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 static ext4_group_t
ext4_meta_bg_first_group(struct super_block
*sb
,
50 return (group
>> EXT4_DESC_PER_BLOCK_BITS(sb
)) <<
51 EXT4_DESC_PER_BLOCK_BITS(sb
);
54 static ext4_fsblk_t
ext4_meta_bg_first_block_no(struct super_block
*sb
,
56 group
= ext4_meta_bg_first_group(sb
, group
);
57 return ext4_group_first_block_no(sb
, group
);
60 static ext4_grpblk_t
ext4_group_overhead_blocks(struct super_block
*sb
,
62 ext4_grpblk_t overhead
;
63 overhead
= ext4_bg_num_gdb(sb
, group
);
64 if (ext4_bg_has_super(sb
, group
))
66 le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
70 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
71 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
73 static int verify_group_input(struct super_block
*sb
,
74 struct ext4_new_group_data
*input
)
76 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
77 struct ext4_super_block
*es
= sbi
->s_es
;
78 ext4_fsblk_t start
= ext4_blocks_count(es
);
79 ext4_fsblk_t end
= start
+ input
->blocks_count
;
80 ext4_group_t group
= input
->group
;
81 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
82 unsigned overhead
= ext4_group_overhead_blocks(sb
, group
);
83 ext4_fsblk_t metaend
= start
+ overhead
;
84 struct buffer_head
*bh
= NULL
;
85 ext4_grpblk_t free_blocks_count
, offset
;
88 input
->free_blocks_count
= free_blocks_count
=
89 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
91 if (test_opt(sb
, DEBUG
))
92 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
93 "(%d free, %u reserved)\n",
94 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
95 "no-super", input
->group
, input
->blocks_count
,
96 free_blocks_count
, input
->reserved_blocks
);
98 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
99 if (group
!= sbi
->s_groups_count
)
100 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
101 input
->group
, sbi
->s_groups_count
);
102 else if (offset
!= 0)
103 ext4_warning(sb
, "Last group not full");
104 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
105 ext4_warning(sb
, "Reserved blocks too high (%u)",
106 input
->reserved_blocks
);
107 else if (free_blocks_count
< 0)
108 ext4_warning(sb
, "Bad blocks count %u",
109 input
->blocks_count
);
110 else if (!(bh
= sb_bread(sb
, end
- 1)))
111 ext4_warning(sb
, "Cannot read last block (%llu)",
113 else if (outside(input
->block_bitmap
, start
, end
))
114 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
115 (unsigned long long)input
->block_bitmap
);
116 else if (outside(input
->inode_bitmap
, start
, end
))
117 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
118 (unsigned long long)input
->inode_bitmap
);
119 else if (outside(input
->inode_table
, start
, end
) ||
120 outside(itend
- 1, start
, end
))
121 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
122 (unsigned long long)input
->inode_table
, itend
- 1);
123 else if (input
->inode_bitmap
== input
->block_bitmap
)
124 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
125 (unsigned long long)input
->block_bitmap
);
126 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
127 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
129 (unsigned long long)input
->block_bitmap
,
130 (unsigned long long)input
->inode_table
, itend
- 1);
131 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
132 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
134 (unsigned long long)input
->inode_bitmap
,
135 (unsigned long long)input
->inode_table
, itend
- 1);
136 else if (inside(input
->block_bitmap
, start
, metaend
))
137 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
138 (unsigned long long)input
->block_bitmap
,
140 else if (inside(input
->inode_bitmap
, start
, metaend
))
141 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
142 (unsigned long long)input
->inode_bitmap
,
144 else if (inside(input
->inode_table
, start
, metaend
) ||
145 inside(itend
- 1, start
, metaend
))
146 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
148 (unsigned long long)input
->inode_table
,
149 itend
- 1, start
, metaend
- 1);
158 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
161 struct ext4_new_flex_group_data
{
162 struct ext4_new_group_data
*groups
; /* new_group_data for groups
164 __u16
*bg_flags
; /* block group flags of groups
166 ext4_group_t count
; /* number of groups in @groups
171 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
174 * Returns NULL on failure otherwise address of the allocated structure.
176 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
178 struct ext4_new_flex_group_data
*flex_gd
;
180 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
184 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_flex_group_data
))
186 flex_gd
->count
= flexbg_size
;
188 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
189 flexbg_size
, GFP_NOFS
);
190 if (flex_gd
->groups
== NULL
)
193 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
194 if (flex_gd
->bg_flags
== NULL
)
200 kfree(flex_gd
->groups
);
207 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
209 kfree(flex_gd
->bg_flags
);
210 kfree(flex_gd
->groups
);
215 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
216 * and inode tables for a flex group.
218 * This function is used by 64bit-resize. Note that this function allocates
219 * group tables from the 1st group of groups contained by @flexgd, which may
220 * be a partial of a flex group.
