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/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 int ext4_resize_begin(struct super_block
*sb
)
23 if (!capable(CAP_SYS_RESOURCE
))
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
30 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
31 ext4_warning(sb
, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
36 if (test_and_set_bit_lock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
))
42 void ext4_resize_end(struct super_block
*sb
)
44 clear_bit_unlock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
);
45 smp_mb__after_clear_bit();
48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
49 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
51 static int verify_group_input(struct super_block
*sb
,
52 struct ext4_new_group_data
*input
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
55 struct ext4_super_block
*es
= sbi
->s_es
;
56 ext4_fsblk_t start
= ext4_blocks_count(es
);
57 ext4_fsblk_t end
= start
+ input
->blocks_count
;
58 ext4_group_t group
= input
->group
;
59 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
60 unsigned overhead
= ext4_bg_has_super(sb
, group
) ?
61 (1 + ext4_bg_num_gdb(sb
, group
) +
62 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
63 ext4_fsblk_t metaend
= start
+ overhead
;
64 struct buffer_head
*bh
= NULL
;
65 ext4_grpblk_t free_blocks_count
, offset
;
68 input
->free_blocks_count
= free_blocks_count
=
69 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
71 if (test_opt(sb
, DEBUG
))
72 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
73 "(%d free, %u reserved)\n",
74 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
75 "no-super", input
->group
, input
->blocks_count
,
76 free_blocks_count
, input
->reserved_blocks
);
78 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
79 if (group
!= sbi
->s_groups_count
)
80 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
81 input
->group
, sbi
->s_groups_count
);
83 ext4_warning(sb
, "Last group not full");
84 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
85 ext4_warning(sb
, "Reserved blocks too high (%u)",
86 input
->reserved_blocks
);
87 else if (free_blocks_count
< 0)
88 ext4_warning(sb
, "Bad blocks count %u",
90 else if (!(bh
= sb_bread(sb
, end
- 1)))
91 ext4_warning(sb
, "Cannot read last block (%llu)",
93 else if (outside(input
->block_bitmap
, start
, end
))
94 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
95 (unsigned long long)input
->block_bitmap
);
96 else if (outside(input
->inode_bitmap
, start
, end
))
97 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
98 (unsigned long long)input
->inode_bitmap
);
99 else if (outside(input
->inode_table
, start
, end
) ||
100 outside(itend
- 1, start
, end
))
101 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
102 (unsigned long long)input
->inode_table
, itend
- 1);
103 else if (input
->inode_bitmap
== input
->block_bitmap
)
104 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
105 (unsigned long long)input
->block_bitmap
);
106 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
107 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
109 (unsigned long long)input
->block_bitmap
,
110 (unsigned long long)input
->inode_table
, itend
- 1);
111 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
112 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
114 (unsigned long long)input
->inode_bitmap
,
115 (unsigned long long)input
->inode_table
, itend
- 1);
116 else if (inside(input
->block_bitmap
, start
, metaend
))
117 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 (unsigned long long)input
->block_bitmap
,
120 else if (inside(input
->inode_bitmap
, start
, metaend
))
121 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 (unsigned long long)input
->inode_bitmap
,
124 else if (inside(input
->inode_table
, start
, metaend
) ||
125 inside(itend
- 1, start
, metaend
))
126 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
128 (unsigned long long)input
->inode_table
,
129 itend
- 1, start
, metaend
- 1);
138 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
141 struct ext4_new_flex_group_data
{
142 struct ext4_new_group_data
*groups
; /* new_group_data for groups
144 __u16
*bg_flags
; /* block group flags of groups
146 ext4_group_t count
; /* number of groups in @groups
151 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
154 * Returns NULL on failure otherwise address of the allocated structure.
156 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
158 struct ext4_new_flex_group_data
*flex_gd
;
160 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
164 flex_gd
->count
= flexbg_size
;
166 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
167 flexbg_size
, GFP_NOFS
);
168 if (flex_gd
->groups
== NULL
)
171 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
172 if (flex_gd
->bg_flags
== NULL
)
178 kfree(flex_gd
->groups
);
185 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
187 kfree(flex_gd
->bg_flags
);
188 kfree(flex_gd
->groups
);
193 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
194 * and inode tables for a flex group.
196 * This function is used by 64bit-resize. Note that this function allocates
197 * group tables from the 1st group of groups contained by @flexgd, which may
198 * be a partial of a flex group.
