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
);
54 * Make sure the credit we accumalated is not really high
56 if (needed
&& ext4_handle_has_enough_credits(handle
,
57 EXT4_RESERVE_TRANS_BLOCKS
)) {
58 up_write((&EXT4_I(inode
)->i_data_sem
));
59 retval
= ext4_journal_restart(handle
, needed
);
60 down_write((&EXT4_I(inode
)->i_data_sem
));
64 retval
= ext4_journal_extend(handle
, needed
);
67 * IF not able to extend the journal restart the journal
69 up_write((&EXT4_I(inode
)->i_data_sem
));
70 retval
= ext4_journal_restart(handle
, needed
);
71 down_write((&EXT4_I(inode
)->i_data_sem
));
76 retval
= ext4_ext_insert_extent(handle
, inode
, &path
, &newext
, 0);
78 up_write((&EXT4_I(inode
)->i_data_sem
));
79 ext4_ext_drop_refs(path
);
85 static int update_extent_range(handle_t
*handle
, struct inode
*inode
,
86 ext4_fsblk_t pblock
, struct migrate_struct
*lb
)
90 * See if we can add on to the existing range (if it exists)
92 if (lb
->first_pblock
&&
93 (lb
->last_pblock
+1 == pblock
) &&
94 (lb
->last_block
+1 == lb
->curr_block
)) {
95 lb
->last_pblock
= pblock
;
96 lb
->last_block
= lb
->curr_block
;
103 retval
= finish_range(handle
, inode
, lb
);
104 lb
->first_pblock
= lb
->last_pblock
= pblock
;
105 lb
->first_block
= lb
->last_block
= lb
->curr_block
;
110 static int update_ind_extent_range(handle_t
*handle
, struct inode
*inode
,
112 struct migrate_struct
*lb
)
114 struct buffer_head
*bh
;
117 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
119 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
123 i_data
= (__le32
*)bh
->b_data
;
124 for (i
= 0; i
< max_entries
; i
++) {
126 retval
= update_extent_range(handle
, inode
,
127 le32_to_cpu(i_data
[i
]), lb
);
139 static int update_dind_extent_range(handle_t
*handle
, struct inode
*inode
,
141 struct migrate_struct
*lb
)
143 struct buffer_head
*bh
;
146 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
148 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
152 i_data
= (__le32
*)bh
->b_data
;
153 for (i
= 0; i
< max_entries
; i
++) {
155 retval
= update_ind_extent_range(handle
, inode
,
156 le32_to_cpu(i_data
[i
]), lb
);
160 /* Only update the file block number */
161 lb
->curr_block
+= max_entries
;
169 static int update_tind_extent_range(handle_t
*handle
, struct inode
*inode
,
171 struct migrate_struct
*lb
)
173 struct buffer_head
*bh
;
176 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
178 bh
= ext4_sb_bread(inode
->i_sb
, pblock
, 0);
182 i_data
= (__le32
*)bh
->b_data
;
183 for (i
= 0; i
< max_entries
; i
++) {
185 retval
= update_dind_extent_range(handle
, inode
,
186 le32_to_cpu(i_data
[i
]), lb
);
190 /* Only update the file block number */
191 lb
->curr_block
+= max_entries
* max_entries
;
199 static int extend_credit_for_blkdel(handle_t
*handle
, struct inode
*inode
)
201 int retval
= 0, needed
;
203 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
206 * We are freeing a blocks. During this we touch
207 * superblock, group descriptor and block bitmap.
208 * So allocate a credit of 3. We may update
209 * quota (user and group).
