2 * Copyright IBM Corporation, 2007
3 * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of version 2.1 of the GNU Lesser General Public License
7 * as published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
15 #include <linux/slab.h>
16 #include "ext4_jbd2.h"
17 #include "ext4_extents.h"
20 * The contiguous blocks details which can be
21 * represented by a single extent
23 struct migrate_struct
{
24 ext4_lblk_t first_block
, last_block
, curr_block
;
25 ext4_fsblk_t first_pblock
, last_pblock
;
28 static int finish_range(handle_t
*handle
, struct inode
*inode
,
29 struct migrate_struct
*lb
)
32 int retval
= 0, needed
;
33 struct ext4_extent newext
;
34 struct ext4_ext_path
*path
;
35 if (lb
->first_pblock
== 0)
38 /* Add the extent to temp inode*/
39 newext
.ee_block
= cpu_to_le32(lb
->first_block
);
40 newext
.ee_len
= cpu_to_le16(lb
->last_block
- lb
->first_block
+ 1);
41 ext4_ext_store_pblock(&newext
, lb
->first_pblock
);
42 path
= ext4_ext_find_extent(inode
, lb
->first_block
, NULL
, 0);
45 retval
= PTR_ERR(path
);
51 * Calculate the credit needed to inserting this extent
52 * Since we are doing this in loop we may accumalate extra
53 * credit. But below we try to not accumalate too much
54 * of them by restarting the journal.
56 needed
= ext4_ext_calc_credits_for_single_extent(inode
,
57 lb
->last_block
- lb
->first_block
+ 1, path
);
60 * Make sure the credit we accumalated is not really high
62 if (needed
&& ext4_handle_has_enough_credits(handle
,
63 EXT4_RESERVE_TRANS_BLOCKS
)) {
64 retval
= ext4_journal_restart(handle
, needed
);
68 retval
= ext4_journal_extend(handle
, needed
);
71 * IF not able to extend the journal restart the journal
73 retval
= ext4_journal_restart(handle
, needed
);
78 retval
= ext4_ext_insert_extent(handle
, inode
, path
, &newext
, 0);
81 ext4_ext_drop_refs(path
);
88 static int update_extent_range(handle_t
*handle
, struct inode
*inode
,
89 ext4_fsblk_t pblock
, struct migrate_struct
*lb
)
93 * See if we can add on to the existing range (if it exists)
95 if (lb
->first_pblock
&&
96 (lb
->last_pblock
+1 == pblock
) &&
97 (lb
->last_block
+1 == lb
->curr_block
)) {
98 lb
->last_pblock
= pblock
;
99 lb
->last_block
= lb
->curr_block
;
106 retval
= finish_range(handle
, inode
, lb
);
107 lb
->first_pblock
= lb
->last_pblock
= pblock
;
108 lb
->first_block
= lb
->last_block
= lb
->curr_block
;
113 static int update_ind_extent_range(handle_t
*handle
, struct inode
*inode
,
115 struct migrate_struct
*lb
)
117 struct buffer_head
*bh
;
120 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
122 bh
= sb_bread(inode
->i_sb
, pblock
);
126 i_data
= (__le32
*)bh
->b_data
;
127 for (i
= 0; i
< max_entries
; i
++) {
129 retval
= update_extent_range(handle
, inode
,
130 le32_to_cpu(i_data
[i
]), lb
);
142 static int update_dind_extent_range(handle_t
*handle
, struct inode
*inode
,
144 struct migrate_struct
*lb
)
146 struct buffer_head
*bh
;
149 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
151 bh
= sb_bread(inode
->i_sb
, pblock
);
155 i_data
= (__le32
*)bh
->b_data
;
156 for (i
= 0; i
< max_entries
; i
++) {
158 retval
= update_ind_extent_range(handle
, inode
,
159 le32_to_cpu(i_data
[i
]), lb
);
163 /* Only update the file block number */
164 lb
->curr_block
+= max_entries
;
172 static int update_tind_extent_range(handle_t
*handle
, struct inode
*inode
,
174 struct migrate_struct
*lb
)
176 struct buffer_head
*bh
;
179 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
181 bh
= sb_bread(inode
->i_sb
, pblock
);
185 i_data
= (__le32
*)bh
->b_data
;
186 for (i
= 0; i
< max_entries
; i
++) {
188 retval
= update_dind_extent_range(handle
, inode
,
189 le32_to_cpu(i_data
[i
]), lb
);
193 /* Only update the file block number */
194 lb
->curr_block
+= max_entries
* max_entries
;
202 static int extend_credit_for_blkdel(handle_t
*handle
, struct inode
*inode
)
204 int retval
= 0, needed
;
206 if (ext4_handle_has_enough_credits(handle
, EXT4_RESERVE_TRANS_BLOCKS
+1))
209 * We are freeing a blocks. During this we touch
210 * superblock, group descriptor and block bitmap.