200 * @sb: super block of fs to which the groups belongs
202 static void ext4_alloc_group_tables(struct super_block
*sb
,
203 struct ext4_new_flex_group_data
*flex_gd
,
206 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
207 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
208 ext4_fsblk_t start_blk
;
209 ext4_fsblk_t last_blk
;
210 ext4_group_t src_group
;
211 ext4_group_t bb_index
= 0;
212 ext4_group_t ib_index
= 0;
213 ext4_group_t it_index
= 0;
215 ext4_group_t last_group
;
218 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
220 src_group
= group_data
[0].group
;
221 last_group
= src_group
+ flex_gd
->count
- 1;
223 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
224 (last_group
& ~(flexbg_size
- 1))));
226 group
= group_data
[0].group
;
227 start_blk
= ext4_group_first_block_no(sb
, src_group
);
228 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
230 overhead
= ext4_bg_has_super(sb
, src_group
) ?
231 (1 + ext4_bg_num_gdb(sb
, src_group
) +
232 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
234 start_blk
+= overhead
;
236 BUG_ON(src_group
>= group_data
[0].group
+ flex_gd
->count
);
237 /* We collect contiguous blocks as much as possible. */
239 for (; src_group
<= last_group
; src_group
++)
240 if (!ext4_bg_has_super(sb
, src_group
))
241 last_blk
+= group_data
[src_group
- group
].blocks_count
;
245 /* Allocate block bitmaps */
246 for (; bb_index
< flex_gd
->count
; bb_index
++) {
247 if (start_blk
>= last_blk
)
249 group_data
[bb_index
].block_bitmap
= start_blk
++;
250 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
251 group
-= group_data
[0].group
;
252 group_data
[group
].free_blocks_count
--;
254 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
257 /* Allocate inode bitmaps */
258 for (; ib_index
< flex_gd
->count
; ib_index
++) {
259 if (start_blk
>= last_blk
)
261 group_data
[ib_index
].inode_bitmap
= start_blk
++;
262 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
263 group
-= group_data
[0].group
;
264 group_data
[group
].free_blocks_count
--;
266 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
269 /* Allocate inode tables */
270 for (; it_index
< flex_gd
->count
; it_index
++) {
271 if (start_blk
+ EXT4_SB(sb
)->s_itb_per_group
> last_blk
)
273 group_data
[it_index
].inode_table
= start_blk
;
274 ext4_get_group_no_and_offset(sb
, start_blk
, &group
, NULL
);
275 group
-= group_data
[0].group
;
276 group_data
[group
].free_blocks_count
-=
277 EXT4_SB(sb
)->s_itb_per_group
;
279 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
281 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
284 if (test_opt(sb
, DEBUG
)) {
286 group
= group_data
[0].group
;
288 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
289 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
292 for (i
= 0; i
< flex_gd
->count
; i
++) {
293 printk(KERN_DEBUG
"adding %s group %u: %u "
294 "blocks (%d free)\n",
295 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
296 "no-super", group
+ i
,
297 group_data
[i
].blocks_count
,
298 group_data
[i
].free_blocks_count
);
303 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
306 struct buffer_head
*bh
;
309 bh
= sb_getblk(sb
, blk
);
311 return ERR_PTR(-EIO
);
312 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
316 memset(bh
->b_data
, 0, sb
->s_blocksize
);
317 set_buffer_uptodate(bh
);
324 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
325 * If that fails, restart the transaction & regain write access for the
326 * buffer head which is used for block_bitmap modifications.
328 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
332 if (ext4_handle_has_enough_credits(handle
, thresh
))
335 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
339 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
348 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
350 * Helper function for ext4_setup_new_group_blocks() which set .
353 * @handle: journal handle
354 * @flex_gd: flex group data
356 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
357 struct ext4_new_flex_group_data
*flex_gd
,
358 ext4_fsblk_t block
, ext4_group_t count
)
362 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
363 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
365 struct buffer_head
*bh
;
369 ext4_get_group_no_and_offset(sb
, block
, &group
, NULL
);
370 start
= ext4_group_first_block_no(sb
, group
);
371 group
-= flex_gd
->groups
[0].group
;
373 count2
= sb
->s_blocksize
* 8 - (block
- start
);
377 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
378 BUG_ON(flex_gd
->count
> 1);
382 err
= extend_or_restart_transaction(handle
, 1);
386 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
390 err
= ext4_journal_get_write_access(handle
, bh
);
393 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
394 block
- start
, count2
);
395 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
397 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
407 * Set up the block and inode bitmaps, and the inode table for the new groups.
408 * This doesn't need to be part of the main transaction, since we are only
409 * changing blocks outside the actual filesystem. We still do journaling to
410 * ensure the recovery is correct in case of a failure just after resize.
411 * If any part of this fails, we simply abort the resize.
413 * setup_new_flex_group_blocks handles a flex group as follow:
414 * 1. copy super block and GDT, and initialize group tables if necessary.
415 * In this step, we only set bits in blocks bitmaps for blocks taken by
416 * super block and GDT.
417 * 2. allocate group tables in block bitmaps, that is, set bits in block
418 * bitmap for blocks taken by group tables.
