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 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_flex_group_data
))
166 flex_gd
->count
= flexbg_size
;
168 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
169 flexbg_size
, GFP_NOFS
);
170 if (flex_gd
->groups
== NULL
)
173 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
174 if (flex_gd
->bg_flags
== NULL
)
180 kfree(flex_gd
->groups
);
187 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
189 kfree(flex_gd
->bg_flags
);
190 kfree(flex_gd
->groups
);
195 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
196 * and inode tables for a flex group.
198 * This function is used by 64bit-resize. Note that this function allocates
199 * group tables from the 1st group of groups contained by @flexgd, which may
200 * be a partial of a flex group.
202 * @sb: super block of fs to which the groups belongs
204 static void ext4_alloc_group_tables(struct super_block
*sb
,
205 struct ext4_new_flex_group_data
*flex_gd
,
208 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
209 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
210 ext4_fsblk_t start_blk
;
211 ext4_fsblk_t last_blk
;
212 ext4_group_t src_group
;
213 ext4_group_t bb_index
= 0;
214 ext4_group_t ib_index
= 0;
215 ext4_group_t it_index
= 0;
217 ext4_group_t last_group
;
220 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
222 src_group
= group_data
[0].group
;
223 last_group
= src_group
+ flex_gd
->count
- 1;
225 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
226 (last_group
& ~(flexbg_size
- 1))));
228 group
= group_data
[0].group
;
229 start_blk
= ext4_group_first_block_no(sb
, src_group
);
230 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
232 overhead
= ext4_bg_has_super(sb
, src_group
) ?
233 (1 + ext4_bg_num_gdb(sb
, src_group
) +
234 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
236 start_blk
+= overhead
;
238 BUG_ON(src_group
>= group_data
[0].group
+ flex_gd
->count
);
239 /* We collect contiguous blocks as much as possible. */
241 for (; src_group
<= last_group
; src_group
++)
242 if (!ext4_bg_has_super(sb
, src_group
))
243 last_blk
+= group_data
[src_group
- group
].blocks_count
;
247 /* Allocate block bitmaps */
248 for (; bb_index
< flex_gd
->count
; bb_index
++) {
249 if (start_blk
>= last_blk
)
251 group_data
[bb_index
].block_bitmap
= start_blk
++;
252 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
253 group
-= group_data
[0].group
;
254 group_data
[group
].free_blocks_count
--;
256 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
259 /* Allocate inode bitmaps */
260 for (; ib_index
< flex_gd
->count
; ib_index
++) {
261 if (start_blk
>= last_blk
)
263 group_data
[ib_index
].inode_bitmap
= start_blk
++;
264 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
265 group
-= group_data
[0].group
;
266 group_data
[group
].free_blocks_count
--;
268 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
271 /* Allocate inode tables */
272 for (; it_index
< flex_gd
->count
; it_index
++) {
273 if (start_blk
+ EXT4_SB(sb
)->s_itb_per_group
> last_blk
)
275 group_data
[it_index
].inode_table
= start_blk
;
276 ext4_get_group_no_and_offset(sb
, start_blk
, &group
, NULL
);
277 group
-= group_data
[0].group
;
278 group_data
[group
].free_blocks_count
-=
279 EXT4_SB(sb
)->s_itb_per_group
;
281 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
283 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
286 if (test_opt(sb
, DEBUG
)) {
288 group
= group_data
[0].group
;
290 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
291 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
294 for (i
= 0; i
< flex_gd
->count
; i
++) {
295 printk(KERN_DEBUG
"adding %s group %u: %u "
296 "blocks (%d free)\n",
297 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
298 "no-super", group
+ i
,
299 group_data
[i
].blocks_count
,
300 group_data
[i
].free_blocks_count
);
305 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
308 struct buffer_head
*bh
;
311 bh
= sb_getblk(sb
, blk
);
313 return ERR_PTR(-EIO
);
314 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
318 memset(bh
->b_data
, 0, sb
->s_blocksize
);
319 set_buffer_uptodate(bh
);
326 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
327 * If that fails, restart the transaction & regain write access for the
328 * buffer head which is used for block_bitmap modifications.
330 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
334 if (ext4_handle_has_enough_credits(handle
, thresh
))
337 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
341 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
350 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
352 * Helper function for ext4_setup_new_group_blocks() which set .
