1 // SPDX-License-Identifier: LGPL-2.1
3 * Copyright IBM Corporation, 2007
4 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
8 #include <linux/slab.h>
10 #include "ext4_extents.h"
13 * The contiguous blocks details which can be
14 * represented by a single extent
16 struct migrate_struct
{
17 ext4_lblk_t first_block
, last_block
, curr_block
;
18 ext4_fsblk_t first_pblock
, last_pblock
;
21 static int finish_range(handle_t
*handle
, struct inode
*inode
,
22 struct migrate_struct
*lb
)
25 int retval
= 0, needed
;
26 struct ext4_extent newext
;
27 struct ext4_ext_path
*path
;
28 if (lb
->first_pblock
== 0)
31 /* Add the extent to temp inode*/
32 newext
.ee_block
= cpu_to_le32(lb
->first_block
);
33 newext
.ee_len
= cpu_to_le16(lb
->last_block
- lb
->first_block
+ 1);
34 ext4_ext_store_pblock(&newext
, lb
->first_pblock
);
35 /* Locking only for convinience since we are operating on temp inode */
36 down_write(&EXT4_I(inode
)->i_data_sem
);
37 path
= ext4_find_extent(inode
, lb
->first_block
, NULL
, 0);
39 retval
= PTR_ERR(path
);
45 * Calculate the credit needed to inserting this extent
46 * Since we are doing this in loop we may accumalate extra
47 * credit. But below we try to not accumalate too much
48 * of them by restarting the journal.
50 needed
= ext4_ext_calc_credits_for_single_extent(inode
,
51 lb
->last_block
- lb
->first_block
+ 1, path
);
53 retval
= ext4_datasem_ensure_credits(handle
, inode
, needed
, needed
, 0);
56 retval
= ext4_ext_insert_extent(handle
, inode
, &path
, &newext
, 0);
58 up_write((&EXT4_I(inode
)->i_data_sem
));
59 ext4_ext_drop_refs(path
);
65 static int update_extent_range(handle_t
*handle
, struct inode
*inode
,
66 ext4_fsblk_t pblock
, struct migrate_struct
*lb
)
70 * See if we can add on to the existing range (if it exists)
72 if (lb
->first_pblock
&&
73 (lb
->last_pblock
+1 == pblock
) &&
74 (lb
->last_block
+1 == lb
->curr_block
)) {
75 lb
->last_pblock
= pblock
;
76 lb
->last_block
= lb
->curr_block
;
83 retval
= finish_range(handle
, inode
, lb
);
84 lb
->first_pblock
= lb
->last_pblock
= pblock
;
85 lb
->first_block
= lb
->last_block
= lb
->curr_block
;
90 static int update_ind_extent_range(handle_t
*handle
, struct inode
*inode
,
92 struct migrate_struct
*lb
)
94 struct buffer_head
*bh
;
97 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
99 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
103 i_data
= (__le32
*)bh
->b_data
;
104 for (i
= 0; i
< max_entries
; i
++) {
106 retval
= update_extent_range(handle
, inode
,
107 le32_to_cpu(i_data
[i
]), lb
);
119 static int update_dind_extent_range(handle_t
*handle
, struct inode
*inode
,
121 struct migrate_struct
*lb
)
123 struct buffer_head
*bh
;
126 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
128 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
132 i_data
= (__le32
*)bh
->b_data
;
133 for (i
= 0; i
< max_entries
; i
++) {
135 retval
= update_ind_extent_range(handle
, inode
,
136 le32_to_cpu(i_data
[i
]), lb
);
140 /* Only update the file block number */
141 lb
->curr_block
+= max_entries
;
149 static int update_tind_extent_range(handle_t
*handle
, struct inode
*inode
,
151 struct migrate_struct
*lb
)
153 struct buffer_head
*bh
;
156 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
158 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
162 i_data
= (__le32
*)bh
->b_data
;
163 for (i
= 0; i
< max_entries
; i
++) {
165 retval
= update_dind_extent_range(handle
, inode
,
166 le32_to_cpu(i_data
[i
]), lb
);
170 /* Only update the file block number */
171 lb
->curr_block