211 * So allocate a credit of 3. We may update
212 * quota (user and group).
214 needed
= 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
);
216 if (ext4_journal_extend(handle
, needed
) != 0)
217 retval
= ext4_journal_restart(handle
, needed
);
222 static int free_dind_blocks(handle_t
*handle
,
223 struct inode
*inode
, __le32 i_data
)
227 struct buffer_head
*bh
;
228 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
230 bh
= sb_bread(inode
->i_sb
, le32_to_cpu(i_data
));
234 tmp_idata
= (__le32
*)bh
->b_data
;
235 for (i
= 0; i
< max_entries
; i
++) {
237 extend_credit_for_blkdel(handle
, inode
);
238 ext4_free_blocks(handle
, inode
, NULL
,
239 le32_to_cpu(tmp_idata
[i
]), 1,
240 EXT4_FREE_BLOCKS_METADATA
|
241 EXT4_FREE_BLOCKS_FORGET
);
245 extend_credit_for_blkdel(handle
, inode
);
246 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
247 EXT4_FREE_BLOCKS_METADATA
|
248 EXT4_FREE_BLOCKS_FORGET
);
252 static int free_tind_blocks(handle_t
*handle
,
253 struct inode
*inode
, __le32 i_data
)
257 struct buffer_head
*bh
;
258 unsigned long max_entries
= inode
->i_sb
->s_blocksize
>> 2;
260 bh
= sb_bread(inode
->i_sb
, le32_to_cpu(i_data
));
264 tmp_idata
= (__le32
*)bh
->b_data
;
265 for (i
= 0; i
< max_entries
; i
++) {
267 retval
= free_dind_blocks(handle
,
268 inode
, tmp_idata
[i
]);
276 extend_credit_for_blkdel(handle
, inode
);
277 ext4_free_blocks(handle
, inode
, NULL
, le32_to_cpu(i_data
), 1,
278 EXT4_FREE_BLOCKS_METADATA
|
279 EXT4_FREE_BLOCKS_FORGET
);
283 static int free_ind_block(handle_t
*handle
, struct inode
*inode
, __le32
*i_data
)
287 /* ei->i_data[EXT4_IND_BLOCK] */
289 extend_credit_for_blkdel(handle
, inode
);
290 ext4_free_blocks(handle
, inode
, NULL
,
291 le32_to_cpu(i_data
[0]), 1,
292 EXT4_FREE_BLOCKS_METADATA
|
293 EXT4_FREE_BLOCKS_FORGET
);
296 /* ei->i_data[EXT4_DIND_BLOCK] */
298 retval
= free_dind_blocks(handle
, inode
, i_data
[1]);
303 /* ei->i_data[EXT4_TIND_BLOCK] */
305 retval
= free_tind_blocks(handle
, inode
, i_data
[2]);
312 static int ext4_ext_swap_inode_data(handle_t
*handle
, struct inode
*inode
,
313 struct inode
*tmp_inode
)
317 struct ext4_inode_info
*ei
= EXT4_I(inode
);
318 struct ext4_inode_info
*tmp_ei
= EXT4_I(tmp_inode
);
321 * One credit accounted for writing the
322 * i_data field of the original inode
324 retval
= ext4_journal_extend(handle
, 1);
326 retval
= ext4_journal_restart(handle
, 1);
331 i_data
[0] = ei
->i_data
[EXT4_IND_BLOCK
];
332 i_data
[1] = ei
->i_data
[EXT4_DIND_BLOCK
];
333 i_data
[2] = ei
->i_data
[EXT4_TIND_BLOCK
];
335 down_write(&EXT4_I(inode
)->i_data_sem
);
337 * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
338 * happened after we started the migrate. We need to
341 if (!ext4_test_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
)) {
343 up_write(&EXT4_I(inode
)->i_data_sem
);
346 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
348 * We have the extent map build with the tmp inode.
349 * Now copy the i_data across
351 ext4_set_inode_flag(inode
, EXT4_INODE_EXTENTS
);
352 memcpy(ei
->i_data
, tmp_ei
->i_data
, sizeof(ei
->i_data
));
355 * Update i_blocks with the new blocks that got
356 * allocated while adding extents for extent index
359 * While converting to extents we need not
360 * update the orignal inode i_blocks for extent blocks
361 * via quota APIs. The quota update happened via tmp_inode already.