355 * @handle: journal handle
356 * @flex_gd: flex group data
358 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
359 struct ext4_new_flex_group_data
*flex_gd
,
360 ext4_fsblk_t block
, ext4_group_t count
)
364 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
365 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
367 struct buffer_head
*bh
;
371 ext4_get_group_no_and_offset(sb
, block
, &group
, NULL
);
372 start
= ext4_group_first_block_no(sb
, group
);
373 group
-= flex_gd
->groups
[0].group
;
375 count2
= sb
->s_blocksize
* 8 - (block
- start
);
379 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
380 BUG_ON(flex_gd
->count
> 1);
384 err
= extend_or_restart_transaction(handle
, 1);
388 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
392 err
= ext4_journal_get_write_access(handle
, bh
);
395 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
396 block
- start
, count2
);
397 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
399 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
409 * Set up the block and inode bitmaps, and the inode table for the new groups.
410 * This doesn't need to be part of the main transaction, since we are only
411 * changing blocks outside the actual filesystem. We still do journaling to
412 * ensure the recovery is correct in case of a failure just after resize.
413 * If any part of this fails, we simply abort the resize.
415 * setup_new_flex_group_blocks handles a flex group as follow:
416 * 1. copy super block and GDT, and initialize group tables if necessary.
417 * In this step, we only set bits in blocks bitmaps for blocks taken by
418 * super block and GDT.
419 * 2. allocate group tables in block bitmaps, that is, set bits in block
420 * bitmap for blocks taken by group tables.
422 static int setup_new_flex_group_blocks(struct super_block
*sb
,
423 struct ext4_new_flex_group_data
*flex_gd
)
425 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
428 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
429 struct ext4_super_block
*es
= sbi
->s_es
;
430 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
431 __u16
*bg_flags
= flex_gd
->bg_flags
;
433 ext4_group_t group
, count
;
434 struct buffer_head
*bh
= NULL
;
435 int reserved_gdb
, i
, j
, err
= 0, err2
;
437 BUG_ON(!flex_gd
->count
|| !group_data
||
438 group_data
[0].group
!= sbi
->s_groups_count
);
440 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
442 /* This transaction may be extended/restarted along the way */
443 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
445 return PTR_ERR(handle
);
447 group
= group_data
[0].group
;
448 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
449 unsigned long gdblocks
;
451 gdblocks
= ext4_bg_num_gdb(sb
, group
);
452 start
= ext4_group_first_block_no(sb
, group
);
454 /* Copy all of the GDT blocks into the backup in this group */
455 for (j
= 0, block
= start
+ 1; j
< gdblocks
; j
++, block
++) {
456 struct buffer_head
*gdb
;
458 ext4_debug("update backup group %#04llx\n", block
);
459 err
= extend_or_restart_transaction(handle
, 1);
463 gdb
= sb_getblk(sb
, block
);
469 err
= ext4_journal_get_write_access(handle
, gdb
);
474 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
476 set_buffer_uptodate(gdb
);
478 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
486 /* Zero out all of the reserved backup group descriptor
489 if (ext4_bg_has_super(sb
, group
)) {
490 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
491 reserved_gdb
, GFP_NOFS
);
496 /* Initialize group tables of the grop @group */
497 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
500 /* Zero out all of the inode table blocks */
501 block
= group_data
[i
].inode_table
;
502 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
503 block
, sbi
->s_itb_per_group
);
504 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
510 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
513 /* Initialize block bitmap of the @group */
514 block
= group_data
[i
].block_bitmap
;
515 err
= extend_or_restart_transaction(handle
, 1);
519 bh
= bclean(handle
, sb
, block
);
524 if (ext4_bg_has_super(sb
, group
)) {
525 ext4_debug("mark backup superblock %#04llx (+0)\n",
527 ext4_set_bits(bh
->b_data
, 0, gdblocks
+ reserved_gdb
+
530 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
531 sb
->s_blocksize
* 8, bh
->b_data
);
532 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
538 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
541 /* Initialize inode bitmap of the @group */
542 block
= group_data
[i
].inode_bitmap
;
543 err
= extend_or_restart_transaction(handle
, 1);
546 /* Mark unused entries in inode bitmap used */
547 bh
= bclean(handle
, sb
, block
);
553 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
554 sb
->s_blocksize
* 8, bh
->b_data
);
555 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
562 /* Mark group tables in block bitmap */
563 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
564 count
= group_table_count
[j
];
565 start
= (&group_data
[0].block_bitmap
)[j
];
567 for (i
= 1; i
< flex_gd
->count
; i
++) {
568 block
+= group_table_count
[j
];
569 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
570 count
+= group_table_count
[j
];
573 err
= set_flexbg_block_bitmap(sb
, handle
,
574 flex_gd
, start
, count
);
577 count
= group_table_count
[j
];
578 start
= group_data
[i
].block_bitmap
;
583 err
= set_flexbg_block_bitmap(sb
, handle
,
584 flex_gd
, start
, count
);
592 err2
= ext4_journal_stop(handle
);
600 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
601 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
602 * calling this for the first time. In a sparse filesystem it will be the
603 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
604 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
606 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
607 unsigned *five
, unsigned *seven
)
609 unsigned *min
= three
;
613 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
614 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
636 * Check that all of the backup GDT blocks are held in the primary GDT block.