222 * @sb: super block of fs to which the groups belongs
224 * Returns 0 on a successful allocation of the metadata blocks in the
227 static int ext4_alloc_group_tables(struct super_block
*sb
,
228 struct ext4_new_flex_group_data
*flex_gd
,
231 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
232 ext4_fsblk_t start_blk
;
233 ext4_fsblk_t last_blk
;
234 ext4_group_t src_group
;
235 ext4_group_t bb_index
= 0;
236 ext4_group_t ib_index
= 0;
237 ext4_group_t it_index
= 0;
239 ext4_group_t last_group
;
242 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
244 src_group
= group_data
[0].group
;
245 last_group
= src_group
+ flex_gd
->count
- 1;
247 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
248 (last_group
& ~(flexbg_size
- 1))));
250 group
= group_data
[0].group
;
251 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
253 start_blk
= ext4_group_first_block_no(sb
, src_group
);
254 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
256 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
258 start_blk
+= overhead
;
260 /* We collect contiguous blocks as much as possible. */
262 for (; src_group
<= last_group
; src_group
++) {
263 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
265 last_blk
+= group_data
[src_group
- group
].blocks_count
;
270 /* Allocate block bitmaps */
271 for (; bb_index
< flex_gd
->count
; bb_index
++) {
272 if (start_blk
>= last_blk
)
274 group_data
[bb_index
].block_bitmap
= start_blk
++;
275 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
276 group
-= group_data
[0].group
;
277 group_data
[group
].free_blocks_count
--;
279 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
282 /* Allocate inode bitmaps */
283 for (; ib_index
< flex_gd
->count
; ib_index
++) {
284 if (start_blk
>= last_blk
)
286 group_data
[ib_index
].inode_bitmap
= start_blk
++;
287 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
288 group
-= group_data
[0].group
;
289 group_data
[group
].free_blocks_count
--;
291 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
294 /* Allocate inode tables */
295 for (; it_index
< flex_gd
->count
; it_index
++) {
296 if (start_blk
+ EXT4_SB(sb
)->s_itb_per_group
> last_blk
)
298 group_data
[it_index
].inode_table
= start_blk
;
299 ext4_get_group_no_and_offset(sb
, start_blk
, &group
, NULL
);
300 group
-= group_data
[0].group
;
301 group_data
[group
].free_blocks_count
-=
302 EXT4_SB(sb
)->s_itb_per_group
;
304 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
306 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
309 if (test_opt(sb
, DEBUG
)) {
311 group
= group_data
[0].group
;
313 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
314 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
317 for (i
= 0; i
< flex_gd
->count
; i
++) {
318 printk(KERN_DEBUG
"adding %s group %u: %u "
319 "blocks (%d free)\n",
320 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
321 "no-super", group
+ i
,
322 group_data
[i
].blocks_count
,
323 group_data
[i
].free_blocks_count
);
329 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
332 struct buffer_head
*bh
;
335 bh
= sb_getblk(sb
, blk
);
337 return ERR_PTR(-EIO
);
338 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
342 memset(bh
->b_data
, 0, sb
->s_blocksize
);
343 set_buffer_uptodate(bh
);
350 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
351 * If that fails, restart the transaction & regain write access for the
352 * buffer head which is used for block_bitmap modifications.
354 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
358 if (ext4_handle_has_enough_credits(handle
, thresh
))
361 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
365 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
374 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
376 * Helper function for ext4_setup_new_group_blocks() which set .
379 * @handle: journal handle
380 * @flex_gd: flex group data
382 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
383 struct ext4_new_flex_group_data
*flex_gd
,
384 ext4_fsblk_t block
, ext4_group_t count
)
388 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
389 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
391 struct buffer_head
*bh
;
395 ext4_get_group_no_and_offset(sb
, block
, &group
, NULL
);
396 start
= ext4_group_first_block_no(sb
, group
);
397 group
-= flex_gd
->groups
[0].group
;
399 count2
= sb
->s_blocksize
* 8 - (block
- start
);
403 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
404 BUG_ON(flex_gd
->count
> 1);
408 err
= extend_or_restart_transaction(handle
, 1);
412 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
416 err
= ext4_journal_get_write_access(handle
, bh
);
419 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
420 block
- start
, count2
);
421 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
423 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
433 * Set up the block and inode bitmaps, and the inode table for the new groups.
434 * This doesn't need to be part of the main transaction, since we are only
435 * changing blocks outside the actual filesystem. We still do journaling to
436 * ensure the recovery is correct in case of a failure just after resize.
437 * If any part of this fails, we simply abort the resize.
439 * setup_new_flex_group_blocks handles a flex group as follow:
440 * 1. copy super block and GDT, and initialize group tables if necessary.
441 * In this step, we only set bits in blocks bitmaps for blocks taken by
442 * super block and GDT.
443 * 2. allocate group tables in block bitmaps, that is, set bits in block
444 * bitmap for blocks taken by group tables.
