks to be moved
941 *
942 * Check the arguments of ext4_move_extents() whether the files can be
943 * exchanged with each other.
944 * Return 0 on success, or a negative error value on failure.
945 */
946static int
947mext_check_arguments(struct inode *orig_inode,
948 struct inode *donor_inode, __u64 orig_start,
949 __u64 donor_start, __u64 *len)
950{
951 ext4_lblk_t orig_blocks, donor_blocks;
952 unsigned int blkbits = orig_inode->i_blkbits;
953 unsigned int blocksize = 1 << blkbits;
954
955 if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
956 ext4_debug("ext4 move extent: suid or sgid is set"
957 " to donor file [ino:orig %lu, donor %lu]\n",
958 orig_inode->i_ino, donor_inode->i_ino);
959 return -EINVAL;
960 }
961
962 if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
963 return -EPERM;
964
965 /* Ext4 move extent does not support swapfile */
966 if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
967 ext4_debug("ext4 move extent: The argument files should "
968 "not be swapfile [ino:orig %lu, donor %lu]\n",
969 orig_inode->i_ino, donor_inode->i_ino);
970 return -EINVAL;
971 }
972
973 /* Files should be in the same ext4 FS */
974 if (orig_inode->i_sb != donor_inode->i_sb) {
975 ext4_debug("ext4 move extent: The argument files "
976 "should be in same FS [ino:orig %lu, donor %lu]\n",
977 orig_inode->i_ino, donor_inode->i_ino);
978 return -EINVAL;
979 }
980
981 /* Ext4 move extent supports only extent based file */
982 if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
983 ext4_debug("ext4 move extent: orig file is not extents "
984 "based file [ino:orig %lu]\n", orig_inode->i_ino);
985 return -EOPNOTSUPP;
986 } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
987 ext4_debug("ext4 move extent: donor file is not extents "
988 "based file [ino:donor %lu]\n", donor_inode->i_ino);
989 return -EOPNOTSUPP;
990 }
991
992 if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
993 ext4_debug("ext4 move extent: File size is 0 byte\n");
994 return -EINVAL;
995 }
996
997 /* Start offset should be same */
998 if (orig_start != donor_start) {
999 ext4_debug("ext4 move extent: orig and donor's start "
1000 "offset are not same [ino:orig %lu, donor %lu]\n",
1001 orig_inode->i_ino, donor_inode->i_ino);
1002 return -EINVAL;
1003 }
1004
1005 if ((orig_start >= EXT_MAX_BLOCKS) ||
1006 (donor_start >= EXT_MAX_BLOCKS) ||
1007 (*len > EXT_MAX_BLOCKS) ||
1008 (orig_start + *len >= EXT_MAX_BLOCKS)) {
1009 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1010 "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1011 orig_inode->i_ino, donor_inode->i_ino);
1012 return -EINVAL;
1013 }
1014
1015 if (orig_inode->i_size > donor_inode->i_size) {
1016 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1017 /* TODO: eliminate this artificial restriction */
1018 if (orig_start >= donor_blocks) {
1019 ext4_debug("ext4 move extent: orig start offset "
1020 "[%llu] should be less than donor file blocks "
1021 "[%u] [ino:orig %lu, donor %lu]\n",
1022 orig_start, donor_blocks,
1023 orig_inode->i_ino, donor_inode->i_ino);
1024 return -EINVAL;
1025 }
1026
1027 /* TODO: eliminate this artificial restriction */
1028 if (orig_start + *len > donor_blocks) {
1029 ext4_debug("ext4 move extent: End offset [%llu] should "
1030 "be less than donor file blocks [%u]."