637 * It is assumed that they are stored in group order. Returns the number of
638 * groups in current filesystem that have BACKUPS, or -ve error code.
640 static int verify_reserved_gdb(struct super_block
*sb
,
642 struct buffer_head
*primary
)
644 const ext4_fsblk_t blk
= primary
->b_blocknr
;
649 __le32
*p
= (__le32
*)primary
->b_data
;
652 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
653 if (le32_to_cpu(*p
++) !=
654 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
655 ext4_warning(sb
, "reserved GDT %llu"
656 " missing grp %d (%llu)",
659 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
663 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
671 * Called when we need to bring a reserved group descriptor table block into
672 * use from the resize inode. The primary copy of the new GDT block currently
673 * is an indirect block (under the double indirect block in the resize inode).
674 * The new backup GDT blocks will be stored as leaf blocks in this indirect
675 * block, in group order. Even though we know all the block numbers we need,
676 * we check to ensure that the resize inode has actually reserved these blocks.
678 * Don't need to update the block bitmaps because the blocks are still in use.
680 * We get all of the error cases out of the way, so that we are sure to not
681 * fail once we start modifying the data on disk, because JBD has no rollback.
683 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
686 struct super_block
*sb
= inode
->i_sb
;
687 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
688 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
689 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
690 struct buffer_head
**o_group_desc
, **n_group_desc
;
691 struct buffer_head
*dind
;
692 struct buffer_head
*gdb_bh
;
694 struct ext4_iloc iloc
;
698 if (test_opt(sb
, DEBUG
))
700 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
704 * If we are not using the primary superblock/GDT copy don't resize,
705 * because the user tools have no way of handling this. Probably a
706 * bad time to do it anyways.
708 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
709 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
710 ext4_warning(sb
, "won't resize using backup superblock at %llu",
711 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
715 gdb_bh
= sb_bread(sb
, gdblock
);
719 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
725 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
726 dind
= sb_bread(sb
, le32_to_cpu(*data
));
732 data
= (__le32
*)dind
->b_data
;
733 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
734 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
740 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
744 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
748 err
= ext4_journal_get_write_access(handle
, dind
);
750 ext4_std_error(sb
, err
);
752 /* ext4_reserve_inode_write() gets a reference on the iloc */
753 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
757 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
758 sizeof(struct buffer_head
*),
762 ext4_warning(sb
, "not enough memory for %lu groups",
768 * Finally, we have all of the possible failures behind us...
770 * Remove new GDT block from inode double-indirect block and clear out
771 * the new GDT block for use (which also "frees" the backup GDT blocks
772 * from the reserved inode). We don't need to change the bitmaps for
773 * these blocks, because they are marked as in-use from being in the
774 * reserved inode, and will become GDT blocks (primary and backup).
