2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
14 #include <linux/ext4_jbd2.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
21 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
22 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
24 static int verify_group_input(struct super_block
*sb
,
25 struct ext4_new_group_data
*input
)
27 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
28 struct ext4_super_block
*es
= sbi
->s_es
;
29 ext4_fsblk_t start
= ext4_blocks_count(es
);
30 ext4_fsblk_t end
= start
+ input
->blocks_count
;
31 ext4_group_t group
= input
->group
;
32 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
33 unsigned overhead
= ext4_bg_has_super(sb
, group
) ?
34 (1 + ext4_bg_num_gdb(sb
, group
) +
35 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
36 ext4_fsblk_t metaend
= start
+ overhead
;
37 struct buffer_head
*bh
= NULL
;
38 ext4_grpblk_t free_blocks_count
, offset
;
41 input
->free_blocks_count
= free_blocks_count
=
42 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
44 if (test_opt(sb
, DEBUG
))
45 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
46 "(%d free, %u reserved)\n",
47 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
48 "no-super", input
->group
, input
->blocks_count
,
49 free_blocks_count
, input
->reserved_blocks
);
51 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
52 if (group
!= sbi
->s_groups_count
)
53 ext4_warning(sb
, __FUNCTION__
,
54 "Cannot add at group %u (only %lu groups)",
55 input
->group
, sbi
->s_groups_count
);
57 ext4_warning(sb
, __FUNCTION__
, "Last group not full");
58 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
59 ext4_warning(sb
, __FUNCTION__
, "Reserved blocks too high (%u)",
60 input
->reserved_blocks
);
61 else if (free_blocks_count
< 0)
62 ext4_warning(sb
, __FUNCTION__
, "Bad blocks count %u",
64 else if (!(bh
= sb_bread(sb
, end
- 1)))
65 ext4_warning(sb
, __FUNCTION__
,
66 "Cannot read last block (%llu)",
68 else if (outside(input
->block_bitmap
, start
, end
))
69 ext4_warning(sb
, __FUNCTION__
,
70 "Block bitmap not in group (block %llu)",
71 (unsigned long long)input
->block_bitmap
);
72 else if (outside(input
->inode_bitmap
, start
, end
))
73 ext4_warning(sb
, __FUNCTION__
,
74 "Inode bitmap not in group (block %llu)",
75 (unsigned long long)input
->inode_bitmap
);
76 else if (outside(input
->inode_table
, start
, end
) ||
77 outside(itend
- 1, start
, end
))
78 ext4_warning(sb
, __FUNCTION__
,
79 "Inode table not in group (blocks %llu-%llu)",
80 (unsigned long long)input
->inode_table
, itend
- 1);
81 else if (input
->inode_bitmap
== input
->block_bitmap
)
82 ext4_warning(sb
, __FUNCTION__
,
83 "Block bitmap same as inode bitmap (%llu)",
84 (unsigned long long)input
->block_bitmap
);
85 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
86 ext4_warning(sb
, __FUNCTION__
,
87 "Block bitmap (%llu) in inode table (%llu-%llu)",
88 (unsigned long long)input
->block_bitmap
,
89 (unsigned long long)input
->inode_table
, itend
- 1);
90 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
91 ext4_warning(sb
, __FUNCTION__
,
92 "Inode bitmap (%llu) in inode table (%llu-%llu)",
93 (unsigned long long)input
->inode_bitmap
,
94 (unsigned long long)input
->inode_table
, itend
- 1);
95 else if (inside(input
->block_bitmap
, start
, metaend
))
96 ext4_warning(sb
, __FUNCTION__
,
97 "Block bitmap (%llu) in GDT table"
99 (unsigned long long)input
->block_bitmap
,
101 else if (inside(input
->inode_bitmap
, start
, metaend
))
102 ext4_warning(sb
, __FUNCTION__
,
103 "Inode bitmap (%llu) in GDT table"
105 (unsigned long long)input
->inode_bitmap
,
107 else if (inside(input
->inode_table
, start
, metaend
) ||
108 inside(itend
- 1, start
, metaend
))
109 ext4_warning(sb
, __FUNCTION__
,
110 "Inode table (%llu-%llu) overlaps"
111 "GDT table (%llu-%llu)",
112 (unsigned long long)input
->inode_table
,
113 itend
- 1, start
, metaend
- 1);
121 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
124 struct buffer_head
*bh
;
127 bh
= sb_getblk(sb
, blk
);
129 return ERR_PTR(-EIO
);
130 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
135 memset(bh
->b_data
, 0, sb
->s_blocksize
);
136 set_buffer_uptodate(bh
);
144 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
145 * If that fails, restart the transaction & regain write access for the
146 * buffer head which is used for block_bitmap modifications.