446 static int setup_new_flex_group_blocks(struct super_block
*sb
,
447 struct ext4_new_flex_group_data
*flex_gd
)
449 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
452 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
453 struct ext4_super_block
*es
= sbi
->s_es
;
454 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
455 __u16
*bg_flags
= flex_gd
->bg_flags
;
457 ext4_group_t group
, count
;
458 struct buffer_head
*bh
= NULL
;
459 int reserved_gdb
, i
, j
, err
= 0, err2
;
462 BUG_ON(!flex_gd
->count
|| !group_data
||
463 group_data
[0].group
!= sbi
->s_groups_count
);
465 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
466 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
468 /* This transaction may be extended/restarted along the way */
469 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
471 return PTR_ERR(handle
);
473 group
= group_data
[0].group
;
474 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
475 unsigned long gdblocks
;
476 ext4_grpblk_t overhead
;
478 gdblocks
= ext4_bg_num_gdb(sb
, group
);
479 start
= ext4_group_first_block_no(sb
, group
);
481 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
485 ext4_group_t first_group
;
486 first_group
= ext4_meta_bg_first_group(sb
, group
);
487 if (first_group
!= group
+ 1 &&
488 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
492 block
= start
+ ext4_bg_has_super(sb
, group
);
493 /* Copy all of the GDT blocks into the backup in this group */
494 for (j
= 0; j
< gdblocks
; j
++, block
++) {
495 struct buffer_head
*gdb
;
497 ext4_debug("update backup group %#04llx\n", block
);
498 err
= extend_or_restart_transaction(handle
, 1);
502 gdb
= sb_getblk(sb
, block
);
508 err
= ext4_journal_get_write_access(handle
, gdb
);
513 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
515 set_buffer_uptodate(gdb
);
517 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
525 /* Zero out all of the reserved backup group descriptor
528 if (ext4_bg_has_super(sb
, group
)) {
529 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
530 reserved_gdb
, GFP_NOFS
);
536 /* Initialize group tables of the grop @group */
537 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
540 /* Zero out all of the inode table blocks */
541 block
= group_data
[i
].inode_table
;
542 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
543 block
, sbi
->s_itb_per_group
);
544 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
550 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
553 /* Initialize block bitmap of the @group */
554 block
= group_data
[i
].block_bitmap
;
555 err
= extend_or_restart_transaction(handle
, 1);
559 bh
= bclean(handle
, sb
, block
);
564 overhead
= ext4_group_overhead_blocks(sb
, group
);
566 ext4_debug("mark backup superblock %#04llx (+0)\n",
568 ext4_set_bits(bh
->b_data
, 0, overhead
);
570 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
571 sb
->s_blocksize
* 8, bh
->b_data
);
572 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
578 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
581 /* Initialize inode bitmap of the @group */
582 block
= group_data
[i
].inode_bitmap
;
583 err
= extend_or_restart_transaction(handle
, 1);
586 /* Mark unused entries in inode bitmap used */
587 bh
= bclean(handle
, sb
, block
);
593 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
594 sb
->s_blocksize
* 8, bh
->b_data
);
595 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
602 /* Mark group tables in block bitmap */
603 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
604 count
= group_table_count
[j
];
605 start
= (&group_data
[0].block_bitmap
)[j
];
607 for (i
= 1; i
< flex_gd
->count
; i
++) {
608 block
+= group_table_count
[j
];
609 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
610 count
+= group_table_count
[j
];
613 err
= set_flexbg_block_bitmap(sb
, handle
,
614 flex_gd
, start
, count
);
617 count
= group_table_count
[j
];
618 start
= group_data
[i
].block_bitmap
;
623 err
= set_flexbg_block_bitmap(sb
, handle
,
624 flex_gd
, start
, count
);
632 err2
= ext4_journal_stop(handle
);
640 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
641 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
642 * calling this for the first time. In a sparse filesystem it will be the
643 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
644 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
646 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
647 unsigned *five
, unsigned *seven
)
649 unsigned *min
= three
;
653 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
654 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
676 * Check that all of the backup GDT blocks are held in the primary GDT block.
677 * It is assumed that they are stored in group order. Returns the number of
678 * groups in current filesystem that have BACKUPS, or -ve error code.
680 static int verify_reserved_gdb(struct super_block
*sb
,
682 struct buffer_head
*primary
)
684 const ext4_fsblk_t blk
= primary
->b_blocknr
;
689 __le32
*p
= (__le32
*)primary
->b_data
;
692 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
693 if (le32_to_cpu(*p
++) !=
694 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
695 ext4_warning(sb
, "reserved GDT %llu"
696 " missing grp %d (%llu)",
699 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
703 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
711 * Called when we need to bring a reserved group descriptor table block into
712 * use from the resize inode. The primary copy of the new GDT block currently
713 * is an indirect block (under the double indirect block in the resize inode).