776 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
777 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
779 ext4_std_error(sb
, err
);
782 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
783 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
784 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
785 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
787 ext4_std_error(sb
, err
);
792 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
793 memcpy(n_group_desc
, o_group_desc
,
794 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
795 n_group_desc
[gdb_num
] = gdb_bh
;
796 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
797 EXT4_SB(sb
)->s_gdb_count
++;
798 ext4_kvfree(o_group_desc
);
800 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
801 err
= ext4_handle_dirty_super_now(handle
, sb
);
803 ext4_std_error(sb
, err
);
808 ext4_kvfree(n_group_desc
);
809 /* ext4_handle_release_buffer(handle, iloc.bh); */
812 /* ext4_handle_release_buffer(handle, dind); */
814 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
820 ext4_debug("leaving with error %d\n", err
);
825 * Called when we are adding a new group which has a backup copy of each of
826 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
827 * We need to add these reserved backup GDT blocks to the resize inode, so
828 * that they are kept for future resizing and not allocated to files.
830 * Each reserved backup GDT block will go into a different indirect block.
831 * The indirect blocks are actually the primary reserved GDT blocks,
832 * so we know in advance what their block numbers are. We only get the
833 * double-indirect block to verify it is pointing to the primary reserved
834 * GDT blocks so we don't overwrite a data block by accident. The reserved
835 * backup GDT blocks are stored in their reserved primary GDT block.
837 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
840 struct super_block
*sb
= inode
->i_sb
;
841 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
842 struct buffer_head
**primary
;
843 struct buffer_head
*dind
;
844 struct ext4_iloc iloc
;
851 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
855 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
856 dind
= sb_bread(sb
, le32_to_cpu(*data
));
862 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
863 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
864 EXT4_ADDR_PER_BLOCK(sb
));
865 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
867 /* Get each reserved primary GDT block and verify it holds backups */
868 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
869 if (le32_to_cpu(*data
) != blk
) {
870 ext4_warning(sb
, "reserved block %llu"
871 " not at offset %ld",
873 (long)(data
- (__le32
*)dind
->b_data
));
877 primary
[res
] = sb_bread(sb
, blk
);
882 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
884 brelse(primary
[res
]);
889 data
= (__le32
*)dind
->b_data
;
892 for (i
= 0; i
< reserved_gdb
; i
++) {
893 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
896 for (j = 0; j < i; j++)
897 ext4_handle_release_buffer(handle, primary[j]);
903 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
907 * Finally we can add each of the reserved backup GDT blocks from
908 * the new group to its reserved primary GDT block.
910 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
911 for (i
= 0; i
< reserved_gdb
; i
++) {
913 data
= (__le32
*)primary
[i
]->b_data
;
914 /* printk("reserving backup %lu[%u] = %lu\n",
915 primary[i]->b_blocknr, gdbackups,
916 blk + primary[i]->b_blocknr); */
917 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
918 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
922 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
923 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
927 brelse(primary
[res
]);
937 * Update the backup copies of the ext4 metadata. These don't need to be part
938 * of the main resize transaction, because e2fsck will re-write them if there
939 * is a problem (basically only OOM will cause a problem). However, we
940 * _should_ update the backups if possible, in case the primary gets trashed
941 * for some reason and we need to run e2fsck from a backup superblock. The
942 * important part is that the new block and inode counts are in the backup
943 * superblocks, and the location of the new group metadata in the GDT backups.
945 * We do not need take the s_resize_lock for this, because these
946 * blocks are not otherwise touched by the filesystem code when it is
947 * mounted. We don't need to worry about last changing from
948 * sbi->s_groups_count, because the worst that can happen is that we
949 * do not copy the full number of backups at this time. The resize
950 * which changed s_groups_count will backup again.
952 static void update_backups(struct super_block
*sb
,
953 int blk_off
, char *data
, int size
)
955 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
956 const ext4_group_t last
= sbi
->s_groups_count
;
957 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
962 int rest
= sb
->s_blocksize
- size
;
966 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
967 if (IS_ERR(handle
)) {
969 err
= PTR_ERR(handle
);
973 ext4_superblock_csum_set(sb
, (struct ext4_super_block
*)data
);
975 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
976 struct buffer_head
*bh
;
978 /* Out of journal space, and can't get more - abort - so sad */
979 if (ext4_handle_valid(handle
) &&
980 handle
->h_buffer_credits
== 0 &&
981 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
982 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
985 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
990 ext4_debug("update metadata backup %#04lx\n",
991 (unsigned long)bh
->b_blocknr
);
992 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
995 memcpy(bh
->b_data
, data
, size
);
997 memset(bh
->b_data
+ size
, 0, rest
);
998 set_buffer_uptodate(bh
);
1000 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1002 ext4_std_error(sb
, err
);
1005 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1009 * Ugh! Need to have e2fsck write the backup copies. It is too
1010 * late to revert the resize, we shouldn't fail just because of
1011 * the backup copies (they are only needed in case of corruption).