148 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
,
149 struct buffer_head
*bh
)
153 if (handle
->h_buffer_credits
>= thresh
)
156 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
160 if ((err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
162 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
170 * Set up the block and inode bitmaps, and the inode table for the new group.
171 * This doesn't need to be part of the main transaction, since we are only
172 * changing blocks outside the actual filesystem. We still do journaling to
173 * ensure the recovery is correct in case of a failure just after resize.
174 * If any part of this fails, we simply abort the resize.
176 static int setup_new_group_blocks(struct super_block
*sb
,
177 struct ext4_new_group_data
*input
)
179 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
180 ext4_fsblk_t start
= ext4_group_first_block_no(sb
, input
->group
);
181 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
182 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
) : 0;
183 unsigned long gdblocks
= ext4_bg_num_gdb(sb
, input
->group
);
184 struct buffer_head
*bh
;
191 /* This transaction may be extended/restarted along the way */
192 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
195 return PTR_ERR(handle
);
198 if (input
->group
!= sbi
->s_groups_count
) {
203 if (IS_ERR(bh
= bclean(handle
, sb
, input
->block_bitmap
))) {
208 if (ext4_bg_has_super(sb
, input
->group
)) {
209 ext4_debug("mark backup superblock %#04llx (+0)\n", start
);
210 ext4_set_bit(0, bh
->b_data
);
213 /* Copy all of the GDT blocks into the backup in this group */
214 for (i
= 0, bit
= 1, block
= start
+ 1;
215 i
< gdblocks
; i
++, block
++, bit
++) {
216 struct buffer_head
*gdb
;
218 ext4_debug("update backup group %#04llx (+%d)\n", block
, bit
);
220 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
223 gdb
= sb_getblk(sb
, block
);
228 if ((err
= ext4_journal_get_write_access(handle
, gdb
))) {
233 memcpy(gdb
->b_data
, sbi
->s_group_desc
[i
]->b_data
, gdb
->b_size
);
234 set_buffer_uptodate(gdb
);
236 ext4_journal_dirty_metadata(handle
, gdb
);
237 ext4_set_bit(bit
, bh
->b_data
);
241 /* Zero out all of the reserved backup group descriptor table blocks */
242 for (i
= 0, bit
= gdblocks
+ 1, block
= start
+ bit
;
243 i
< reserved_gdb
; i
++, block
++, bit
++) {
244 struct buffer_head
*gdb
;
246 ext4_debug("clear reserved block %#04llx (+%d)\n", block
, bit
);
248 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
251 if (IS_ERR(gdb
= bclean(handle
, sb
, block
))) {
255 ext4_journal_dirty_metadata(handle
, gdb
);
256 ext4_set_bit(bit
, bh
->b_data
);
259 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input
->block_bitmap
,
260 input
->block_bitmap
- start
);
261 ext4_set_bit(input
->block_bitmap
- start
, bh
->b_data
);
262 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input
->inode_bitmap
,
263 input
->inode_bitmap
- start
);
264 ext4_set_bit(input
->inode_bitmap
- start
, bh
->b_data
);
266 /* Zero out all of the inode table blocks */
267 for (i
= 0, block
= input
->inode_table
, bit
= block
- start
;
268 i
< sbi
->s_itb_per_group
; i
++, bit
++, block
++) {
269 struct buffer_head
*it
;
271 ext4_debug("clear inode block %#04llx (+%d)\n", block
, bit
);
273 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
276 if (IS_ERR(it
= bclean(handle
, sb
, block
))) {
280 ext4_journal_dirty_metadata(handle
, it
);
282 ext4_set_bit(bit
, bh
->b_data
);
285 if ((err
= extend_or_restart_transaction(handle
, 2, bh
)))
288 mark_bitmap_end(input
->blocks_count
, EXT4_BLOCKS_PER_GROUP(sb
),
290 ext4_journal_dirty_metadata(handle
, bh
);
293 /* Mark unused entries in inode bitmap used */
294 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
295 input
->inode_bitmap
, input
->inode_bitmap
- start
);
296 if (IS_ERR(bh
= bclean(handle
, sb
, input
->inode_bitmap
))) {
301 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), EXT4_BLOCKS_PER_GROUP(sb
),
303 ext4_journal_dirty_metadata(handle
, bh
);
309 if ((err2
= ext4_journal_stop(handle
)) && !err
)
316 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
317 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
318 * calling this for the first time. In a sparse filesystem it will be the
319 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
320 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
322 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
323 unsigned *five
, unsigned *seven
)
325 unsigned *min
= three
;
329 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
330 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
352 * Check that all of the backup GDT blocks are held in the primary GDT block.