363 spin_lock(&inode
->i_lock
);
364 inode
->i_blocks
+= tmp_inode
->i_blocks
;
365 spin_unlock(&inode
->i_lock
);
366 up_write(&EXT4_I(inode
)->i_data_sem
);
369 * We mark the inode dirty after, because we decrement the
370 * i_blocks when freeing the indirect meta-data blocks
372 retval
= free_ind_block(handle
, inode
, i_data
);
373 ext4_mark_inode_dirty(handle
, inode
);
379 static int free_ext_idx(handle_t
*handle
, struct inode
*inode
,
380 struct ext4_extent_idx
*ix
)
384 struct buffer_head
*bh
;
385 struct ext4_extent_header
*eh
;
387 block
= ext4_idx_pblock(ix
);
388 bh
= sb_bread(inode
->i_sb
, block
);
392 eh
= (struct ext4_extent_header
*)bh
->b_data
;
393 if (eh
->eh_depth
!= 0) {
394 ix
= EXT_FIRST_INDEX(eh
);
395 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
396 retval
= free_ext_idx(handle
, inode
, ix
);
402 extend_credit_for_blkdel(handle
, inode
);
403 ext4_free_blocks(handle
, inode
, NULL
, block
, 1,
404 EXT4_FREE_BLOCKS_METADATA
| EXT4_FREE_BLOCKS_FORGET
);
409 * Free the extent meta data blocks only
411 static int free_ext_block(handle_t
*handle
, struct inode
*inode
)
414 struct ext4_inode_info
*ei
= EXT4_I(inode
);
415 struct ext4_extent_header
*eh
= (struct ext4_extent_header
*)ei
->i_data
;
416 struct ext4_extent_idx
*ix
;
417 if (eh
->eh_depth
== 0)
419 * No extra blocks allocated for extent meta data
422 ix
= EXT_FIRST_INDEX(eh
);
423 for (i
= 0; i
< le16_to_cpu(eh
->eh_entries
); i
++, ix
++) {
424 retval
= free_ext_idx(handle
, inode
, ix
);
431 int ext4_ext_migrate(struct inode
*inode
)
436 struct ext4_inode_info
*ei
;
437 struct inode
*tmp_inode
= NULL
;
438 struct migrate_struct lb
;
439 unsigned long max_entries
;
444 * If the filesystem does not support extents, or the inode
445 * already is extent-based, error out.
447 if (!EXT4_HAS_INCOMPAT_FEATURE(inode
->i_sb
,
448 EXT4_FEATURE_INCOMPAT_EXTENTS
) ||
449 (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
452 if (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
== 0)
454 * don't migrate fast symlink
459 * Worst case we can touch the allocation bitmaps, a bgd
460 * block, and a block to link in the orphan list. We do need
461 * need to worry about credits for modifying the quota inode.
463 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
,
464 4 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode
->i_sb
));
466 if (IS_ERR(handle
)) {
467 retval
= PTR_ERR(handle
);
470 goal
= (((inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(inode
->i_sb
)) *
471 EXT4_INODES_PER_GROUP(inode
->i_sb
)) + 1;
472 owner
[0] = i_uid_read(inode
);
473 owner
[1] = i_gid_read(inode
);
474 tmp_inode
= ext4_new_inode(handle
, inode
->i_sb
->s_root
->d_inode
,
475 S_IFREG
, NULL
, goal
, owner
);
476 if (IS_ERR(tmp_inode
)) {
477 retval
= PTR_ERR(tmp_inode
);
478 ext4_journal_stop(handle
);
481 i_size_write(tmp_inode
, i_size_read(inode
));
483 * Set the i_nlink to zero so it will be deleted later
484 * when we drop inode reference.
486 clear_nlink(tmp_inode
);
488 ext4_ext_tree_init(handle
, tmp_inode
);
489 ext4_orphan_add(handle
, tmp_inode
);
490 ext4_journal_stop(handle
);
493 * start with one credit accounted for
494 * superblock modification.