1013 * However, if we got here we have a journal problem too, so we
1014 * can't really start a transaction to mark the superblock.
1015 * Chicken out and just set the flag on the hope it will be written
1016 * to disk, and if not - we will simply wait until next fsck.
1020 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1021 "forcing fsck on next reboot", group
, err
);
1022 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1023 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1024 mark_buffer_dirty(sbi
->s_sbh
);
1029 * ext4_add_new_descs() adds @count group descriptor of groups
1030 * starting at @group
1032 * @handle: journal handle
1034 * @group: the group no. of the first group desc to be added
1035 * @resize_inode: the resize inode
1036 * @count: number of group descriptors to be added
1038 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1039 ext4_group_t group
, struct inode
*resize_inode
,
1042 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1043 struct ext4_super_block
*es
= sbi
->s_es
;
1044 struct buffer_head
*gdb_bh
;
1045 int i
, gdb_off
, gdb_num
, err
= 0;
1047 for (i
= 0; i
< count
; i
++, group
++) {
1048 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1049 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1051 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1052 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1055 * We will only either add reserved group blocks to a backup group
1056 * or remove reserved blocks for the first group in a new group block.
1057 * Doing both would be mean more complex code, and sane people don't
1058 * use non-sparse filesystems anymore. This is already checked above.
1061 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1062 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1064 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1065 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1067 err
= add_new_gdb(handle
, resize_inode
, group
);
1074 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1076 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1080 if (bitmap_uptodate(bh
))
1084 if (bh_submit_read(bh
) < 0) {
1094 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1096 struct ext4_group_desc
*gdp
,
1097 struct ext4_new_group_data
*group_data
)
1099 struct buffer_head
*bh
;
1101 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1102 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1105 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1108 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1109 EXT4_INODES_PER_GROUP(sb
) / 8);
1112 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1115 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
,
1116 EXT4_BLOCKS_PER_GROUP(sb
) / 8);
1123 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1125 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1126 struct ext4_new_flex_group_data
*flex_gd
)
1128 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1129 struct ext4_group_desc
*gdp
;
1130 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1131 struct buffer_head
*gdb_bh
;
1133 __u16
*bg_flags
= flex_gd
->bg_flags
;
1134 int i
, gdb_off
, gdb_num
, err
= 0;
1137 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1138 group
= group_data
->group
;
1140 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1141 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1144 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1146 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1147 /* Update group descriptor block for new group */
1148 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1149 gdb_off
* EXT4_DESC_SIZE(sb
));
1151 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1152 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1153 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1154 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1156 ext4_std_error(sb
, err
);
1160 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1161 ext4_free_group_clusters_set(sb
, gdp
,
1162 EXT4_B2C(sbi
, group_data
->free_blocks_count
));
1163 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1164 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1165 ext4_group_desc_csum_set(sb
, group
, gdp
);
1167 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1168 if (unlikely(err
)) {
1169 ext4_std_error(sb
, err
);
1174 * We can allocate memory for mb_alloc based on the new group
1177 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1185 * ext4_update_super() updates the super block so that the newly added
1186 * groups can be seen by the filesystem.
1189 * @flex_gd: new added groups
1191 static void ext4_update_super(struct super_block
*sb
,
1192 struct ext4_new_flex_group_data
*flex_gd
)
1194 ext4_fsblk_t blocks_count
= 0;
1195 ext4_fsblk_t free_blocks
= 0;
1196 ext4_fsblk_t reserved_blocks
= 0;
1197 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1198 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1199 struct ext4_super_block
*es
= sbi
->s_es
;
1202 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1204 * Make the new blocks and inodes valid next. We do this before
1205 * increasing the group count so that once the group is enabled,
1206 * all of its blocks and inodes are already valid.
1208 * We always allocate group-by-group, then block-by-block or
1209 * inode-by-inode within a group, so enabling these
1210 * blocks/inodes before the group is live won't actually let us
1211 * allocate the new space yet.