+= max_entries
* max_entries
;
179 static int free_dind_blocks(handle_t
*handle
,
180 struct inode
*inode
, __le32 i_data
)
184 struct buffer_head
*bh
;
185 struct super_block
*sb
= inode
->i_sb
;
186 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
189 bh
= ext4_sb_bread(sb
, le32_to_cpu(i_data
), 0);
193 tmp_idata
= (__le32
*)bh
->b_data
;
194 for (i
= 0; i
< max_entries
; i
++) {
196 err
= ext4_journal_ensure_credits(handle
,
197 EXT4_RESERVE_TRANS_BLOCKS
,
198 ext4_free_metadata_revoke_credits(sb
, 1));
203 ext4_free_blocks(handle
, inode
, NULL
,
204 le32_to_cpu(tmp_idata
[i
]), 1,
205 EXT4_FREE_BLOCKS_METADATA
|
206 EXT4_FREE_BLOCKS_FORGET
);
210 err
= ext4_journal_ensure_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
,
211 ext4_free_metadata_revoke_credits(sb
, 1));
214 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
215 EXT4_FREE_BLOCKS_METADATA
|
216 EXT4_FREE_BLOCKS_FORGET
);
220 static int free_tind_blocks(handle_t
*handle
,
221 struct inode
*inode
, __le32 i_data
)
225 struct buffer_head
*bh
;
226 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
228 bh
= ext4_sb_bread(inode
->i_sb
, le32_to_cpu(i_data
), 0);
232 tmp_idata
= (__le32
*)bh
->b_data
;
233 for (i
= 0; i
< max_entries
; i
++) {
235 retval
= free_dind_blocks(handle
,
236 inode
, tmp_idata
[i
]);
244 retval
= ext4_journal_ensure_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
,
245 ext4_free_metadata_revoke_credits(inode
->i_sb
, 1));
248 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
249 EXT4_FREE_BLOCKS_METADATA
|
250 EXT4_FREE_BLOCKS_FORGET
);
254 static int free_ind_block(handle_t
*handle
, struct inode
*inode
, __le32
*i_data
)
258 /* ei->i_data[EXT4_IND_BLOCK] */
260 retval
= ext4_journal_ensure_credits(handle
,
261 EXT4_RESERVE_TRANS_BLOCKS
,
262 ext4_free_metadata_revoke_credits(inode
->i_sb
, 1));
265 ext4_free_blocks(handle
, inode
, NULL
,
266 le32_to_cpu(i_data
[0]), 1,
267 EXT4_FREE_BLOCKS_METADATA
|
268 EXT4_FREE_BLOCKS_FORGET
);
271 /* ei->i_data[EXT4_DIND_BLOCK] */
273 retval
= free_dind_blocks(handle
, inode
, i_data
[1]);
278 /* ei->i_data[EXT4_TIND_BLOCK] */
280 retval
= free_tind_blocks(handle
, inode
, i_data
[2]);
287 static int ext4_ext_swap_inode_data(handle_t
*handle
, struct inode
*inode
,
288 struct inode
*tmp_inode
)
290 int retval
, retval2
= 0;
292 struct ext4_inode_info
*ei
= EXT4_I(inode
);
293 struct ext4_inode_info
*tmp_ei
= EXT4_I(tmp_inode
);
296 * One credit accounted for writing the
297 * i_data field of the original inode
299 retval
= ext4_journal_ensure_credits(handle
, 1, 0);
303 i_data
[0] = ei
->i_data
[EXT4_IND_BLOCK
];
304 i_data
[1] = ei
->i_data
[EXT4_DIND_BLOCK
];
305 i_data
[2] = ei
->i_data
[EXT4_TIND_BLOCK
];
307 down_write(&EXT4_I(inode
)->i_data_sem
);
309 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
310 * happened after we started the migrate. We need to
313 if (!ext4_test_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
)) {
315 up_write(&EXT4_I(inode
)->i_data_sem
);
318 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
320 * We have the extent map build with the tmp inode.
321 * Now copy the i_data across
323 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
324 memcpy(ei
->i_data
, tmp_ei
->i_data
, sizeof(ei
->i_data
));
327 * Update i_blocks with the new blocks that got
328 * allocated while adding extents for extent index
331 * While converting to extents we need not
332 * update the original inode i_blocks for extent blocks
333 * via quota APIs. The quota update happened via tmp_inode already.