1031 "So adjust length from %llu to %llu "
1032 "[ino:orig %lu, donor %lu]\n",
1033 orig_start + *len, donor_blocks,
1034 *len, donor_blocks - orig_start,
1035 orig_inode->i_ino, donor_inode->i_ino);
1036 *len = donor_blocks - orig_start;
1037 }
1038 } else {
1039 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1040 if (orig_start >= orig_blocks) {
1041 ext4_debug("ext4 move extent: start offset [%llu] "
1042 "should be less than original file blocks "
1043 "[%u] [ino:orig %lu, donor %lu]\n",
1044 orig_start, orig_blocks,
1045 orig_inode->i_ino, donor_inode->i_ino);
1046 return -EINVAL;
1047 }
1048
1049 if (orig_start + *len > orig_blocks) {
1050 ext4_debug("ext4 move extent: Adjust length "
1051 "from %llu to %llu. Because it should be "
1052 "less than original file blocks "
1053 "[ino:orig %lu, donor %lu]\n",
1054 *len, orig_blocks - orig_start,
1055 orig_inode->i_ino, donor_inode->i_ino);
1056 *len = orig_blocks - orig_start;
1057 }
1058 }
1059
1060 if (!*len) {
1061 ext4_debug("ext4 move extent: len should not be 0 "
1062 "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1063 donor_inode->i_ino);
1064 return -EINVAL;
1065 }
1066
1067 return 0;
1068}
1069
1070/**
1071 * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1072 *
1073 * @inode1: the inode structure
1074 * @inode2: the inode structure
1075 *
1076 * Lock two inodes' i_mutex by i_ino order.
1077 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1078 */
1079static int
1080mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1081{
1082 int ret = 0;
1083
1084 BUG_ON(inode1 == NULL && inode2 == NULL);
1085
1086 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1087 if (ret < 0)
1088 goto out;
1089
1090 if (inode1 == inode2) {
1091 mutex_lock(&inode1->i_mutex);
1092 goto out;
1093 }
1094
1095 if (inode1->i_ino < inode2->i_ino) {
1096 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1097 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1098 } else {
1099 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1100 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1101 }
1102
1103out:
1104 return ret;
1105}
1106
1107/**
1108 * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1109 *
1110 * @inode1: the inode that is released first
1111 * @inode2: the inode that is released second
1112 *
1113 * If inode1 or inode2 is NULL, return -EIO. Otherwise, return 0.
1114 */
1115
1116static int
1117mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1118{
1119 int ret = 0;
1120
1121 BUG_ON(inode1 == NULL && inode2 == NULL);
1122
1123 ret = mext_check_null_inode(inode1, inode2, __func__, __LINE__);
1124 if (ret < 0)
1125 goto out;
1126
1127 if (inode1)
1128 mutex_unlock(&inode1->i_mutex);
1129
1130 if (inode2 && inode2 != inode1)
1131 mutex_unlock(&inode2->i_mutex);
1132
1133out:
1134 return ret;
1135}
1136
1137/**
1138 * ext4_move_extents - Exchange the specified range of a file
1139 *
1140 * @o_filp: file structure of the original file
1141 * @d_filp: file structure of the donor file
1142 * @orig_start: start offset in block for orig
1143 * @donor_start: start offset in block for donor
1144 * @len: the number of blocks to be moved
1145 * @moved_len: moved block length
1146 *
1147 * This function returns 0 and moved block length is set in moved_len
1148 * if succeed, otherwise returns error value.
1149 *
1150 * Note: ext4_move_extents() proceeds the following order.
1151 * 1:ext4_move_extents() calculates the last block number of moving extent
1152 * function by the start block number (orig_start) and the number of blocks
1153 * to be moved (len) specified as arguments.
1154 * If the {orig, donor}_start points a hole, the extent's start offset
1155 * pointed by ext_cur (current extent), holecheck_path, orig_path are set
1156 * after hole behind.
1157 * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1158 * or the ext_cur exceeds the block_end which is last logical block number.
1159 * 3:To get the length of continues area, call mext_next_extent()
1160 * specified with the ext_cur (initial value is holecheck_path) re-cursive,
1161 * until find un-continuous extent, the start logical block number exceeds
1162 * the block_end or the extent points to the last extent.
1163 * 4:Exchange the original inode data with donor inode data
1164 * from orig_page_offset to seq_end_page.
1165 * The start indexes of data are specified as arguments.
1166 * That of the original inode is orig_page_offset,
1167 * and the donor inode is also orig_page_offset
1168 * (To easily handle blocksize != pagesize case, the offset for the
1169 * donor inode is block unit).
1170 * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1171 * then returns to step 2.
1172 * 6:Release holecheck_path, orig_path and set the len to moved_len
1173 * which shows the number of moved blocks.
1174 * The moved_len is useful for the command to calculate the file offset
1175 * for starting next move extent ioctl.
1176 * 7:Return 0 on success, or a negative error value on failure.