1213 for (i
= 0; i
< flex_gd
->count
; i
++) {
1214 blocks_count
+= group_data
[i
].blocks_count
;
1215 free_blocks
+= group_data
[i
].free_blocks_count
;
1218 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1219 do_div(reserved_blocks
, ext4_blocks_count(es
));
1220 reserved_blocks
*= blocks_count
;
1221 do_div(reserved_blocks
, 100);
1223 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1224 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1225 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1227 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1231 * We need to protect s_groups_count against other CPUs seeing
1232 * inconsistent state in the superblock.
1234 * The precise rules we use are:
1236 * * Writers must perform a smp_wmb() after updating all
1237 * dependent data and before modifying the groups count
1239 * * Readers must perform an smp_rmb() after reading the groups
1240 * count and before reading any dependent data.
1242 * NB. These rules can be relaxed when checking the group count
1243 * while freeing data, as we can only allocate from a block
1244 * group after serialising against the group count, and we can
1245 * only then free after serialising in turn against that
1250 /* Update the global fs size fields */
1251 sbi
->s_groups_count
+= flex_gd
->count
;
1253 /* Update the reserved block counts only once the new group is
1255 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1258 /* Update the free space counts */
1259 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1260 EXT4_B2C(sbi
, free_blocks
));
1261 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1262 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1264 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
1265 EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1266 sbi
->s_log_groups_per_flex
) {
1267 ext4_group_t flex_group
;
1268 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1269 atomic_add(EXT4_B2C(sbi
, free_blocks
),
1270 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1271 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1272 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1275 if (test_opt(sb
, DEBUG
))
1276 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1277 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1278 blocks_count
, free_blocks
, reserved_blocks
);
1281 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1282 * _before_ we start modifying the filesystem, because we cannot abort the
1283 * transaction and not have it write the data to disk.
1285 static int ext4_flex_group_add(struct super_block
*sb
,
1286 struct inode
*resize_inode
,
1287 struct ext4_new_flex_group_data
*flex_gd
)
1289 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1290 struct ext4_super_block
*es
= sbi
->s_es
;
1291 ext4_fsblk_t o_blocks_count
;
1295 unsigned reserved_gdb
;
1296 int err
= 0, err2
= 0, credit
;
1298 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1300 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1301 o_blocks_count
= ext4_blocks_count(es
);
1302 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1305 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1309 * We will always be modifying at least the superblock and GDT
1310 * block. If we are adding a group past the last current GDT block,
1311 * we will also modify the inode and the dindirect block. If we
1312 * are adding a group with superblock/GDT backups we will also
1313 * modify each of the reserved GDT dindirect blocks.
1315 credit
= flex_gd
->count
* 4 + reserved_gdb
;
1316 handle
= ext4_journal_start_sb(sb
, credit
);
1317 if (IS_ERR(handle
)) {
1318 err
= PTR_ERR(handle
);
1322 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1326 group
= flex_gd
->groups
[0].group
;
1327 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1328 err
= ext4_add_new_descs(handle
, sb
, group
,
1329 resize_inode
, flex_gd
->count
);
1333 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1337 ext4_update_super(sb
, flex_gd
);
1339 err
= ext4_handle_dirty_super(handle
, sb
);
1342 err2
= ext4_journal_stop(handle
);
1348 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1349 sizeof(struct ext4_super_block
));
1350 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
1351 struct buffer_head
*gdb_bh
;
1353 gdb_num
= group
/ EXT4_BLOCKS_PER_GROUP(sb
);
1354 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1355 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1363 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1364 struct ext4_new_flex_group_data
*flex_gd
,
1365 ext4_fsblk_t n_blocks_count
,
1366 unsigned long flexbg_size
)
1368 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1369 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1370 ext4_fsblk_t o_blocks_count
;
1371 ext4_group_t n_group
;
1373 ext4_group_t last_group
;
1375 ext4_grpblk_t blocks_per_group
;
1378 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1380 o_blocks_count
= ext4_blocks_count(es
);
1382 if (o_blocks_count
== n_blocks_count
)
1385 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1387 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1389 last_group
= group
| (flexbg_size
- 1);
1390 if (last_group
> n_group
)
1391 last_group
= n_group
;
1393 flex_gd
->count
= last_group
- group
+ 1;
1395 for (i
= 0; i
< flex_gd
->count
; i
++) {
1398 group_data
[i
].group
= group
+ i
;
1399 group_data
[i
].blocks_count
= blocks_per_group
;
1400 overhead
= ext4_bg_has_super(sb
, group
+ i
) ?