714 * The new backup GDT blocks will be stored as leaf blocks in this indirect
715 * block, in group order. Even though we know all the block numbers we need,
716 * we check to ensure that the resize inode has actually reserved these blocks.
718 * Don't need to update the block bitmaps because the blocks are still in use.
720 * We get all of the error cases out of the way, so that we are sure to not
721 * fail once we start modifying the data on disk, because JBD has no rollback.
723 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
726 struct super_block
*sb
= inode
->i_sb
;
727 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
728 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
729 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
730 struct buffer_head
**o_group_desc
, **n_group_desc
;
731 struct buffer_head
*dind
;
732 struct buffer_head
*gdb_bh
;
734 struct ext4_iloc iloc
;
738 if (test_opt(sb
, DEBUG
))
740 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
744 * If we are not using the primary superblock/GDT copy don't resize,
745 * because the user tools have no way of handling this. Probably a
746 * bad time to do it anyways.
748 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
749 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
750 ext4_warning(sb
, "won't resize using backup superblock at %llu",
751 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
755 gdb_bh
= sb_bread(sb
, gdblock
);
759 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
765 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
766 dind
= sb_bread(sb
, le32_to_cpu(*data
));
772 data
= (__le32
*)dind
->b_data
;
773 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
774 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
780 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
784 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
788 err
= ext4_journal_get_write_access(handle
, dind
);
790 ext4_std_error(sb
, err
);
792 /* ext4_reserve_inode_write() gets a reference on the iloc */
793 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
797 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
798 sizeof(struct buffer_head
*),
802 ext4_warning(sb
, "not enough memory for %lu groups",
808 * Finally, we have all of the possible failures behind us...
810 * Remove new GDT block from inode double-indirect block and clear out
811 * the new GDT block for use (which also "frees" the backup GDT blocks
812 * from the reserved inode). We don't need to change the bitmaps for
813 * these blocks, because they are marked as in-use from being in the
814 * reserved inode, and will become GDT blocks (primary and backup).
816 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
817 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
819 ext4_std_error(sb
, err
);
822 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
823 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
824 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
825 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
827 ext4_std_error(sb
, err
);
832 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
833 memcpy(n_group_desc
, o_group_desc
,
834 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
835 n_group_desc
[gdb_num
] = gdb_bh
;
836 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
837 EXT4_SB(sb
)->s_gdb_count
++;
838 ext4_kvfree(o_group_desc
);
840 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
841 err
= ext4_handle_dirty_super(handle
, sb
);
843 ext4_std_error(sb
, err
);
848 ext4_kvfree(n_group_desc
);
849 /* ext4_handle_release_buffer(handle, iloc.bh); */
852 /* ext4_handle_release_buffer(handle, dind); */
854 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
860 ext4_debug("leaving with error %d\n", err
);
865 * add_new_gdb_meta_bg is the sister of add_new_gdb.
867 static int add_new_gdb_meta_bg(struct super_block
*sb
,
868 handle_t
*handle
, ext4_group_t group
) {
869 ext4_fsblk_t gdblock
;
870 struct buffer_head
*gdb_bh
;
871 struct buffer_head
**o_group_desc
, **n_group_desc
;
872 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
875 gdblock
= ext4_meta_bg_first_block_no(sb
, group
) +
876 ext4_bg_has_super(sb
, group
);
877 gdb_bh
= sb_bread(sb
, gdblock
);
880 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
881 sizeof(struct buffer_head
*),
885 ext4_warning(sb
, "not enough memory for %lu groups",
890 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
891 memcpy(n_group_desc
, o_group_desc
,
892 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
893 n_group_desc
[gdb_num
] = gdb_bh
;
894 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
895 EXT4_SB(sb
)->s_gdb_count
++;
896 ext4_kvfree(o_group_desc
);
897 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
904 * Called when we are adding a new group which has a backup copy of each of
905 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
906 * We need to add these reserved backup GDT blocks to the resize inode, so
907 * that they are kept for future resizing and not allocated to files.
909 * Each reserved backup GDT block will go into a different indirect block.
910 * The indirect blocks are actually the primary reserved GDT blocks,
911 * so we know in advance what their block numbers are. We only get the
912 * double-indirect block to verify it is pointing to the primary reserved
913 * GDT blocks so we don't overwrite a data block by accident. The reserved
914 * backup GDT blocks are stored in their reserved primary GDT block.