1177 */
1178int
1179ext4_move_extents(struct file *o_filp, struct file *d_filp,
1180 __u64 orig_start, __u64 donor_start, __u64 len,
1181 __u64 *moved_len)
1182{
1183 struct inode *orig_inode = o_filp->f_dentry->d_inode;
1184 struct inode *donor_inode = d_filp->f_dentry->d_inode;
1185 struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1186 struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1187 ext4_lblk_t block_start = orig_start;
1188 ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1189 ext4_lblk_t rest_blocks;
1190 pgoff_t orig_page_offset = 0, seq_end_page;
1191 int ret1, ret2, depth, last_extent = 0;
1192 int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1193 int data_offset_in_page;
1194 int block_len_in_page;
1195 int uninit;
1196
1197 /* orig and donor should be different file */
1198 if (orig_inode->i_ino == donor_inode->i_ino) {
1199 ext4_debug("ext4 move extent: The argument files should not "
1200 "be same file [ino:orig %lu, donor %lu]\n",
1201 orig_inode->i_ino, donor_inode->i_ino);
1202 return -EINVAL;
1203 }
1204
1205 /* Regular file check */
1206 if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1207 ext4_debug("ext4 move extent: The argument files should be "
1208 "regular file [ino:orig %lu, donor %lu]\n",
1209 orig_inode->i_ino, donor_inode->i_ino);
1210 return -EINVAL;
1211 }
1212
1213 /* Protect orig and donor inodes against a truncate */
1214 ret1 = mext_inode_double_lock(orig_inode, donor_inode);
1215 if (ret1 < 0)
1216 return ret1;
1217
1218 /* Protect extent tree against block allocations via delalloc */
1219 double_down_write_data_sem(orig_inode, donor_inode);
1220 /* Check the filesystem environment whether move_extent can be done */
1221 ret1 = mext_check_arguments(orig_inode, donor_inode, orig_start,
1222 donor_start, &len);
1223 if (ret1)
1224 goto out;
1225
1226 file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1227 block_end = block_start + len - 1;
1228 if (file_end < block_end)
1229 len -= block_end - file_end;
1230
1231 ret1 = get_ext_path(orig_inode, block_start, &orig_path);
1232 if (ret1)
1233 goto out;
1234
1235 /* Get path structure to check the hole */
1236 ret1 = get_ext_path(orig_inode, block_start, &holecheck_path);
1237 if (ret1)
1238 goto out;
1239
1240 depth = ext_depth(orig_inode);
1241 ext_cur = holecheck_path[depth].p_ext;
1242
1243 /*
1244 * Get proper starting location of block replacement if block_start was
1245 * within the hole.
1246 */
1247 if (le32_to_cpu(ext_cur->ee_block) +
1248 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1249 /*
1250 * The hole exists between extents or the tail of
1251 * original file.
1252 */
1253 last_extent = mext_next_extent(orig_inode,
1254 holecheck_path, &ext_cur);
1255 if (last_extent < 0) {
1256 ret1 = last_extent;
1257 goto out;
1258 }
1259 last_extent = mext_next_extent(orig_inode, orig_path,
1260 &ext_dummy);
1261 if (last_extent < 0) {
1262 ret1 = last_extent;
1263 goto out;
1264 }
1265 seq_start = le32_to_cpu(ext_cur->ee_block);
1266 } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1267 /* The hole exists at the beginning of original file. */
1268 seq_start = le32_to_cpu(ext_cur->ee_block);
1269 else
1270 seq_start = block_start;
1271
1272 /* No blocks within the specified range. */
1273 if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1274 ext4_debug("ext4 move extent: The specified range of file "
1275 "may be the hole\n");
1276 ret1 = -EINVAL;
1277 goto out;
1278 }
1279
1280 /* Adjust start blocks */
1281 add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1282 ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1283 max(le32_to_cpu(ext_cur->ee_block), block_start);
1284
1285 while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1286 seq_blocks += add_blocks;
1287
1288 /* Adjust tail blocks */
1289 if (seq_start + seq_blocks - 1 > block_end)
1290 seq_blocks = block_end - seq_start + 1;
1291
1292 ext_prev = ext_cur;
1293 last_extent = mext_next_extent(orig_inode, holecheck_path,
1294 &ext_cur);
1295 if (last_extent < 0) {
1296 ret1 = last_extent;
1297 break;
1298 }
1299 add_blocks = ext4_ext_get_actual_len(ext_cur);
1300
1301 /*
1302 * Extend the length of contiguous block (seq_blocks)
1303 * if extents are contiguous.