1401 (1 + ext4_bg_num_gdb(sb
, group
+ i
) +
1402 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
1403 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1404 if (ext4_has_group_desc_csum(sb
))
1405 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1406 EXT4_BG_INODE_UNINIT
;
1408 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1411 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1412 /* We need to initialize block bitmap of last group. */
1413 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1415 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1416 group_data
[i
- 1].blocks_count
= last
+ 1;
1417 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1424 /* Add group descriptor data to an existing or new group descriptor block.
1425 * Ensure we handle all possible error conditions _before_ we start modifying
1426 * the filesystem, because we cannot abort the transaction and not have it
1427 * write the data to disk.
1429 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1430 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1432 * We only need to hold the superblock lock while we are actually adding
1433 * in the new group's counts to the superblock. Prior to that we have
1434 * not really "added" the group at all. We re-check that we are still
1435 * adding in the last group in case things have changed since verifying.
1437 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1439 struct ext4_new_flex_group_data flex_gd
;
1440 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1441 struct ext4_super_block
*es
= sbi
->s_es
;
1442 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1443 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1444 struct inode
*inode
= NULL
;
1445 int gdb_off
, gdb_num
;
1449 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
1450 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1452 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1453 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
1454 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1458 if (ext4_blocks_count(es
) + input
->blocks_count
<
1459 ext4_blocks_count(es
)) {
1460 ext4_warning(sb
, "blocks_count overflow");
1464 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1465 le32_to_cpu(es
->s_inodes_count
)) {
1466 ext4_warning(sb
, "inodes_count overflow");
1470 if (reserved_gdb
|| gdb_off
== 0) {
1471 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
1472 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
1473 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1475 "No reserved GDT blocks, can't resize");
1478 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1479 if (IS_ERR(inode
)) {
1480 ext4_warning(sb
, "Error opening resize inode");
1481 return PTR_ERR(inode
);
1486 err
= verify_group_input(sb
, input
);
1491 flex_gd
.groups
= input
;
1492 flex_gd
.bg_flags
= &bg_flags
;
1493 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1497 } /* ext4_group_add */
1500 * extend a group without checking assuming that checking has been done.
1502 static int ext4_group_extend_no_check(struct super_block
*sb
,
1503 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1505 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1509 /* We will update the superblock, one block bitmap, and
1510 * one group descriptor via ext4_group_add_blocks().
1512 handle
= ext4_journal_start_sb(sb
, 3);
1513 if (IS_ERR(handle
)) {
1514 err
= PTR_ERR(handle
);
1515 ext4_warning(sb
, "error %d on journal start", err
);
1519 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1521 ext4_warning(sb
, "error %d on journal write access", err
);
1525 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1526 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1527 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1528 o_blocks_count
+ add
);
1529 /* We add the blocks to the bitmap and set the group need init bit */
1530 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1533 ext4_handle_dirty_super(handle
, sb
);
1534 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1535 o_blocks_count
+ add
);
1537 err2
= ext4_journal_stop(handle
);
1542 if (test_opt(sb
, DEBUG
))
1543 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1544 "blocks\n", ext4_blocks_count(es
));
1545 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1546 sizeof(struct ext4_super_block
));
1552 * Extend the filesystem to the new number of blocks specified. This entry
1553 * point is only used to extend the current filesystem to the end of the last
1554 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1555 * for emergencies (because it has no dependencies on reserved blocks).
1557 * If we _really_ wanted, we could use default values to call ext4_group_add()
1558 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1559 * GDT blocks are reserved to grow to the desired size.