916 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
919 struct super_block
*sb
= inode
->i_sb
;
920 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
921 struct buffer_head
**primary
;
922 struct buffer_head
*dind
;
923 struct ext4_iloc iloc
;
930 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
934 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
935 dind
= sb_bread(sb
, le32_to_cpu(*data
));
941 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
942 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
943 EXT4_ADDR_PER_BLOCK(sb
));
944 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
946 /* Get each reserved primary GDT block and verify it holds backups */
947 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
948 if (le32_to_cpu(*data
) != blk
) {
949 ext4_warning(sb
, "reserved block %llu"
950 " not at offset %ld",
952 (long)(data
- (__le32
*)dind
->b_data
));
956 primary
[res
] = sb_bread(sb
, blk
);
961 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
963 brelse(primary
[res
]);
968 data
= (__le32
*)dind
->b_data
;
971 for (i
= 0; i
< reserved_gdb
; i
++) {
972 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
975 for (j = 0; j < i; j++)
976 ext4_handle_release_buffer(handle, primary[j]);
982 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
986 * Finally we can add each of the reserved backup GDT blocks from
987 * the new group to its reserved primary GDT block.
989 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
990 for (i
= 0; i
< reserved_gdb
; i
++) {
992 data
= (__le32
*)primary
[i
]->b_data
;
993 /* printk("reserving backup %lu[%u] = %lu\n",
994 primary[i]->b_blocknr, gdbackups,
995 blk + primary[i]->b_blocknr); */
996 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
997 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1001 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
1002 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1006 brelse(primary
[res
]);
1016 * Update the backup copies of the ext4 metadata. These don't need to be part
1017 * of the main resize transaction, because e2fsck will re-write them if there
1018 * is a problem (basically only OOM will cause a problem). However, we
1019 * _should_ update the backups if possible, in case the primary gets trashed
1020 * for some reason and we need to run e2fsck from a backup superblock. The
1021 * important part is that the new block and inode counts are in the backup
1022 * superblocks, and the location of the new group metadata in the GDT backups.
1024 * We do not need take the s_resize_lock for this, because these
1025 * blocks are not otherwise touched by the filesystem code when it is
1026 * mounted. We don't need to worry about last changing from
1027 * sbi->s_groups_count, because the worst that can happen is that we
1028 * do not copy the full number of backups at this time. The resize
1029 * which changed s_groups_count will backup again.
1031 static void update_backups(struct super_block
*sb
, int blk_off
, char *data
,
1032 int size
, int meta_bg
)
1034 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1036 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1040 ext4_group_t group
= 0;
1041 int rest
= sb
->s_blocksize
- size
;
1045 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
1046 if (IS_ERR(handle
)) {
1048 err
= PTR_ERR(handle
);
1053 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1054 last
= sbi
->s_groups_count
;
1056 group
= ext4_meta_bg_first_group(sb
, group
) + 1;
1057 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1060 while (group
< sbi
->s_groups_count
) {
1061 struct buffer_head
*bh
;
1062 ext4_fsblk_t backup_block
;
1064 /* Out of journal space, and can't get more - abort - so sad */
1065 if (ext4_handle_valid(handle
) &&
1066 handle
->h_buffer_credits
== 0 &&
1067 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1068 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1072 backup_block
= group
* bpg
+ blk_off
;
1074 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1075 ext4_bg_has_super(sb
, group
));
1077 bh
= sb_getblk(sb
, backup_block
);
1082 ext4_debug("update metadata backup %llu(+%llu)\n",
1083 backup_block
, backup_block
-
1084 ext4_group_first_block_no(sb
, group
));
1085 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
1088 memcpy(bh
->b_data
, data
, size
);
1090 memset(bh
->b_data
+ size
, 0, rest
);
1091 set_buffer_uptodate(bh
);
1093 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1095 ext4_std_error(sb
, err
);
1099 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1100 else if (group
== last
)
1105 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1109 * Ugh! Need to have e2fsck write the backup copies. It is too
1110 * late to revert the resize, we shouldn't fail just because of
1111 * the backup copies (they are only needed in case of corruption).
1113 * However, if we got here we have a journal problem too, so we
1114 * can't really start a transaction to mark the superblock.
1115 * Chicken out and just set the flag on the hope it will be written
1116 * to disk, and if not - we will simply wait until next fsck.
1120 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1121 "forcing fsck on next reboot", group
, err
);
1122 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1123 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1124 mark_buffer_dirty(sbi
->s_sbh
);
1129 * ext4_add_new_descs() adds @count group descriptor of groups
1130 * starting at @group
1132 * @handle: journal handle
1134 * @group: the group no. of the first group desc to be added
1135 * @resize_inode: the resize inode
1136 * @count: number of group descriptors to be added
1138 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1139 ext4_group_t group
, struct inode
*resize_inode
,
1142 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1143 struct ext4_super_block
*es
= sbi
->s_es
;
1144 struct buffer_head
*gdb_bh
;
1145 int i
, gdb_off
, gdb_num
, err
= 0;
1148 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1149 for (i
= 0; i
< count
; i
++, group
++) {
1150 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1151 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1153 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1154 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1157 * We will only either add reserved group blocks to a backup group
1158 * or remove reserved blocks for the first group in a new group block.
1159 * Doing both would be mean more complex code, and sane people don't
1160 * use non-sparse filesystems anymore. This is already checked above.