420 static int setup_new_flex_group_blocks(struct super_block
*sb
,
421 struct ext4_new_flex_group_data
*flex_gd
)
423 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
426 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
427 struct ext4_super_block
*es
= sbi
->s_es
;
428 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
429 __u16
*bg_flags
= flex_gd
->bg_flags
;
431 ext4_group_t group
, count
;
432 struct buffer_head
*bh
= NULL
;
433 int reserved_gdb
, i
, j
, err
= 0, err2
;
435 BUG_ON(!flex_gd
->count
|| !group_data
||
436 group_data
[0].group
!= sbi
->s_groups_count
);
438 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
440 /* This transaction may be extended/restarted along the way */
441 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
443 return PTR_ERR(handle
);
445 group
= group_data
[0].group
;
446 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
447 unsigned long gdblocks
;
449 gdblocks
= ext4_bg_num_gdb(sb
, group
);
450 start
= ext4_group_first_block_no(sb
, group
);
452 /* Copy all of the GDT blocks into the backup in this group */
453 for (j
= 0, block
= start
+ 1; j
< gdblocks
; j
++, block
++) {
454 struct buffer_head
*gdb
;
456 ext4_debug("update backup group %#04llx\n", block
);
457 err
= extend_or_restart_transaction(handle
, 1);
461 gdb
= sb_getblk(sb
, block
);
467 err
= ext4_journal_get_write_access(handle
, gdb
);
472 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
474 set_buffer_uptodate(gdb
);
476 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
484 /* Zero out all of the reserved backup group descriptor
487 if (ext4_bg_has_super(sb
, group
)) {
488 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
489 reserved_gdb
, GFP_NOFS
);
494 /* Initialize group tables of the grop @group */
495 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
498 /* Zero out all of the inode table blocks */
499 block
= group_data
[i
].inode_table
;
500 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
501 block
, sbi
->s_itb_per_group
);
502 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
508 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
511 /* Initialize block bitmap of the @group */
512 block
= group_data
[i
].block_bitmap
;
513 err
= extend_or_restart_transaction(handle
, 1);
517 bh
= bclean(handle
, sb
, block
);
522 if (ext4_bg_has_super(sb
, group
)) {
523 ext4_debug("mark backup superblock %#04llx (+0)\n",
525 ext4_set_bits(bh
->b_data
, 0, gdblocks
+ reserved_gdb
+
528 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
529 sb
->s_blocksize
* 8, bh
->b_data
);
530 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
536 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
539 /* Initialize inode bitmap of the @group */
540 block
= group_data
[i
].inode_bitmap
;
541 err
= extend_or_restart_transaction(handle
, 1);
544 /* Mark unused entries in inode bitmap used */
545 bh
= bclean(handle
, sb
, block
);
551 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
552 sb
->s_blocksize
* 8, bh
->b_data
);
553 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
560 /* Mark group tables in block bitmap */
561 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
562 count
= group_table_count
[j
];
563 start
= (&group_data
[0].block_bitmap
)[j
];
565 for (i
= 1; i
< flex_gd
->count
; i
++) {
566 block
+= group_table_count
[j
];
567 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
568 count
+= group_table_count
[j
];
571 err
= set_flexbg_block_bitmap(sb
, handle
,
572 flex_gd
, start
, count
);
575 count
= group_table_count
[j
];
576 start
= group_data
[i
].block_bitmap
;
581 err
= set_flexbg_block_bitmap(sb
, handle
,
582 flex_gd
, start
, count
);
590 err2
= ext4_journal_stop(handle
);
598 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
599 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
600 * calling this for the first time. In a sparse filesystem it will be the
601 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
602 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
604 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
605 unsigned *five
, unsigned *seven
)
607 unsigned *min
= three
;
611 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
612 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
634 * Check that all of the backup GDT blocks are held in the primary GDT block.
635 * It is assumed that they are stored in group order. Returns the number of
636 * groups in current filesystem that have BACKUPS, or -ve error code.
638 static int verify_reserved_gdb(struct super_block
*sb
,
640 struct buffer_head
*primary
)
642 const ext4_fsblk_t blk
= primary
->b_blocknr
;
647 __le32
*p
= (__le32
*)primary
->b_data
;
650 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
651 if (le32_to_cpu(*p
++) !=
652 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
653 ext4_warning(sb
, "reserved GDT %llu"
654 " missing grp %d (%llu)",
657 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
661 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
669 * Called when we need to bring a reserved group descriptor table block into
670 * use from the resize inode. The primary copy of the new GDT block currently
671 * is an indirect block (under the double indirect block in the resize inode).