496 * For the tmp_inode we already have committed the
497 * transaction that created the inode. Later as and
498 * when we add extents we extent the journal
501 * Even though we take i_mutex we can still cause block
502 * allocation via mmap write to holes. If we have allocated
503 * new blocks we fail migrate. New block allocation will
504 * clear EXT4_STATE_EXT_MIGRATE flag. The flag is updated
505 * with i_data_sem held to prevent racing with block
508 down_read((&EXT4_I(inode
)->i_data_sem
));
509 ext4_set_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
510 up_read((&EXT4_I(inode
)->i_data_sem
));
512 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
513 if (IS_ERR(handle
)) {
515 * It is impossible to update on-disk structures without
516 * a handle, so just rollback in-core changes and live other
517 * work to orphan_list_cleanup()
519 ext4_orphan_del(NULL
, tmp_inode
);
520 retval
= PTR_ERR(handle
);
526 memset(&lb
, 0, sizeof(lb
));
528 /* 32 bit block address 4 bytes */
529 max_entries
= inode
->i_sb
->s_blocksize
>> 2;
530 for (i
= 0; i
< EXT4_NDIR_BLOCKS
; i
++) {
532 retval
= update_extent_range(handle
, tmp_inode
,
533 le32_to_cpu(i_data
[i
]), &lb
);
539 if (i_data
[EXT4_IND_BLOCK
]) {
540 retval
= update_ind_extent_range(handle
, tmp_inode
,
541 le32_to_cpu(i_data
[EXT4_IND_BLOCK
]), &lb
);
545 lb
.curr_block
+= max_entries
;
546 if (i_data
[EXT4_DIND_BLOCK
]) {
547 retval
= update_dind_extent_range(handle
, tmp_inode
,
548 le32_to_cpu(i_data
[EXT4_DIND_BLOCK
]), &lb
);
552 lb
.curr_block
+= max_entries
* max_entries
;
553 if (i_data
[EXT4_TIND_BLOCK
]) {
554 retval
= update_tind_extent_range(handle
, tmp_inode
,
555 le32_to_cpu(i_data
[EXT4_TIND_BLOCK
]), &lb
);
560 * Build the last extent
562 retval
= finish_range(handle
, tmp_inode
, &lb
);
566 * Failure case delete the extent information with the
569 free_ext_block(handle
, tmp_inode
);
571 retval
= ext4_ext_swap_inode_data(handle
, inode
, tmp_inode
);
574 * if we fail to swap inode data free the extent
575 * details of the tmp inode
577 free_ext_block(handle
, tmp_inode
);
580 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
581 if (ext4_journal_extend(handle
, 1) != 0)
582 ext4_journal_restart(handle
, 1);
585 * Mark the tmp_inode as of size zero
587 i_size_write(tmp_inode
, 0);
590 * set the i_blocks count to zero
591 * so that the ext4_delete_inode does the
594 * We don't need to take the i_lock because
595 * the inode is not visible to user space.
597 tmp_inode
->i_blocks
= 0;
599 /* Reset the extent details */
600 ext4_ext_tree_init(handle
, tmp_inode
);
601 ext4_journal_stop(handle
);
603 unlock_new_inode(tmp_inode
);
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_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
616 struct ext4_inode_info
*ei
= EXT4_I(inode
);
617 struct ext4_extent
*ex
;
623 if (!EXT4_HAS_INCOMPAT_FEATURE(inode
->i_sb
,
624 EXT4_FEATURE_INCOMPAT_EXTENTS
) ||
625 (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
628 if (EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
629 EXT4_FEATURE_RO_COMPAT_BIGALLOC
))
632 handle
= ext4_journal_start(inode
, EXT4_HT_MIGRATE
, 1);
634 return PTR_ERR(handle
);
636 down_write(&EXT4_I(inode
)->i_data_sem
);
637 ret
= ext4_ext_check_inode(inode
);
641 eh
= ext_inode_hdr(inode
);
642 ex
= EXT_FIRST_EXTENT(eh
);
643 if (ext4_blocks_count(es
) > EXT4_MAX_BLOCK_FILE_PHYS
||
644 eh
->eh_depth
!= 0 || le16_to_cpu(eh
->eh_entries
) > 1) {
648 if (eh
->eh_entries
== 0)
651 len
= le16_to_cpu(ex
->ee_len
);
652 blk
= ext4_ext_pblock(ex
);
653 if (len
> EXT4_NDIR_BLOCKS
) {
659 ext4_clear_inode_flag(inode
, EXT4_INODE_EXTENTS
);
660 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
661 for (i
=0; i
< len
; i
++)
662 ei
->i_data
[i
] = cpu_to_le32(blk
++);
663 ext4_mark_inode_dirty(handle
, inode
);
665 ext4_journal_stop(handle
);
666 up_write(&EXT4_I(inode
)->i_data_sem
);