353 * It is assumed that they are stored in group order. Returns the number of
354 * groups in current filesystem that have BACKUPS, or -ve error code.
356 static int verify_reserved_gdb(struct super_block
*sb
,
357 struct buffer_head
*primary
)
359 const ext4_fsblk_t blk
= primary
->b_blocknr
;
360 const ext4_group_t end
= EXT4_SB(sb
)->s_groups_count
;
365 __le32
*p
= (__le32
*)primary
->b_data
;
368 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
369 if (le32_to_cpu(*p
++) !=
370 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
371 ext4_warning(sb
, __FUNCTION__
,
373 " missing grp %d (%llu)",
376 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
380 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
388 * Called when we need to bring a reserved group descriptor table block into
389 * use from the resize inode. The primary copy of the new GDT block currently
390 * is an indirect block (under the double indirect block in the resize inode).
391 * The new backup GDT blocks will be stored as leaf blocks in this indirect
392 * block, in group order. Even though we know all the block numbers we need,
393 * we check to ensure that the resize inode has actually reserved these blocks.
395 * Don't need to update the block bitmaps because the blocks are still in use.
397 * We get all of the error cases out of the way, so that we are sure to not
398 * fail once we start modifying the data on disk, because JBD has no rollback.
400 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
401 struct ext4_new_group_data
*input
,
402 struct buffer_head
**primary
)
404 struct super_block
*sb
= inode
->i_sb
;
405 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
406 unsigned long gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
407 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
408 struct buffer_head
**o_group_desc
, **n_group_desc
;
409 struct buffer_head
*dind
;
411 struct ext4_iloc iloc
;
415 if (test_opt(sb
, DEBUG
))
417 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
421 * If we are not using the primary superblock/GDT copy don't resize,
422 * because the user tools have no way of handling this. Probably a
423 * bad time to do it anyways.
425 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
426 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
427 ext4_warning(sb
, __FUNCTION__
,
428 "won't resize using backup superblock at %llu",
429 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
433 *primary
= sb_bread(sb
, gdblock
);
437 if ((gdbackups
= verify_reserved_gdb(sb
, *primary
)) < 0) {
442 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
443 dind
= sb_bread(sb
, le32_to_cpu(*data
));
449 data
= (__le32
*)dind
->b_data
;
450 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
451 ext4_warning(sb
, __FUNCTION__
,
452 "new group %u GDT block %llu not reserved",
453 input
->group
, gdblock
);
458 if ((err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
)))
461 if ((err
= ext4_journal_get_write_access(handle
, *primary
)))
464 if ((err
= ext4_journal_get_write_access(handle
, dind
)))
467 /* ext4_reserve_inode_write() gets a reference on the iloc */
468 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
471 n_group_desc
= kmalloc((gdb_num
+ 1) * sizeof(struct buffer_head
*),
475 ext4_warning (sb
, __FUNCTION__
,
476 "not enough memory for %lu groups", gdb_num
+ 1);
481 * Finally, we have all of the possible failures behind us...
483 * Remove new GDT block from inode double-indirect block and clear out
484 * the new GDT block for use (which also "frees" the backup GDT blocks
485 * from the reserved inode). We don't need to change the bitmaps for
486 * these blocks, because they are marked as in-use from being in the
487 * reserved inode, and will become GDT blocks (primary and backup).