672 * The new backup GDT blocks will be stored as leaf blocks in this indirect
673 * block, in group order. Even though we know all the block numbers we need,
674 * we check to ensure that the resize inode has actually reserved these blocks.
676 * Don't need to update the block bitmaps because the blocks are still in use.
678 * We get all of the error cases out of the way, so that we are sure to not
679 * fail once we start modifying the data on disk, because JBD has no rollback.
681 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
684 struct super_block
*sb
= inode
->i_sb
;
685 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
686 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
687 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
688 struct buffer_head
**o_group_desc
, **n_group_desc
;
689 struct buffer_head
*dind
;
690 struct buffer_head
*gdb_bh
;
692 struct ext4_iloc iloc
;
696 if (test_opt(sb
, DEBUG
))
698 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
702 * If we are not using the primary superblock/GDT copy don't resize,
703 * because the user tools have no way of handling this. Probably a
704 * bad time to do it anyways.
706 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
707 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
708 ext4_warning(sb
, "won't resize using backup superblock at %llu",
709 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
713 gdb_bh
= sb_bread(sb
, gdblock
);
717 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
723 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
724 dind
= sb_bread(sb
, le32_to_cpu(*data
));
730 data
= (__le32
*)dind
->b_data
;
731 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
732 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
738 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
742 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
746 err
= ext4_journal_get_write_access(handle
, dind
);
748 ext4_std_error(sb
, err
);
750 /* ext4_reserve_inode_write() gets a reference on the iloc */
751 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
755 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
756 sizeof(struct buffer_head
*),
760 ext4_warning(sb
, "not enough memory for %lu groups",
766 * Finally, we have all of the possible failures behind us...
768 * Remove new GDT block from inode double-indirect block and clear out
769 * the new GDT block for use (which also "frees" the backup GDT blocks
770 * from the reserved inode). We don't need to change the bitmaps for
771 * these blocks, because they are marked as in-use from being in the
772 * reserved inode, and will become GDT blocks (primary and backup).
774 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
775 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
777 ext4_std_error(sb
, err
);
780 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
781 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
782 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
783 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
785 ext4_std_error(sb
, err
);
790 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
791 memcpy(n_group_desc
, o_group_desc
,
792 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
793 n_group_desc
[gdb_num
] = gdb_bh
;
794 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
795 EXT4_SB(sb
)->s_gdb_count
++;
796 ext4_kvfree(o_group_desc
);
798 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
799 err
= ext4_handle_dirty_super_now(handle
, sb
);
801 ext4_std_error(sb
, err
);
806 ext4_kvfree(n_group_desc
);
807 /* ext4_handle_release_buffer(handle, iloc.bh); */
810 /* ext4_handle_release_buffer(handle, dind); */
812 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
818 ext4_debug("leaving with error %d\n", err
);
823 * Called when we are adding a new group which has a backup copy of each of
824 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
825 * We need to add these reserved backup GDT blocks to the resize inode, so
826 * that they are kept for future resizing and not allocated to files.
828 * Each reserved backup GDT block will go into a different indirect block.
829 * The indirect blocks are actually the primary reserved GDT blocks,
830 * so we know in advance what their block numbers are. We only get the
831 * double-indirect block to verify it is pointing to the primary reserved
832 * GDT blocks so we don't overwrite a data block by accident. The reserved
833 * backup GDT blocks are stored in their reserved primary GDT block.
835 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
838 struct super_block
*sb
= inode
->i_sb
;
839 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
840 struct buffer_head
**primary
;
841 struct buffer_head
*dind
;
842 struct ext4_iloc iloc
;
849 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
853 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
854 dind
= sb_bread(sb
, le32_to_cpu(*data
));
860 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
861 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
862 EXT4_ADDR_PER_BLOCK(sb
));
863 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
865 /* Get each reserved primary GDT block and verify it holds backups */
866 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
867 if (le32_to_cpu(*data
) != blk
) {
868 ext4_warning(sb
, "reserved block %llu"
869 " not at offset %ld",
871 (long)(data
- (__le32
*)dind
->b_data
));
875 primary
[res
] = sb_bread(sb
, blk
);
880 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
882 brelse(primary
[res
]);
887 data
= (__le32
*)dind
->b_data
;
890 for (i
= 0; i
< reserved_gdb
; i
++) {
891 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
894 for (j = 0; j < i; j++)
895 ext4_handle_release_buffer(handle, primary[j]);
901 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
905 * Finally we can add each of the reserved backup GDT blocks from
906 * the new group to its reserved primary GDT block.