211 needed
= 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
213 if (ext4_journal_extend(handle
, needed
) != 0)
214 retval
= ext4_journal_restart(handle
, needed
);
219 static int free_dind_blocks(handle_t
*handle
,
220 struct inode
*inode
, __le32 i_data
)
224 struct buffer_head
*bh
;
225 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
227 bh
= ext4_sb_bread(inode
->i_sb
, le32_to_cpu(i_data
), 0);
231 tmp_idata
= (__le32
*)bh
->b_data
;
232 for (i
= 0; i
< max_entries
; i
++) {
234 extend_credit_for_blkdel(handle
, inode
);
235 ext4_free_blocks(handle
, inode
, NULL
,
236 le32_to_cpu(tmp_idata
[i
]), 1,
237 EXT4_FREE_BLOCKS_METADATA
|
238 EXT4_FREE_BLOCKS_FORGET
);
242 extend_credit_for_blkdel(handle
, inode
);
243 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
244 EXT4_FREE_BLOCKS_METADATA
|
245 EXT4_FREE_BLOCKS_FORGET
);
249 static int free_tind_blocks(handle_t
*handle
,
250 struct inode
*inode
, __le32 i_data
)
254 struct buffer_head
*bh
;
255 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
257 bh
= ext4_sb_bread(inode
->i_sb
, le32_to_cpu(i_data
), 0);
261 tmp_idata
= (__le32
*)bh
->b_data
;
262 for (i
= 0; i
< max_entries
; i
++) {
264 retval
= free_dind_blocks(handle
,
265 inode
, tmp_idata
[i
]);
273 extend_credit_for_blkdel(handle
, inode
);
274 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
275 EXT4_FREE_BLOCKS_METADATA
|
276 EXT4_FREE_BLOCKS_FORGET
);
280 static int free_ind_block(handle_t
*handle
, struct inode
*inode
, __le32
*i_data
)
284 /* ei->i_data[EXT4_IND_BLOCK] */
286 extend_credit_for_blkdel(handle
, inode
);
287 ext4_free_blocks(handle
, inode
, NULL
,
288 le32_to_cpu(i_data
[0]), 1,
289 EXT4_FREE_BLOCKS_METADATA
|
290 EXT4_FREE_BLOCKS_FORGET
);
293 /* ei->i_data[EXT4_DIND_BLOCK] */
295 retval
= free_dind_blocks(handle
, inode
, i_data
[1]);
300 /* ei->i_data[EXT4_TIND_BLOCK] */
302 retval
= free_tind_blocks(handle
, inode
, i_data
[2]);
309 static int ext4_ext_swap_inode_data(handle_t
*handle
, struct inode
*inode
,
310 struct inode
*tmp_inode
)
314 struct ext4_inode_info
*ei
= EXT4_I(inode
);
315 struct ext4_inode_info
*tmp_ei
= EXT4_I(tmp_inode
);
318 * One credit accounted for writing the
319 * i_data field of the original inode
321 retval
= ext4_journal_extend(handle
, 1);
323 retval
= ext4_journal_restart(handle
, 1);
328 i_data
[0] = ei
->i_data
[EXT4_IND_BLOCK
];
329 i_data
[1] = ei
->i_data
[EXT4_DIND_BLOCK
];
330 i_data
[2] = ei
->i_data
[EXT4_TIND_BLOCK
];
332 down_write(&EXT4_I(inode
)->i_data_sem
);
334 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
335 * happened after we started the migrate. We need to
338 if (!ext4_test_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
)) {
340 up_write(&EXT4_I(inode
)->i_data_sem
);
343 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
345 * We have the extent map build with the tmp inode.
346 * Now copy the i_data across
348 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
349 memcpy(ei
->i_data
, tmp_ei
->i_data
, sizeof(ei
->i_data
));
352 * Update i_blocks with the new blocks that got
353 * allocated while adding extents for extent index
356 * While converting to extents we need not
357 * update the original inode i_blocks for extent blocks
358 * via quota APIs. The quota update happened via tmp_inode already.