1561 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1562 ext4_fsblk_t n_blocks_count
)
1564 ext4_fsblk_t o_blocks_count
;
1567 struct buffer_head
*bh
;
1571 o_blocks_count
= ext4_blocks_count(es
);
1573 if (test_opt(sb
, DEBUG
))
1574 ext4_msg(sb
, KERN_DEBUG
,
1575 "extending last group from %llu to %llu blocks",
1576 o_blocks_count
, n_blocks_count
);
1578 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1581 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1582 ext4_msg(sb
, KERN_ERR
,
1583 "filesystem too large to resize to %llu blocks safely",
1585 if (sizeof(sector_t
) < 8)
1586 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1590 if (n_blocks_count
< o_blocks_count
) {
1591 ext4_warning(sb
, "can't shrink FS - resize aborted");
1595 /* Handle the remaining blocks in the last group only. */
1596 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1599 ext4_warning(sb
, "need to use ext2online to resize further");
1603 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1605 if (o_blocks_count
+ add
< o_blocks_count
) {
1606 ext4_warning(sb
, "blocks_count overflow");
1610 if (o_blocks_count
+ add
> n_blocks_count
)
1611 add
= n_blocks_count
- o_blocks_count
;
1613 if (o_blocks_count
+ add
< n_blocks_count
)
1614 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1615 o_blocks_count
+ add
, add
);
1617 /* See if the device is actually as big as what was requested */
1618 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1620 ext4_warning(sb
, "can't read last block, resize aborted");
1625 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1627 } /* ext4_group_extend */
1630 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1632 * @sb: super block of the fs to be resized
1633 * @n_blocks_count: the number of blocks resides in the resized fs
1635 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1637 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1638 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1639 struct ext4_super_block
*es
= sbi
->s_es
;
1640 struct buffer_head
*bh
;
1641 struct inode
*resize_inode
;
1642 ext4_fsblk_t o_blocks_count
;
1643 ext4_group_t o_group
;
1644 ext4_group_t n_group
;
1645 ext4_grpblk_t offset
, add
;
1646 unsigned long n_desc_blocks
;
1647 unsigned long o_desc_blocks
;
1648 unsigned long desc_blocks
;
1649 int err
= 0, flexbg_size
= 1;
1651 o_blocks_count
= ext4_blocks_count(es
);
1653 if (test_opt(sb
, DEBUG
))
1654 ext4_msg(sb
, KERN_DEBUG
, "resizing filesystem from %llu "
1655 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1657 if (n_blocks_count
< o_blocks_count
) {
1658 /* On-line shrinking not supported */
1659 ext4_warning(sb
, "can't shrink FS - resize aborted");
1663 if (n_blocks_count
== o_blocks_count
)
1664 /* Nothing need to do */
1667 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &offset
);
1668 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1670 n_desc_blocks
= (n_group
+ EXT4_DESC_PER_BLOCK(sb
)) /
1671 EXT4_DESC_PER_BLOCK(sb
);
1672 o_desc_blocks
= (sbi
->s_groups_count
+ EXT4_DESC_PER_BLOCK(sb
) - 1) /
1673 EXT4_DESC_PER_BLOCK(sb
);
1674 desc_blocks
= n_desc_blocks
- o_desc_blocks
;
1677 (!EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
) ||
1678 le16_to_cpu(es
->s_reserved_gdt_blocks
) < desc_blocks
)) {
1679 ext4_warning(sb
, "No reserved GDT blocks, can't resize");
1683 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1684 if (IS_ERR(resize_inode
)) {
1685 ext4_warning(sb
, "Error opening resize inode");
1686 return PTR_ERR(resize_inode
);
1689 /* See if the device is actually as big as what was requested */
1690 bh
= sb_bread(sb
, n_blocks_count
- 1);
1692 ext4_warning(sb
, "can't read last block, resize aborted");
1697 /* extend the last group */
1698 if (n_group
== o_group
)
1699 add
= n_blocks_count
- o_blocks_count
;
1701 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1703 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1708 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1709 es
->s_log_groups_per_flex
)
1710 flexbg_size
= 1 << es
->s_log_groups_per_flex
;
1712 o_blocks_count
= ext4_blocks_count(es
);
1713 if (o_blocks_count
== n_blocks_count
)
1716 flex_gd
= alloc_flex_gd(flexbg_size
);
1717 if (flex_gd
== NULL
) {
1722 /* Add flex groups. Note that a regular group is a
1723 * flex group with 1 group.
1725 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
1727 ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
);
1728 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
1735 free_flex_gd(flex_gd
);
1738 if (test_opt(sb
, DEBUG
))
1739 ext4_msg(sb
, KERN_DEBUG
, "resized filesystem from %llu "
1740 "upto %llu blocks", o_blocks_count
, n_blocks_count
);