908 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
909 for (i
= 0; i
< reserved_gdb
; i
++) {
911 data
= (__le32
*)primary
[i
]->b_data
;
912 /* printk("reserving backup %lu[%u] = %lu\n",
913 primary[i]->b_blocknr, gdbackups,
914 blk + primary[i]->b_blocknr); */
915 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
916 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
920 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
921 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
925 brelse(primary
[res
]);
935 * Update the backup copies of the ext4 metadata. These don't need to be part
936 * of the main resize transaction, because e2fsck will re-write them if there
937 * is a problem (basically only OOM will cause a problem). However, we
938 * _should_ update the backups if possible, in case the primary gets trashed
939 * for some reason and we need to run e2fsck from a backup superblock. The
940 * important part is that the new block and inode counts are in the backup
941 * superblocks, and the location of the new group metadata in the GDT backups.
943 * We do not need take the s_resize_lock for this, because these
944 * blocks are not otherwise touched by the filesystem code when it is
945 * mounted. We don't need to worry about last changing from
946 * sbi->s_groups_count, because the worst that can happen is that we
947 * do not copy the full number of backups at this time. The resize
948 * which changed s_groups_count will backup again.
950 static void update_backups(struct super_block
*sb
,
951 int blk_off
, char *data
, int size
)
953 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
954 const ext4_group_t last
= sbi
->s_groups_count
;
955 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
960 int rest
= sb
->s_blocksize
- size
;
964 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
965 if (IS_ERR(handle
)) {
967 err
= PTR_ERR(handle
);
971 ext4_superblock_csum_set(sb
, (struct ext4_super_block
*)data
);
973 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
974 struct buffer_head
*bh
;
976 /* Out of journal space, and can't get more - abort - so sad */
977 if (ext4_handle_valid(handle
) &&
978 handle
->h_buffer_credits
== 0 &&
979 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
980 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
983 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
988 ext4_debug("update metadata backup %#04lx\n",
989 (unsigned long)bh
->b_blocknr
);
990 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
993 memcpy(bh
->b_data
, data
, size
);
995 memset(bh
->b_data
+ size
, 0, rest
);
996 set_buffer_uptodate(bh
);
998 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1000 ext4_std_error(sb
, err
);
1003 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1007 * Ugh! Need to have e2fsck write the backup copies. It is too
1008 * late to revert the resize, we shouldn't fail just because of
1009 * the backup copies (they are only needed in case of corruption).
1011 * However, if we got here we have a journal problem too, so we
1012 * can't really start a transaction to mark the superblock.
1013 * Chicken out and just set the flag on the hope it will be written
1014 * to disk, and if not - we will simply wait until next fsck.
1018 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1019 "forcing fsck on next reboot", group
, err
);
1020 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1021 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1022 mark_buffer_dirty(sbi
->s_sbh
);
1027 * ext4_add_new_descs() adds @count group descriptor of groups
1028 * starting at @group
1030 * @handle: journal handle
1032 * @group: the group no. of the first group desc to be added
1033 * @resize_inode: the resize inode
1034 * @count: number of group descriptors to be added
1036 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1037 ext4_group_t group
, struct inode
*resize_inode
,
1040 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1041 struct ext4_super_block
*es
= sbi
->s_es
;
1042 struct buffer_head
*gdb_bh
;
1043 int i
, gdb_off
, gdb_num
, err
= 0;
1045 for (i
= 0; i
< count
; i
++, group
++) {
1046 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1047 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1049 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1050 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1053 * We will only either add reserved group blocks to a backup group
1054 * or remove reserved blocks for the first group in a new group block.
1055 * Doing both would be mean more complex code, and sane people don't
1056 * use non-sparse filesystems anymore. This is already checked above.
1059 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1060 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1062 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1063 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1065 err
= add_new_gdb(handle
, resize_inode
, group
);
1072 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1074 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1078 if (bitmap_uptodate(bh
))
1082 if (bh_submit_read(bh
) < 0) {
1092 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1094 struct ext4_group_desc
*gdp
,
1095 struct ext4_new_group_data
*group_data
)
1097 struct buffer_head
*bh
;
1099 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1100 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1103 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1106 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1107 EXT4_INODES_PER_GROUP(sb
) / 8);
1110 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1113 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
,
1114 EXT4_BLOCKS_PER_GROUP(sb
) / 8);
1121 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1123 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1124 struct ext4_new_flex_group_data
*flex_gd
)
1126 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1127 struct ext4_group_desc
*gdp
;
1128 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1129 struct buffer_head
*gdb_bh
;
1131 __u16
*bg_flags
= flex_gd
->bg_flags
;
1132 int i
, gdb_off
, gdb_num
, err
= 0;
1135 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1136 group
= group_data
->group
;
1138 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1139 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1142 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1144 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1145 /* Update group descriptor block for new group */
1146 gdp
= (struct ext4_group_desc
*)((char *)gdb_bh
->b_data
+
1147 gdb_off
* EXT4_DESC_SIZE(sb
));
1149 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1150 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1151 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1152 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1154 ext4_std_error(sb
, err
);
1158 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1159 ext4_free_group_clusters_set(sb
, gdp
,
1160 EXT4_B2C(sbi
, group_data
->free_blocks_count
));
1161 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1162 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1163 ext4_group_desc_csum_set(sb
, group
, gdp
);
1165 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1166 if (unlikely(err
)) {
1167 ext4_std_error(sb
, err
);
1172 * We can allocate memory for mb_alloc based on the new group
1175 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1183 * ext4_update_super() updates the super block so that the newly added
1184 * groups can be seen by the filesystem.