1304 */
1305 if (ext4_can_extents_be_merged(orig_inode,
1306 ext_prev, ext_cur) &&
1307 block_end >= le32_to_cpu(ext_cur->ee_block) &&
1308 !last_extent)
1309 continue;
1310
1311 /* Is original extent is uninitialized */
1312 uninit = ext4_ext_is_uninitialized(ext_prev);
1313
1314 data_offset_in_page = seq_start % blocks_per_page;
1315
1316 /*
1317 * Calculate data blocks count that should be swapped
1318 * at the first page.
1319 */
1320 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1321 /* Swapped blocks are across pages */
1322 block_len_in_page =
1323 blocks_per_page - data_offset_in_page;
1324 } else {
1325 /* Swapped blocks are in a page */
1326 block_len_in_page = seq_blocks;
1327 }
1328
1329 orig_page_offset = seq_start >>
1330 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1331 seq_end_page = (seq_start + seq_blocks - 1) >>
1332 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1333 seq_start = le32_to_cpu(ext_cur->ee_block);
1334 rest_blocks = seq_blocks;
1335
1336 /*
1337 * Up semaphore to avoid following problems:
1338 * a. transaction deadlock among ext4_journal_start,
1339 * ->write_begin via pagefault, and jbd2_journal_commit
1340 * b. racing with ->readpage, ->write_begin, and ext4_get_block
1341 * in move_extent_per_page
1342 */
1343 double_up_write_data_sem(orig_inode, donor_inode);
1344
1345 while (orig_page_offset <= seq_end_page) {
1346
1347 /* Swap original branches with new branches */
1348 block_len_in_page = move_extent_per_page(
1349 o_filp, donor_inode,
1350 orig_page_offset,
1351 data_offset_in_page,
1352 block_len_in_page, uninit,
1353 &ret1);
1354
1355 /* Count how many blocks we have exchanged */
1356 *moved_len += block_len_in_page;
1357 if (ret1 < 0)
1358 break;
1359 if (*moved_len > len) {
1360 EXT4_ERROR_INODE(orig_inode,
1361 "We replaced blocks too much! "
1362 "sum of replaced: %llu requested: %llu",
1363 *moved_len, len);
1364 ret1 = -EIO;
1365 break;
1366 }
1367
1368 orig_page_offset++;
1369 data_offset_in_page = 0;
1370 rest_blocks -= block_len_in_page;
1371 if (rest_blocks > blocks_per_page)
1372 block_len_in_page = blocks_per_page;
1373 else
1374 block_len_in_page = rest_blocks;
1375 }
1376
1377 double_down_write_data_sem(orig_inode, donor_inode);
1378 if (ret1 < 0)
1379 break;
1380
1381 /* Decrease buffer counter */
1382 if (holecheck_path)
1383 ext4_ext_drop_refs(holecheck_path);
1384 ret1 = get_ext_path(orig_inode, seq_start, &holecheck_path);
1385 if (ret1)
1386 break;
1387 depth = holecheck_path->p_depth;
1388
1389 /* Decrease buffer counter */
1390 if (orig_path)
1391 ext4_ext_drop_refs(orig_path);
1392 ret1 = get_ext_path(orig_inode, seq_start, &orig_path);
1393 if (ret1)
1394 break;
1395
1396 ext_cur = holecheck_path[depth].p_ext;
1397 add_blocks = ext4_ext_get_actual_len(ext_cur);
1398 seq_blocks = 0;
1399
1400 }
1401out:
1402 if (*moved_len) {
1403 ext4_discard_preallocations(orig_inode);
1404 ext4_discard_preallocations(donor_inode);
1405 }
1406
1407 if (orig_path) {
1408 ext4_ext_drop_refs(orig_path);
1409 kfree(orig_path);
1410 }
1411 if (holecheck_path) {
1412 ext4_ext_drop_refs(holecheck_path);
1413 kfree(holecheck_path);
1414 }
1415 double_up_write_data_sem(orig_inode, donor_inode);
1416 ret2 = mext_inode_double_unlock(orig_inode, donor_inode);
1417
1418 if (ret1)
1419 return ret1;
1420 else if (ret2)
1421 return ret2;
1422
1423 return 0;
1424}
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}