1163 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1164 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1166 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1167 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1168 } else if (meta_bg
!= 0) {
1169 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1171 err
= add_new_gdb(handle
, resize_inode
, group
);
1179 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1181 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1184 if (!bh_uptodate_or_lock(bh
)) {
1185 if (bh_submit_read(bh
) < 0) {
1194 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1196 struct ext4_group_desc
*gdp
,
1197 struct ext4_new_group_data
*group_data
)
1199 struct buffer_head
*bh
;
1201 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1202 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1205 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1208 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1209 EXT4_INODES_PER_GROUP(sb
) / 8);
1212 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1215 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1222 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1224 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1225 struct ext4_new_flex_group_data
*flex_gd
)
1227 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1228 struct ext4_group_desc
*gdp
;
1229 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1230 struct buffer_head
*gdb_bh
;
1232 __u16
*bg_flags
= flex_gd
->bg_flags
;
1233 int i
, gdb_off
, gdb_num
, err
= 0;
1236 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1237 group
= group_data
->group
;
1239 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1240 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1243 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1245 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1246 /* Update group descriptor block for new group */
1247 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1248 gdb_off
* EXT4_DESC_SIZE(sb
));
1250 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1251 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1252 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1253 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1255 ext4_std_error(sb
, err
);
1259 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1260 ext4_free_group_clusters_set(sb
, gdp
,
1261 EXT4_B2C(sbi
, group_data
->free_blocks_count
));
1262 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1263 if (ext4_has_group_desc_csum(sb
))
1264 ext4_itable_unused_set(sb
, gdp
,
1265 EXT4_INODES_PER_GROUP(sb
));
1266 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1267 ext4_group_desc_csum_set(sb
, group
, gdp
);
1269 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1270 if (unlikely(err
)) {
1271 ext4_std_error(sb
, err
);
1276 * We can allocate memory for mb_alloc based on the new group
1279 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1287 * ext4_update_super() updates the super block so that the newly added
1288 * groups can be seen by the filesystem.
1291 * @flex_gd: new added groups
1293 static void ext4_update_super(struct super_block
*sb
,
1294 struct ext4_new_flex_group_data
*flex_gd
)
1296 ext4_fsblk_t blocks_count
= 0;
1297 ext4_fsblk_t free_blocks
= 0;
1298 ext4_fsblk_t reserved_blocks
= 0;
1299 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1300 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1301 struct ext4_super_block
*es
= sbi
->s_es
;
1304 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1306 * Make the new blocks and inodes valid next. We do this before
1307 * increasing the group count so that once the group is enabled,
1308 * all of its blocks and inodes are already valid.
1310 * We always allocate group-by-group, then block-by-block or
1311 * inode-by-inode within a group, so enabling these
1312 * blocks/inodes before the group is live won't actually let us
1313 * allocate the new space yet.
1315 for (i
= 0; i
< flex_gd
->count
; i
++) {
1316 blocks_count
+= group_data
[i
].blocks_count
;
1317 free_blocks
+= group_data
[i
].free_blocks_count
;
1320 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1321 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1322 reserved_blocks
*= blocks_count
;
1323 do_div(reserved_blocks
, 100);
1325 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1326 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1327 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1329 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1332 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1334 * We need to protect s_groups_count against other CPUs seeing
1335 * inconsistent state in the superblock.
1337 * The precise rules we use are:
1339 * * Writers must perform a smp_wmb() after updating all
1340 * dependent data and before modifying the groups count
1342 * * Readers must perform an smp_rmb() after reading the groups
1343 * count and before reading any dependent data.
1345 * NB. These rules can be relaxed when checking the group count
1346 * while freeing data, as we can only allocate from a block
1347 * group after serialising against the group count, and we can
1348 * only then free after serialising in turn against that
1353 /* Update the global fs size fields */
1354 sbi
->s_groups_count
+= flex_gd
->count
;
1356 /* Update the reserved block counts only once the new group is
1358 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1361 /* Update the free space counts */
1362 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1363 EXT4_B2C(sbi
, free_blocks
));
1364 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1365 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1367 ext4_debug("free blocks count %llu",
1368 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1369 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
1370 EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1371 sbi
->s_log_groups_per_flex
) {
1372 ext4_group_t flex_group
;
1373 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1374 atomic_add(EXT4_B2C(sbi
, free_blocks
),
1375 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1376 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1377 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1381 * Update the fs overhead information
1383 ext4_calculate_overhead(sb
);
1385 if (test_opt(sb
, DEBUG
))
1386 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1387 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1388 blocks_count
, free_blocks
, reserved_blocks
);
1391 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1392 * _before_ we start modifying the filesystem, because we cannot abort the
1393 * transaction and not have it write the data to disk.