489 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
490 ext4_journal_dirty_metadata(handle
, dind
);
492 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
493 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
494 memset((*primary
)->b_data
, 0, sb
->s_blocksize
);
495 ext4_journal_dirty_metadata(handle
, *primary
);
497 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
498 memcpy(n_group_desc
, o_group_desc
,
499 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
500 n_group_desc
[gdb_num
] = *primary
;
501 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
502 EXT4_SB(sb
)->s_gdb_count
++;
505 es
->s_reserved_gdt_blocks
=
506 cpu_to_le16(le16_to_cpu(es
->s_reserved_gdt_blocks
) - 1);
507 ext4_journal_dirty_metadata(handle
, EXT4_SB(sb
)->s_sbh
);
512 //ext4_journal_release_buffer(handle, iloc.bh);
515 //ext4_journal_release_buffer(handle, dind);
517 //ext4_journal_release_buffer(handle, *primary);
519 //ext4_journal_release_buffer(handle, *primary);
525 ext4_debug("leaving with error %d\n", err
);
530 * Called when we are adding a new group which has a backup copy of each of
531 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
532 * We need to add these reserved backup GDT blocks to the resize inode, so
533 * that they are kept for future resizing and not allocated to files.
535 * Each reserved backup GDT block will go into a different indirect block.
536 * The indirect blocks are actually the primary reserved GDT blocks,
537 * so we know in advance what their block numbers are. We only get the
538 * double-indirect block to verify it is pointing to the primary reserved
539 * GDT blocks so we don't overwrite a data block by accident. The reserved
540 * backup GDT blocks are stored in their reserved primary GDT block.
542 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
543 struct ext4_new_group_data
*input
)
545 struct super_block
*sb
= inode
->i_sb
;
546 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
547 struct buffer_head
**primary
;
548 struct buffer_head
*dind
;
549 struct ext4_iloc iloc
;
556 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_KERNEL
);
560 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
561 dind
= sb_bread(sb
, le32_to_cpu(*data
));
567 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
568 data
= (__le32
*)dind
->b_data
+ EXT4_SB(sb
)->s_gdb_count
;
569 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
571 /* Get each reserved primary GDT block and verify it holds backups */
572 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
573 if (le32_to_cpu(*data
) != blk
) {
574 ext4_warning(sb
, __FUNCTION__
,
575 "reserved block %llu"
576 " not at offset %ld",
578 (long)(data
- (__le32
*)dind
->b_data
));
582 primary
[res
] = sb_bread(sb
, blk
);
587 if ((gdbackups
= verify_reserved_gdb(sb
, primary
[res
])) < 0) {
588 brelse(primary
[res
]);
593 data
= (__le32
*)dind
->b_data
;
596 for (i
= 0; i
< reserved_gdb
; i
++) {
597 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
600 for (j = 0; j < i; j++)
601 ext4_journal_release_buffer(handle, primary[j]);
607 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
611 * Finally we can add each of the reserved backup GDT blocks from
612 * the new group to its reserved primary GDT block.
614 blk
= input
->group
* EXT4_BLOCKS_PER_GROUP(sb
);
615 for (i
= 0; i
< reserved_gdb
; i
++) {
617 data
= (__le32
*)primary
[i
]->b_data
;
618 /* printk("reserving backup %lu[%u] = %lu\n",
619 primary[i]->b_blocknr, gdbackups,
620 blk + primary[i]->b_blocknr); */
621 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
622 err2
= ext4_journal_dirty_metadata(handle
, primary
[i
]);
626 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
627 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
631 brelse(primary
[res
]);
641 * Update the backup copies of the ext4 metadata. These don't need to be part
642 * of the main resize transaction, because e2fsck will re-write them if there
643 * is a problem (basically only OOM will cause a problem). However, we
644 * _should_ update the backups if possible, in case the primary gets trashed
645 * for some reason and we need to run e2fsck from a backup superblock. The
646 * important part is that the new block and inode counts are in the backup
647 * superblocks, and the location of the new group metadata in the GDT backups.
649 * We do not need lock_super() for this, because these blocks are not
650 * otherwise touched by the filesystem code when it is mounted. We don't
651 * need to worry about last changing from sbi->s_groups_count, because the
652 * worst that can happen is that we do not copy the full number of backups
653 * at this time. The resize which changed s_groups_count will backup again.