1187 * @flex_gd: new added groups
1189 static void ext4_update_super(struct super_block
*sb
,
1190 struct ext4_new_flex_group_data
*flex_gd
)
1192 ext4_fsblk_t blocks_count
= 0;
1193 ext4_fsblk_t free_blocks
= 0;
1194 ext4_fsblk_t reserved_blocks
= 0;
1195 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1196 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1197 struct ext4_super_block
*es
= sbi
->s_es
;
1200 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1202 * Make the new blocks and inodes valid next. We do this before
1203 * increasing the group count so that once the group is enabled,
1204 * all of its blocks and inodes are already valid.
1206 * We always allocate group-by-group, then block-by-block or
1207 * inode-by-inode within a group, so enabling these
1208 * blocks/inodes before the group is live won't actually let us
1209 * allocate the new space yet.
1211 for (i
= 0; i
< flex_gd
->count
; i
++) {
1212 blocks_count
+= group_data
[i
].blocks_count
;
1213 free_blocks
+= group_data
[i
].free_blocks_count
;
1216 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1217 do_div(reserved_blocks
, ext4_blocks_count(es
));
1218 reserved_blocks
*= blocks_count
;
1219 do_div(reserved_blocks
, 100);
1221 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1222 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1223 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1225 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1229 * We need to protect s_groups_count against other CPUs seeing
1230 * inconsistent state in the superblock.
1232 * The precise rules we use are:
1234 * * Writers must perform a smp_wmb() after updating all
1235 * dependent data and before modifying the groups count
1237 * * Readers must perform an smp_rmb() after reading the groups
1238 * count and before reading any dependent data.
1240 * NB. These rules can be relaxed when checking the group count
1241 * while freeing data, as we can only allocate from a block
1242 * group after serialising against the group count, and we can
1243 * only then free after serialising in turn against that
1248 /* Update the global fs size fields */
1249 sbi
->s_groups_count
+= flex_gd
->count
;
1251 /* Update the reserved block counts only once the new group is
1253 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1256 /* Update the free space counts */
1257 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1258 EXT4_B2C(sbi
, free_blocks
));
1259 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1260 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1262 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
1263 EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1264 sbi
->s_log_groups_per_flex
) {
1265 ext4_group_t flex_group
;
1266 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1267 atomic_add(EXT4_B2C(sbi
, free_blocks
),
1268 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1269 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1270 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1273 if (test_opt(sb
, DEBUG
))
1274 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1275 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1276 blocks_count
, free_blocks
, reserved_blocks
);
1279 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1280 * _before_ we start modifying the filesystem, because we cannot abort the
1281 * transaction and not have it write the data to disk.
1283 static int ext4_flex_group_add(struct super_block
*sb
,
1284 struct inode
*resize_inode
,
1285 struct ext4_new_flex_group_data
*flex_gd
)
1287 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1288 struct ext4_super_block
*es
= sbi
->s_es
;
1289 ext4_fsblk_t o_blocks_count
;
1293 unsigned reserved_gdb
;
1294 int err
= 0, err2
= 0, credit
;
1296 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1298 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1299 o_blocks_count
= ext4_blocks_count(es
);
1300 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1303 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1307 * We will always be modifying at least the superblock and GDT
1308 * block. If we are adding a group past the last current GDT block,
1309 * we will also modify the inode and the dindirect block. If we
1310 * are adding a group with superblock/GDT backups we will also
1311 * modify each of the reserved GDT dindirect blocks.