1395 static int ext4_flex_group_add(struct super_block
*sb
,
1396 struct inode
*resize_inode
,
1397 struct ext4_new_flex_group_data
*flex_gd
)
1399 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1400 struct ext4_super_block
*es
= sbi
->s_es
;
1401 ext4_fsblk_t o_blocks_count
;
1405 unsigned reserved_gdb
;
1406 int err
= 0, err2
= 0, credit
;
1408 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1410 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1411 o_blocks_count
= ext4_blocks_count(es
);
1412 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1415 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1419 * We will always be modifying at least the superblock and GDT
1420 * block. If we are adding a group past the last current GDT block,
1421 * we will also modify the inode and the dindirect block. If we
1422 * are adding a group with superblock/GDT backups we will also
1423 * modify each of the reserved GDT dindirect blocks.
1425 credit
= flex_gd
->count
* 4 + reserved_gdb
;
1426 handle
= ext4_journal_start_sb(sb
, credit
);
1427 if (IS_ERR(handle
)) {
1428 err
= PTR_ERR(handle
);
1432 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1436 group
= flex_gd
->groups
[0].group
;
1437 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1438 err
= ext4_add_new_descs(handle
, sb
, group
,
1439 resize_inode
, flex_gd
->count
);
1443 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1447 ext4_update_super(sb
, flex_gd
);
1449 err
= ext4_handle_dirty_super(handle
, sb
);
1452 err2
= ext4_journal_stop(handle
);
1457 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1458 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1459 EXT4_DESC_PER_BLOCK(sb
));
1460 int meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
,
1461 EXT4_FEATURE_INCOMPAT_META_BG
);
1462 sector_t old_gdb
= 0;
1464 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1465 sizeof(struct ext4_super_block
), 0);
1466 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1467 struct buffer_head
*gdb_bh
;
1469 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1470 if (old_gdb
== gdb_bh
->b_blocknr
)
1472 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1473 gdb_bh
->b_size
, meta_bg
);
1474 old_gdb
= gdb_bh
->b_blocknr
;
1481 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1482 struct ext4_new_flex_group_data
*flex_gd
,
1483 ext4_fsblk_t n_blocks_count
,
1484 unsigned long flexbg_size
)
1486 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1487 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1488 ext4_fsblk_t o_blocks_count
;
1489 ext4_group_t n_group
;
1491 ext4_group_t last_group
;
1493 ext4_grpblk_t blocks_per_group
;
1496 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1498 o_blocks_count
= ext4_blocks_count(es
);
1500 if (o_blocks_count
== n_blocks_count
)
1503 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1505 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1507 last_group
= group
| (flexbg_size
- 1);
1508 if (last_group
> n_group
)
1509 last_group
= n_group
;
1511 flex_gd
->count
= last_group
- group
+ 1;
1513 for (i
= 0; i
< flex_gd
->count
; i
++) {
1516 group_data
[i
].group
= group
+ i
;
1517 group_data
[i
].blocks_count
= blocks_per_group
;
1518 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1519 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1520 if (ext4_has_group_desc_csum(sb
))
1521 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1522 EXT4_BG_INODE_UNINIT
;
1524 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1527 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1528 /* We need to initialize block bitmap of last group. */
1529 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1531 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1532 group_data
[i
- 1].blocks_count
= last
+ 1;
1533 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1540 /* Add group descriptor data to an existing or new group descriptor block.
1541 * Ensure we handle all possible error conditions _before_ we start modifying
1542 * the filesystem, because we cannot abort the transaction and not have it
1543 * write the data to disk.
1545 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1546 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1548 * We only need to hold the superblock lock while we are actually adding
1549 * in the new group's counts to the superblock. Prior to that we have
1550 * not really "added" the group at all. We re-check that we are still
1551 * adding in the last group in case things have changed since verifying.
1553 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1555 struct ext4_new_flex_group_data flex_gd
;
1556 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1557 struct ext4_super_block
*es
= sbi
->s_es
;
1558 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1559 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1560 struct inode
*inode
= NULL
;
1561 int gdb_off
, gdb_num
;
1565 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
1566 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1568 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1569 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
1570 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1574 if (ext4_blocks_count(es
) + input
->blocks_count
<
1575 ext4_blocks_count(es
)) {
1576 ext4_warning(sb
, "blocks_count overflow");
1580 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1581 le32_to_cpu(es
->s_inodes_count
)) {
1582 ext4_warning(sb
, "inodes_count overflow");
1586 if (reserved_gdb
|| gdb_off
== 0) {
1587 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
1588 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
1589 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1591 "No reserved GDT blocks, can't resize");
1594 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1595 if (IS_ERR(inode
)) {
1596 ext4_warning(sb
, "Error opening resize inode");
1597 return PTR_ERR(inode
);
1602 err
= verify_group_input(sb
, input
);
1606 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1610 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1615 flex_gd
.groups
= input
;
1616 flex_gd
.bg_flags
= &bg_flags
;
1617 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1621 } /* ext4_group_add */
1624 * extend a group without checking assuming that checking has been done.