655 static void update_backups(struct super_block
*sb
,
656 int blk_off
, char *data
, int size
)
658 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
659 const ext4_group_t last
= sbi
->s_groups_count
;
660 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
665 int rest
= sb
->s_blocksize
- size
;
669 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
670 if (IS_ERR(handle
)) {
672 err
= PTR_ERR(handle
);
676 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
677 struct buffer_head
*bh
;
679 /* Out of journal space, and can't get more - abort - so sad */
680 if (handle
->h_buffer_credits
== 0 &&
681 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
682 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
685 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
690 ext4_debug("update metadata backup %#04lx\n",
691 (unsigned long)bh
->b_blocknr
);
692 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
695 memcpy(bh
->b_data
, data
, size
);
697 memset(bh
->b_data
+ size
, 0, rest
);
698 set_buffer_uptodate(bh
);
700 ext4_journal_dirty_metadata(handle
, bh
);
703 if ((err2
= ext4_journal_stop(handle
)) && !err
)
707 * Ugh! Need to have e2fsck write the backup copies. It is too
708 * late to revert the resize, we shouldn't fail just because of
709 * the backup copies (they are only needed in case of corruption).
711 * However, if we got here we have a journal problem too, so we
712 * can't really start a transaction to mark the superblock.
713 * Chicken out and just set the flag on the hope it will be written
714 * to disk, and if not - we will simply wait until next fsck.
718 ext4_warning(sb
, __FUNCTION__
,
719 "can't update backup for group %lu (err %d), "
720 "forcing fsck on next reboot", group
, err
);
721 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
722 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
723 mark_buffer_dirty(sbi
->s_sbh
);
727 /* Add group descriptor data to an existing or new group descriptor block.
728 * Ensure we handle all possible error conditions _before_ we start modifying
729 * the filesystem, because we cannot abort the transaction and not have it
730 * write the data to disk.
732 * If we are on a GDT block boundary, we need to get the reserved GDT block.
733 * Otherwise, we may need to add backup GDT blocks for a sparse group.
735 * We only need to hold the superblock lock while we are actually adding
736 * in the new group's counts to the superblock. Prior to that we have
737 * not really "added" the group at all. We re-check that we are still
738 * adding in the last group in case things have changed since verifying.
740 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
742 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
743 struct ext4_super_block
*es
= sbi
->s_es
;
744 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
745 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
746 struct buffer_head
*primary
= NULL
;
747 struct ext4_group_desc
*gdp
;
748 struct inode
*inode
= NULL
;
750 int gdb_off
, gdb_num
;
753 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
754 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
756 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
757 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
758 ext4_warning(sb
, __FUNCTION__
,
759 "Can't resize non-sparse filesystem further");
763 if (ext4_blocks_count(es
) + input
->blocks_count
<
764 ext4_blocks_count(es
)) {
765 ext4_warning(sb
, __FUNCTION__
, "blocks_count overflow\n");
769 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
770 le32_to_cpu(es
->s_inodes_count
)) {
771 ext4_warning(sb
, __FUNCTION__
, "inodes_count overflow\n");
775 if (reserved_gdb
|| gdb_off
== 0) {
776 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
777 EXT4_FEATURE_COMPAT_RESIZE_INODE
)){
778 ext4_warning(sb
, __FUNCTION__
,
779 "No reserved GDT blocks, can't resize");
782 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
784 ext4_warning(sb
, __FUNCTION__
,
785 "Error opening resize inode");
786 return PTR_ERR(inode
);
790 if ((err
= verify_group_input(sb
, input
)))
793 if ((err
= setup_new_group_blocks(sb
, input
)))
797 * We will always be modifying at least the superblock and a GDT
798 * block. If we are adding a group past the last current GDT block,
799 * we will also modify the inode and the dindirect block. If we
800 * are adding a group with superblock/GDT backups we will also
801 * modify each of the reserved GDT dindirect blocks.
803 handle
= ext4_journal_start_sb(sb
,
804 ext4_bg_has_super(sb
, input
->group
) ?