335 spin_lock(&inode
->i_lock
);
336 inode
->i_blocks
+= tmp_inode
->i_blocks
;
337 spin_unlock(&inode
->i_lock
);
338 up_write(&EXT4_I(inode
)->i_data_sem
);
341 * We mark the inode dirty after, because we decrement the
342 * i_blocks when freeing the indirect meta-data blocks
344 retval
= free_ind_block(handle
, inode
, i_data
);
345 retval2
= ext4_mark_inode_dirty(handle
, inode
);
346 if (unlikely(retval2
&& !retval
))
353 static int free_ext_idx(handle_t
*handle
, struct inode
*inode
,
354 struct ext4_extent_idx
*ix
)
358 struct buffer_head
*bh
;
359 struct ext4_extent_header
*eh
;
361 block
= ext4_idx_pblock(ix
);
362 bh
= ext4_sb_bread(inode
->i_sb
, block
, 0);
366 eh
= (struct ext4_extent_header
*)bh
->b_data
;
367 if (eh
->eh_depth
!= 0) {
368 ix
= EXT_FIRST_INDEX(eh
);
369 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
370 retval
= free_ext_idx(handle
, inode
, ix
);
378 retval
= ext4_journal_ensure_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
,
379 ext4_free_metadata_revoke_credits(inode
->i_sb
, 1));
382 ext4_free_blocks(handle
, inode
, NULL
, block
, 1,
383 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
388 * Free the extent meta data blocks only
390 static int free_ext_block(handle_t
*handle
, struct inode
*inode
)
393 struct ext4_inode_info
*ei
= EXT4_I(inode
);
394 struct ext4_extent_header
*eh
= (struct ext4_extent_header
*)ei
->i_data
;
395 struct ext4_extent_idx
*ix
;
396 if (eh
->eh_depth
== 0)
398 * No extra blocks allocated for extent meta data
401 ix
= EXT_FIRST_INDEX(eh
);
402 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
403 retval
= free_ext_idx(handle
, inode
, ix
);
410 int ext4_ext_migrate(struct inode
*inode
)
412 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
416 struct ext4_inode_info
*ei
;
417 struct inode
*tmp_inode
= NULL
;
418 struct migrate_struct lb
;
419 unsigned long max_entries
;
424 * If the filesystem does not support extents, or the inode
425 * already is extent-based, error out.
427 if (!ext4_has_feature_extents(inode
->i_sb
) ||
428 (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
431 if (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
== 0)
433 * don't migrate fast symlink
437 percpu_down_write(&sbi
->s_writepages_rwsem
);
440 * Worst case we can touch the allocation bitmaps, a bgd
441 * block, and a block to link in the orphan list. We do need
442 * need to worry about credits for modifying the quota inode.
444 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
,
445 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
447 if (IS_ERR(handle
)) {
448 retval
= PTR_ERR(handle
);
451 goal
= (((inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(inode
->i_sb
)) *
452 EXT4_INODES_PER_GROUP(inode
->i_sb
)) + 1;
453 owner
[0] = i_uid_read(inode
);
454 owner
[1] = i_gid_read(inode
);
455 tmp_inode
= ext4_new_inode(handle
, d_inode(inode
->i_sb
->s_root
),
456 S_IFREG
, NULL
, goal
, owner
, 0);
457 if (IS_ERR(tmp_inode
)) {
458 retval
= PTR_ERR(tmp_inode
);
459 ext4_journal_stop(handle
);
462 i_size_write(tmp_inode
, i_size_read(inode
));
464 * Set the i_nlink to zero so it will be deleted later
465 * when we drop inode reference.
467 clear_nlink(tmp_inode
);
469 ext4_ext_tree_init(handle
, tmp_inode
);
470 ext4_orphan_add(handle
, tmp_inode
);
471 ext4_journal_stop(handle
);
474 * start with one credit accounted for
475 * superblock modification.