1313 credit
= flex_gd
->count
* 4 + reserved_gdb
;
1314 handle
= ext4_journal_start_sb(sb
, credit
);
1315 if (IS_ERR(handle
)) {
1316 err
= PTR_ERR(handle
);
1320 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1324 group
= flex_gd
->groups
[0].group
;
1325 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1326 err
= ext4_add_new_descs(handle
, sb
, group
,
1327 resize_inode
, flex_gd
->count
);
1331 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1335 ext4_update_super(sb
, flex_gd
);
1337 err
= ext4_handle_dirty_super(handle
, sb
);
1340 err2
= ext4_journal_stop(handle
);
1346 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1347 sizeof(struct ext4_super_block
));
1348 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
1349 struct buffer_head
*gdb_bh
;
1351 gdb_num
= group
/ EXT4_BLOCKS_PER_GROUP(sb
);
1352 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1353 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1361 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1362 struct ext4_new_flex_group_data
*flex_gd
,
1363 ext4_fsblk_t n_blocks_count
,
1364 unsigned long flexbg_size
)
1366 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1367 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1368 ext4_fsblk_t o_blocks_count
;
1369 ext4_group_t n_group
;
1371 ext4_group_t last_group
;
1373 ext4_grpblk_t blocks_per_group
;
1376 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1378 o_blocks_count
= ext4_blocks_count(es
);
1380 if (o_blocks_count
== n_blocks_count
)
1383 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1385 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1387 last_group
= group
| (flexbg_size
- 1);
1388 if (last_group
> n_group
)
1389 last_group
= n_group
;
1391 flex_gd
->count
= last_group
- group
+ 1;
1393 for (i
= 0; i
< flex_gd
->count
; i
++) {
1396 group_data
[i
].group
= group
+ i
;
1397 group_data
[i
].blocks_count
= blocks_per_group
;
1398 overhead
= ext4_bg_has_super(sb
, group
+ i
) ?
1399 (1 + ext4_bg_num_gdb(sb
, group
+ i
) +
1400 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
1401 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1402 if (ext4_has_group_desc_csum(sb
))
1403 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1404 EXT4_BG_INODE_UNINIT
;
1406 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1409 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1410 /* We need to initialize block bitmap of last group. */
1411 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1413 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1414 group_data
[i
- 1].blocks_count
= last
+ 1;
1415 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1422 /* Add group descriptor data to an existing or new group descriptor block.
1423 * Ensure we handle all possible error conditions _before_ we start modifying
1424 * the filesystem, because we cannot abort the transaction and not have it
1425 * write the data to disk.
1427 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1428 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1430 * We only need to hold the superblock lock while we are actually adding
1431 * in the new group's counts to the superblock. Prior to that we have
1432 * not really "added" the group at all. We re-check that we are still
1433 * adding in the last group in case things have changed since verifying.
1435 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1437 struct ext4_new_flex_group_data flex_gd
;
1438 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1439 struct ext4_super_block
*es
= sbi
->s_es
;
1440 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1441 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1442 struct inode
*inode
= NULL
;
1443 int gdb_off
, gdb_num
;
1447 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
1448 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1450 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1451 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
1452 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1456 if (ext4_blocks_count(es
) + input
->blocks_count
<
1457 ext4_blocks_count(es
)) {
1458 ext4_warning(sb
, "blocks_count overflow");
1462 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1463 le32_to_cpu(es
->s_inodes_count
)) {
1464 ext4_warning(sb
, "inodes_count overflow");
1468 if (reserved_gdb
|| gdb_off
== 0) {
1469 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
1470 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
1471 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1473 "No reserved GDT blocks, can't resize");
1476 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1477 if (IS_ERR(inode
)) {
1478 ext4_warning(sb
, "Error opening resize inode");
1479 return PTR_ERR(inode
);
1484 err
= verify_group_input(sb
, input
);
1489 flex_gd
.groups
= input
;
1490 flex_gd
.bg_flags
= &bg_flags
;
1491 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1495 } /* ext4_group_add */
1498 * extend a group without checking assuming that checking has been done.
1500 static int ext4_group_extend_no_check(struct super_block
*sb
,
1501 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1503 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1507 /* We will update the superblock, one block bitmap, and
1508 * one group descriptor via ext4_group_add_blocks().
1510 handle
= ext4_journal_start_sb(sb
, 3);
1511 if (IS_ERR(handle
)) {
1512 err
= PTR_ERR(handle
);
1513 ext4_warning(sb
, "error %d on journal start", err
);
1517 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1519 ext4_warning(sb
, "error %d on journal write access", err
);
1523 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1524 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1525 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1526 o_blocks_count
+ add
);
1527 /* We add the blocks to the bitmap and set the group need init bit */
1528 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1531 ext4_handle_dirty_super(handle
, sb
);
1532 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1533 o_blocks_count
+ add
);
1535 err2
= ext4_journal_stop(handle
);
1540 if (test_opt(sb
, DEBUG
))
1541 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1542 "blocks\n", ext4_blocks_count(es
));
1543 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1544 sizeof(struct ext4_super_block
));
1550 * Extend the filesystem to the new number of blocks specified. This entry
1551 * point is only used to extend the current filesystem to the end of the last
1552 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1553 * for emergencies (because it has no dependencies on reserved blocks).
1555 * If we _really_ wanted, we could use default values to call ext4_group_add()
1556 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1557 * GDT blocks are reserved to grow to the desired size.