360 spin_lock(&inode
->i_lock
);
361 inode
->i_blocks
+= tmp_inode
->i_blocks
;
362 spin_unlock(&inode
->i_lock
);
363 up_write(&EXT4_I(inode
)->i_data_sem
);
366 * We mark the inode dirty after, because we decrement the
367 * i_blocks when freeing the indirect meta-data blocks
369 retval
= free_ind_block(handle
, inode
, i_data
);
370 ext4_mark_inode_dirty(handle
, inode
);
376 static int free_ext_idx(handle_t
*handle
, struct inode
*inode
,
377 struct ext4_extent_idx
*ix
)
381 struct buffer_head
*bh
;
382 struct ext4_extent_header
*eh
;
384 block
= ext4_idx_pblock(ix
);
385 bh
= ext4_sb_bread(inode
->i_sb
, block
, 0);
389 eh
= (struct ext4_extent_header
*)bh
->b_data
;
390 if (eh
->eh_depth
!= 0) {
391 ix
= EXT_FIRST_INDEX(eh
);
392 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
393 retval
= free_ext_idx(handle
, inode
, ix
);
399 extend_credit_for_blkdel(handle
, inode
);
400 ext4_free_blocks(handle
, inode
, NULL
, block
, 1,
401 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
406 * Free the extent meta data blocks only
408 static int free_ext_block(handle_t
*handle
, struct inode
*inode
)
411 struct ext4_inode_info
*ei
= EXT4_I(inode
);
412 struct ext4_extent_header
*eh
= (struct ext4_extent_header
*)ei
->i_data
;
413 struct ext4_extent_idx
*ix
;
414 if (eh
->eh_depth
== 0)
416 * No extra blocks allocated for extent meta data
419 ix
= EXT_FIRST_INDEX(eh
);
420 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
421 retval
= free_ext_idx(handle
, inode
, ix
);
428 int ext4_ext_migrate(struct inode
*inode
)
430 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
434 struct ext4_inode_info
*ei
;
435 struct inode
*tmp_inode
= NULL
;
436 struct migrate_struct lb
;
437 unsigned long max_entries
;
442 * If the filesystem does not support extents, or the inode
443 * already is extent-based, error out.
445 if (!ext4_has_feature_extents(inode
->i_sb
) ||
446 (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
449 if (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
== 0)
451 * don't migrate fast symlink
455 percpu_down_write(&sbi
->s_writepages_rwsem
);
458 * Worst case we can touch the allocation bitmaps, a bgd
459 * block, and a block to link in the orphan list. We do need
460 * need to worry about credits for modifying the quota inode.
462 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
,
463 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
465 if (IS_ERR(handle
)) {
466 retval
= PTR_ERR(handle
);
469 goal
= (((inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(inode
->i_sb
)) *
470 EXT4_INODES_PER_GROUP(inode
->i_sb
)) + 1;
471 owner
[0] = i_uid_read(inode
);
472 owner
[1] = i_gid_read(inode
);
473 tmp_inode
= ext4_new_inode(handle
, d_inode(inode
->i_sb
->s_root
),
474 S_IFREG
, NULL
, goal
, owner
, 0);
475 if (IS_ERR(tmp_inode
)) {
476 retval
= PTR_ERR(tmp_inode
);
477 ext4_journal_stop(handle
);
480 i_size_write(tmp_inode
, i_size_read(inode
));
482 * Set the i_nlink to zero so it will be deleted later
483 * when we drop inode reference.
485 clear_nlink(tmp_inode
);
487 ext4_ext_tree_init(handle
, tmp_inode
);
488 ext4_orphan_add(handle
, tmp_inode
);
489 ext4_journal_stop(handle
);
492 * start with one credit accounted for
493 * superblock modification.
495 * For the tmp_inode we already have committed the
496 * transaction that created the inode. Later as and
497 * when we add extents we extent the journal
500 * Even though we take i_mutex we can still cause block
501 * allocation via mmap write to holes. If we have allocated
502 * new blocks we fail migrate. New block allocation will
503 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
504 * with i_data_sem held to prevent racing with block
507 down_read(&EXT4_I(inode
)->i_data_sem
);
508 ext4_set_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
509 up_read((&EXT4_I(inode
)->i_data_sem
));
511 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