477 * For the tmp_inode we already have committed the
478 * transaction that created the inode. Later as and
479 * when we add extents we extent the journal
482 * Even though we take i_mutex we can still cause block
483 * allocation via mmap write to holes. If we have allocated
484 * new blocks we fail migrate. New block allocation will
485 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
486 * with i_data_sem held to prevent racing with block
489 down_read(&EXT4_I(inode
)->i_data_sem
);
490 ext4_set_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
491 up_read((&EXT4_I(inode
)->i_data_sem
));
493 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
494 if (IS_ERR(handle
)) {
496 * It is impossible to update on-disk structures without
497 * a handle, so just rollback in-core changes and live other
498 * work to orphan_list_cleanup()
500 ext4_orphan_del(NULL
, tmp_inode
);
501 retval
= PTR_ERR(handle
);
507 memset(&lb
, 0, sizeof(lb
));
509 /* 32 bit block address 4 bytes */
510 max_entries
= inode
->i_sb
->s_blocksize
>> 2;
511 for (i
= 0; i
< EXT4_NDIR_BLOCKS
; i
++) {
513 retval
= update_extent_range(handle
, tmp_inode
,
514 le32_to_cpu(i_data
[i
]), &lb
);
520 if (i_data
[EXT4_IND_BLOCK
]) {
521 retval
= update_ind_extent_range(handle
, tmp_inode
,
522 le32_to_cpu(i_data
[EXT4_IND_BLOCK
]), &lb
);
526 lb
.curr_block
+= max_entries
;
527 if (i_data
[EXT4_DIND_BLOCK
]) {
528 retval
= update_dind_extent_range(handle
, tmp_inode
,
529 le32_to_cpu(i_data
[EXT4_DIND_BLOCK
]), &lb
);
533 lb
.curr_block
+= max_entries
* max_entries
;
534 if (i_data
[EXT4_TIND_BLOCK
]) {
535 retval
= update_tind_extent_range(handle
, tmp_inode
,
536 le32_to_cpu(i_data
[EXT4_TIND_BLOCK
]), &lb
);
541 * Build the last extent
543 retval
= finish_range(handle
, tmp_inode
, &lb
);
547 * Failure case delete the extent information with the
550 free_ext_block(handle
, tmp_inode
);
552 retval
= ext4_ext_swap_inode_data(handle
, inode
, tmp_inode
);
555 * if we fail to swap inode data free the extent
556 * details of the tmp inode
558 free_ext_block(handle
, tmp_inode
);
561 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
562 retval
= ext4_journal_ensure_credits(handle
, 1, 0);
566 * Mark the tmp_inode as of size zero
568 i_size_write(tmp_inode
, 0);
571 * set the i_blocks count to zero
572 * so that the ext4_evict_inode() does the
575 * We don't need to take the i_lock because
576 * the inode is not visible to user space.
578 tmp_inode
->i_blocks
= 0;
580 /* Reset the extent details */
581 ext4_ext_tree_init(handle
, tmp_inode
);
583 ext4_journal_stop(handle
);
585 unlock_new_inode(tmp_inode
);
588 percpu_up_write(&sbi
->s_writepages_rwsem
);
593 * Migrate a simple extent-based inode to use the i_blocks[] array
595 int ext4_ind_migrate(struct inode
*inode
)
597 struct ext4_extent_header
*eh
;
598 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
599 struct ext4_super_block
*es
= sbi
->s_es
;
600 struct ext4_inode_info
*ei
= EXT4_I(inode
);
601 struct ext4_extent
*ex
;
603 ext4_lblk_t start
, end
;
608 if (!ext4_has_feature_extents(inode
->i_sb
) ||
609 (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
612 if (ext4_has_feature_bigalloc(inode
->i_sb
))
616 * In order to get correct extent info, force all delayed allocation
617 * blocks to be allocated, otherwise delayed allocation blocks may not
618 * be reflected and bypass the checks on extent header.
620 if (test_opt(inode
->i_sb
, DELALLOC
))
621 ext4_alloc_da_blocks(inode
);
623 percpu_down_write(&sbi
->s_writepages_rwsem
);
625 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
626 if (IS_ERR(handle
)) {
627 ret
= PTR_ERR(handle
);
631 down_write(&EXT4_I(inode
)->i_data_sem
);
632 ret
= ext4_ext_check_inode(inode
);
636 eh
= ext_inode_hdr(inode
);
637 ex
= EXT_FIRST_EXTENT(eh
);
638 if (ext4_blocks_count(es
) > EXT4_MAX_BLOCK_FILE_PHYS
||
639 eh
->eh_depth
!= 0 || le16_to_cpu(eh
->eh_entries
) > 1) {
643 if (eh
->eh_entries
== 0)
644 blk
= len
= start
= end
= 0;
646 len
= le16_to_cpu(ex
->ee_len
);
647 blk
= ext4_ext_pblock(ex
);
648 start
= le32_to_cpu(ex
->ee_block
);
649 end
= start
+ len
- 1;
650 if (end
>= EXT4_NDIR_BLOCKS
) {
656 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
657 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
658 for (i
= start
; i
<= end
; i
++)
659 ei
->i_data
[i
] = cpu_to_le32(blk
++);
660 ret2
= ext4_mark_inode_dirty(handle
, inode
);
661 if (unlikely(ret2
&& !ret
))
664 ext4_journal_stop(handle
);
665 up_write(&EXT4_I(inode
)->i_data_sem
);
667 percpu_up_write(&sbi
->s_writepages_rwsem
);