805 3 + reserved_gdb
: 4);
806 if (IS_ERR(handle
)) {
807 err
= PTR_ERR(handle
);
812 if (input
->group
!= sbi
->s_groups_count
) {
813 ext4_warning(sb
, __FUNCTION__
,
814 "multiple resizers run on filesystem!");
819 if ((err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
)))
823 * We will only either add reserved group blocks to a backup group
824 * or remove reserved blocks for the first group in a new group block.
825 * Doing both would be mean more complex code, and sane people don't
826 * use non-sparse filesystems anymore. This is already checked above.
829 primary
= sbi
->s_group_desc
[gdb_num
];
830 if ((err
= ext4_journal_get_write_access(handle
, primary
)))
833 if (reserved_gdb
&& ext4_bg_num_gdb(sb
, input
->group
) &&
834 (err
= reserve_backup_gdb(handle
, inode
, input
)))
836 } else if ((err
= add_new_gdb(handle
, inode
, input
, &primary
)))
840 * OK, now we've set up the new group. Time to make it active.
842 * Current kernels don't lock all allocations via lock_super(),
843 * so we have to be safe wrt. concurrent accesses the group
844 * data. So we need to be careful to set all of the relevant
845 * group descriptor data etc. *before* we enable the group.
847 * The key field here is sbi->s_groups_count: as long as
848 * that retains its old value, nobody is going to access the new
851 * So first we update all the descriptor metadata for the new
852 * group; then we update the total disk blocks count; then we
853 * update the groups count to enable the group; then finally we
854 * update the free space counts so that the system can start
855 * using the new disk blocks.
858 /* Update group descriptor block for new group */
859 gdp
= (struct ext4_group_desc
*)primary
->b_data
+ gdb_off
;
861 ext4_block_bitmap_set(sb
, gdp
, input
->block_bitmap
); /* LV FIXME */
862 ext4_inode_bitmap_set(sb
, gdp
, input
->inode_bitmap
); /* LV FIXME */
863 ext4_inode_table_set(sb
, gdp
, input
->inode_table
); /* LV FIXME */
864 gdp
->bg_free_blocks_count
= cpu_to_le16(input
->free_blocks_count
);
865 gdp
->bg_free_inodes_count
= cpu_to_le16(EXT4_INODES_PER_GROUP(sb
));
866 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, input
->group
, gdp
);
869 * Make the new blocks and inodes valid next. We do this before
870 * increasing the group count so that once the group is enabled,
871 * all of its blocks and inodes are already valid.
873 * We always allocate group-by-group, then block-by-block or
874 * inode-by-inode within a group, so enabling these
875 * blocks/inodes before the group is live won't actually let us
876 * allocate the new space yet.
878 ext4_blocks_count_set(es
, ext4_blocks_count(es
) +
879 input
->blocks_count
);
880 es
->s_inodes_count
= cpu_to_le32(le32_to_cpu(es
->s_inodes_count
) +
881 EXT4_INODES_PER_GROUP(sb
));
884 * We need to protect s_groups_count against other CPUs seeing
885 * inconsistent state in the superblock.
887 * The precise rules we use are:
889 * * Writers of s_groups_count *must* hold lock_super
891 * * Writers must perform a smp_wmb() after updating all dependent
892 * data and before modifying the groups count
894 * * Readers must hold lock_super() over the access
896 * * Readers must perform an smp_rmb() after reading the groups count
897 * and before reading any dependent data.
899 * NB. These rules can be relaxed when checking the group count
900 * while freeing data, as we can only allocate from a block
901 * group after serialising against the group count, and we can
902 * only then free after serialising in turn against that
907 /* Update the global fs size fields */
908 sbi
->s_groups_count
++;
910 ext4_journal_dirty_metadata(handle
, primary
);
912 /* Update the reserved block counts only once the new group is
914 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
915 input
->reserved_blocks
);
917 /* Update the free space counts */
918 percpu_counter_add(&sbi
->s_freeblocks_counter
,
919 input
->free_blocks_count
);
920 percpu_counter_add(&sbi
->s_freeinodes_counter
,
921 EXT4_INODES_PER_GROUP(sb
));
923 ext4_journal_dirty_metadata(handle
, sbi
->s_sbh
);
928 if ((err2
= ext4_journal_stop(handle
)) && !err
)
931 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
932 sizeof(struct ext4_super_block
));
933 update_backups(sb
, primary
->b_blocknr
, primary
->b_data
,
939 } /* ext4_group_add */
941 /* Extend the filesystem to the new number of blocks specified. This entry
942 * point is only used to extend the current filesystem to the end of the last
943 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
944 * for emergencies (because it has no dependencies on reserved blocks).