512 if (IS_ERR(handle
)) {
514 * It is impossible to update on-disk structures without
515 * a handle, so just rollback in-core changes and live other
516 * work to orphan_list_cleanup()
518 ext4_orphan_del(NULL
, tmp_inode
);
519 retval
= PTR_ERR(handle
);
525 memset(&lb
, 0, sizeof(lb
));
527 /* 32 bit block address 4 bytes */
528 max_entries
= inode
->i_sb
->s_blocksize
>> 2;
529 for (i
= 0; i
< EXT4_NDIR_BLOCKS
; i
++) {
531 retval
= update_extent_range(handle
, tmp_inode
,
532 le32_to_cpu(i_data
[i
]), &lb
);
538 if (i_data
[EXT4_IND_BLOCK
]) {
539 retval
= update_ind_extent_range(handle
, tmp_inode
,
540 le32_to_cpu(i_data
[EXT4_IND_BLOCK
]), &lb
);
544 lb
.curr_block
+= max_entries
;
545 if (i_data
[EXT4_DIND_BLOCK
]) {
546 retval
= update_dind_extent_range(handle
, tmp_inode
,
547 le32_to_cpu(i_data
[EXT4_DIND_BLOCK
]), &lb
);
551 lb
.curr_block
+= max_entries
* max_entries
;
552 if (i_data
[EXT4_TIND_BLOCK
]) {
553 retval
= update_tind_extent_range(handle
, tmp_inode
,
554 le32_to_cpu(i_data
[EXT4_TIND_BLOCK
]), &lb
);
559 * Build the last extent
561 retval
= finish_range(handle
, tmp_inode
, &lb
);
565 * Failure case delete the extent information with the
568 free_ext_block(handle
, tmp_inode
);
570 retval
= ext4_ext_swap_inode_data(handle
, inode
, tmp_inode
);
573 * if we fail to swap inode data free the extent
574 * details of the tmp inode
576 free_ext_block(handle
, tmp_inode
);
579 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
580 if (ext4_journal_extend(handle
, 1) != 0)
581 ext4_journal_restart(handle
, 1);
584 * Mark the tmp_inode as of size zero
586 i_size_write(tmp_inode
, 0);
589 * set the i_blocks count to zero
590 * so that the ext4_evict_inode() does the
593 * We don't need to take the i_lock because
594 * the inode is not visible to user space.
596 tmp_inode
->i_blocks
= 0;
598 /* Reset the extent details */
599 ext4_ext_tree_init(handle
, tmp_inode
);
600 ext4_journal_stop(handle
);
602 unlock_new_inode(tmp_inode
);
605 percpu_up_write(&sbi
->s_writepages_rwsem
);
610 * Migrate a simple extent-based inode to use the i_blocks[] array
612 int ext4_ind_migrate(struct inode
*inode
)
614 struct ext4_extent_header
*eh
;
615 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
616 struct ext4_super_block
*es
= sbi
->s_es
;
617 struct ext4_inode_info
*ei
= EXT4_I(inode
);
618 struct ext4_extent
*ex
;
620 ext4_lblk_t start
, end
;
625 if (!ext4_has_feature_extents(inode
->i_sb
) ||
626 (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
629 if (ext4_has_feature_bigalloc(inode
->i_sb
))
633 * In order to get correct extent info, force all delayed allocation
634 * blocks to be allocated, otherwise delayed allocation blocks may not
635 * be reflected and bypass the checks on extent header.
637 if (test_opt(inode
->i_sb
, DELALLOC
))
638 ext4_alloc_da_blocks(inode
);
640 percpu_down_write(&sbi
->s_writepages_rwsem
);
642 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
643 if (IS_ERR(handle
)) {
644 ret
= PTR_ERR(handle
);
648 down_write(&EXT4_I(inode
)->i_data_sem
);
649 ret
= ext4_ext_check_inode(inode
);
653 eh
= ext_inode_hdr(inode
);
654 ex
= EXT_FIRST_EXTENT(eh
);
655 if (ext4_blocks_count(es
) > EXT4_MAX_BLOCK_FILE_PHYS
||
656 eh
->eh_depth
!= 0 || le16_to_cpu(eh
->eh_entries
) > 1) {
660 if (eh
->eh_entries
== 0)
661 blk
= len
= start
= end
= 0;
663 len
= le16_to_cpu(ex
->ee_len
);
664 blk
= ext4_ext_pblock(ex
);
665 start
= le32_to_cpu(ex
->ee_block
);
666 end
= start
+ len
- 1;
667 if (end
>= EXT4_NDIR_BLOCKS
) {
673 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
674 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
675 for (i
= start
; i
<= end
; i
++)
676 ei
->i_data
[i
] = cpu_to_le32(blk
++);
677 ext4_mark_inode_dirty(handle
, inode
);
679 ext4_journal_stop(handle
);
680 up_write(&EXT4_I(inode
)->i_data_sem
);
682 percpu_up_write(&sbi
->s_writepages_rwsem
);