1559 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1560 ext4_fsblk_t n_blocks_count
)
1562 ext4_fsblk_t o_blocks_count
;
1565 struct buffer_head
*bh
;
1569 o_blocks_count
= ext4_blocks_count(es
);
1571 if (test_opt(sb
, DEBUG
))
1572 ext4_msg(sb
, KERN_DEBUG
,
1573 "extending last group from %llu to %llu blocks",
1574 o_blocks_count
, n_blocks_count
);
1576 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1579 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1580 ext4_msg(sb
, KERN_ERR
,
1581 "filesystem too large to resize to %llu blocks safely",
1583 if (sizeof(sector_t
) < 8)
1584 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1588 if (n_blocks_count
< o_blocks_count
) {
1589 ext4_warning(sb
, "can't shrink FS - resize aborted");
1593 /* Handle the remaining blocks in the last group only. */
1594 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1597 ext4_warning(sb
, "need to use ext2online to resize further");
1601 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1603 if (o_blocks_count
+ add
< o_blocks_count
) {
1604 ext4_warning(sb
, "blocks_count overflow");
1608 if (o_blocks_count
+ add
> n_blocks_count
)
1609 add
= n_blocks_count
- o_blocks_count
;
1611 if (o_blocks_count
+ add
< n_blocks_count
)
1612 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1613 o_blocks_count
+ add
, add
);
1615 /* See if the device is actually as big as what was requested */
1616 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1618 ext4_warning(sb
, "can't read last block, resize aborted");
1623 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1625 } /* ext4_group_extend */
1628 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1630 * @sb: super block of the fs to be resized
1631 * @n_blocks_count: the number of blocks resides in the resized fs
1633 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1635 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1636 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1637 struct ext4_super_block
*es
= sbi
->s_es
;
1638 struct buffer_head
*bh
;
1639 struct inode
*resize_inode
;
1640 ext4_fsblk_t o_blocks_count
;
1641 ext4_group_t o_group
;
1642 ext4_group_t n_group
;
1643 ext4_grpblk_t offset
, add
;
1644 unsigned long n_desc_blocks
;
1645 unsigned long o_desc_blocks
;
1646 unsigned long desc_blocks
;
1647 int err
= 0, flexbg_size
= 1;
1649 o_blocks_count
= ext4_blocks_count(es
);
1651 if (test_opt(sb
, DEBUG
))
1652 ext4_msg(sb
, KERN_DEBUG
, "resizing filesystem from %llu "
1653 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1655 if (n_blocks_count
< o_blocks_count
) {
1656 /* On-line shrinking not supported */
1657 ext4_warning(sb
, "can't shrink FS - resize aborted");
1661 if (n_blocks_count
== o_blocks_count
)
1662 /* Nothing need to do */
1665 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &offset
);
1666 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1668 n_desc_blocks
= (n_group
+ EXT4_DESC_PER_BLOCK(sb
)) /
1669 EXT4_DESC_PER_BLOCK(sb
);
1670 o_desc_blocks
= (sbi
->s_groups_count
+ EXT4_DESC_PER_BLOCK(sb
) - 1) /
1671 EXT4_DESC_PER_BLOCK(sb
);
1672 desc_blocks
= n_desc_blocks
- o_desc_blocks
;
1675 (!EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
) ||
1676 le16_to_cpu(es
->s_reserved_gdt_blocks
) < desc_blocks
)) {
1677 ext4_warning(sb
, "No reserved GDT blocks, can't resize");
1681 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1682 if (IS_ERR(resize_inode
)) {
1683 ext4_warning(sb
, "Error opening resize inode");
1684 return PTR_ERR(resize_inode
);
1687 /* See if the device is actually as big as what was requested */
1688 bh
= sb_bread(sb
, n_blocks_count
- 1);
1690 ext4_warning(sb
, "can't read last block, resize aborted");
1695 /* extend the last group */
1696 if (n_group
== o_group
)
1697 add
= n_blocks_count
- o_blocks_count
;
1699 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1701 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1706 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1707 es
->s_log_groups_per_flex
)
1708 flexbg_size
= 1 << es
->s_log_groups_per_flex
;
1710 o_blocks_count
= ext4_blocks_count(es
);
1711 if (o_blocks_count
== n_blocks_count
)
1714 flex_gd
= alloc_flex_gd(flexbg_size
);
1715 if (flex_gd
== NULL
) {
1720 /* Add flex groups. Note that a regular group is a
1721 * flex group with 1 group.
1723 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
1725 ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
);
1726 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
1733 free_flex_gd(flex_gd
);
1736 if (test_opt(sb
, DEBUG
))
1737 ext4_msg(sb
, KERN_DEBUG
, "resized filesystem from %llu "
1738 "upto %llu blocks", o_blocks_count
, n_blocks_count
);