946 * If we _really_ wanted, we could use default values to call ext4_group_add()
947 * allow the "remount" trick to work for arbitrary resizing, assuming enough
948 * GDT blocks are reserved to grow to the desired size.
950 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
951 ext4_fsblk_t n_blocks_count
)
953 ext4_fsblk_t o_blocks_count
;
954 ext4_group_t o_groups_count
;
957 struct buffer_head
* bh
;
960 unsigned long freed_blocks
;
962 /* We don't need to worry about locking wrt other resizers just
963 * yet: we're going to revalidate es->s_blocks_count after
964 * taking lock_super() below. */
965 o_blocks_count
= ext4_blocks_count(es
);
966 o_groups_count
= EXT4_SB(sb
)->s_groups_count
;
968 if (test_opt(sb
, DEBUG
))
969 printk(KERN_DEBUG
"EXT4-fs: extending last group from %llu uto %llu blocks\n",
970 o_blocks_count
, n_blocks_count
);
972 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
975 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
976 printk(KERN_ERR
"EXT4-fs: filesystem on %s:"
977 " too large to resize to %llu blocks safely\n",
978 sb
->s_id
, n_blocks_count
);
979 if (sizeof(sector_t
) < 8)
980 ext4_warning(sb
, __FUNCTION__
,
981 "CONFIG_LBD not enabled\n");
985 if (n_blocks_count
< o_blocks_count
) {
986 ext4_warning(sb
, __FUNCTION__
,
987 "can't shrink FS - resize aborted");
991 /* Handle the remaining blocks in the last group only. */
992 ext4_get_group_no_and_offset(sb
, o_blocks_count
, NULL
, &last
);
995 ext4_warning(sb
, __FUNCTION__
,
996 "need to use ext2online to resize further");
1000 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1002 if (o_blocks_count
+ add
< o_blocks_count
) {
1003 ext4_warning(sb
, __FUNCTION__
, "blocks_count overflow");
1007 if (o_blocks_count
+ add
> n_blocks_count
)
1008 add
= n_blocks_count
- o_blocks_count
;
1010 if (o_blocks_count
+ add
< n_blocks_count
)
1011 ext4_warning(sb
, __FUNCTION__
,
1012 "will only finish group (%llu"
1014 o_blocks_count
+ add
, add
);
1016 /* See if the device is actually as big as what was requested */
1017 bh
= sb_bread(sb
, o_blocks_count
+ add
-1);
1019 ext4_warning(sb
, __FUNCTION__
,
1020 "can't read last block, resize aborted");
1025 /* We will update the superblock, one block bitmap, and
1026 * one group descriptor via ext4_free_blocks().
1028 handle
= ext4_journal_start_sb(sb
, 3);
1029 if (IS_ERR(handle
)) {
1030 err
= PTR_ERR(handle
);
1031 ext4_warning(sb
, __FUNCTION__
, "error %d on journal start",err
);
1036 if (o_blocks_count
!= ext4_blocks_count(es
)) {
1037 ext4_warning(sb
, __FUNCTION__
,
1038 "multiple resizers run on filesystem!");
1040 ext4_journal_stop(handle
);
1045 if ((err
= ext4_journal_get_write_access(handle
,
1046 EXT4_SB(sb
)->s_sbh
))) {
1047 ext4_warning(sb
, __FUNCTION__
,
1048 "error %d on journal write access", err
);
1050 ext4_journal_stop(handle
);
1053 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1054 ext4_journal_dirty_metadata(handle
, EXT4_SB(sb
)->s_sbh
);
1057 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1058 o_blocks_count
+ add
);
1059 ext4_free_blocks_sb(handle
, sb
, o_blocks_count
, add
, &freed_blocks
);
1060 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1061 o_blocks_count
+ add
);
1062 if ((err
= ext4_journal_stop(handle
)))
1064 if (test_opt(sb
, DEBUG
))
1065 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu blocks\n",
1066 ext4_blocks_count(es
));
1067 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1068 sizeof(struct ext4_super_block
));
1071 } /* ext4_group_extend */