1626 static int ext4_group_extend_no_check(struct super_block
*sb
,
1627 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1629 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1633 /* We will update the superblock, one block bitmap, and
1634 * one group descriptor via ext4_group_add_blocks().
1636 handle
= ext4_journal_start_sb(sb
, 3);
1637 if (IS_ERR(handle
)) {
1638 err
= PTR_ERR(handle
);
1639 ext4_warning(sb
, "error %d on journal start", err
);
1643 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1645 ext4_warning(sb
, "error %d on journal write access", err
);
1649 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1650 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1651 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1652 o_blocks_count
+ add
);
1653 /* We add the blocks to the bitmap and set the group need init bit */
1654 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1657 ext4_handle_dirty_super(handle
, sb
);
1658 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1659 o_blocks_count
+ add
);
1661 err2
= ext4_journal_stop(handle
);
1666 ext4_fsblk_t first_block
;
1667 first_block
= ext4_group_first_block_no(sb
, 0);
1668 if (test_opt(sb
, DEBUG
))
1669 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1670 "blocks\n", ext4_blocks_count(es
));
1671 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
- first_block
,
1672 (char *)es
, sizeof(struct ext4_super_block
), 0);
1678 * Extend the filesystem to the new number of blocks specified. This entry
1679 * point is only used to extend the current filesystem to the end of the last
1680 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1681 * for emergencies (because it has no dependencies on reserved blocks).
1683 * If we _really_ wanted, we could use default values to call ext4_group_add()
1684 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1685 * GDT blocks are reserved to grow to the desired size.
1687 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1688 ext4_fsblk_t n_blocks_count
)
1690 ext4_fsblk_t o_blocks_count
;
1693 struct buffer_head
*bh
;
1697 o_blocks_count
= ext4_blocks_count(es
);
1699 if (test_opt(sb
, DEBUG
))
1700 ext4_msg(sb
, KERN_DEBUG
,
1701 "extending last group from %llu to %llu blocks",
1702 o_blocks_count
, n_blocks_count
);
1704 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1707 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1708 ext4_msg(sb
, KERN_ERR
,
1709 "filesystem too large to resize to %llu blocks safely",
1711 if (sizeof(sector_t
) < 8)
1712 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1716 if (n_blocks_count
< o_blocks_count
) {
1717 ext4_warning(sb
, "can't shrink FS - resize aborted");
1721 /* Handle the remaining blocks in the last group only. */
1722 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1725 ext4_warning(sb
, "need to use ext2online to resize further");
1729 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1731 if (o_blocks_count
+ add
< o_blocks_count
) {
1732 ext4_warning(sb
, "blocks_count overflow");
1736 if (o_blocks_count
+ add
> n_blocks_count
)
1737 add
= n_blocks_count
- o_blocks_count
;
1739 if (o_blocks_count
+ add
< n_blocks_count
)
1740 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1741 o_blocks_count
+ add
, add
);
1743 /* See if the device is actually as big as what was requested */
1744 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1746 ext4_warning(sb
, "can't read last block, resize aborted");
1751 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1753 } /* ext4_group_extend */
1756 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1758 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1762 * Release the resize inode and drop the resize_inode feature if there
1763 * are no more reserved gdt blocks, and then convert the file system
1766 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1769 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1770 struct ext4_super_block
*es
= sbi
->s_es
;
1771 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1773 int i
, ret
, err
= 0;
1776 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1778 if (es
->s_reserved_gdt_blocks
) {
1779 ext4_error(sb
, "Unexpected non-zero "
1780 "s_reserved_gdt_blocks");
1784 /* Do a quick sanity check of the resize inode */
1785 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
- 9))
1786 goto invalid_resize_inode
;
1787 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1788 if (i
== EXT4_DIND_BLOCK
) {
1792 goto invalid_resize_inode
;
1795 goto invalid_resize_inode
;
1797 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1800 handle
= ext4_journal_start_sb(sb
, credits
);
1802 return PTR_ERR(handle
);
1804 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1808 EXT4_CLEAR_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
);
1809 EXT4_SET_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1810 sbi
->s_es
->s_first_meta_bg
=
1811 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1813 err
= ext4_handle_dirty_super(handle
, sb
);
1815 ext4_std_error(sb
, err
);
1820 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1821 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1822 EXT4_FREE_BLOCKS_METADATA
|
1823 EXT4_FREE_BLOCKS_FORGET
);
1824 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1825 inode
->i_blocks
= 0;
1827 err
= ext4_mark_inode_dirty(handle
, inode
);
1829 ext4_std_error(sb
, err
);
1833 ret
= ext4_journal_stop(handle
);
1838 invalid_resize_inode
:
1839 ext4_error(sb
, "corrupted/inconsistent resize inode");
1844 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1846 * @sb: super block of the fs to be resized
1847 * @n_blocks_count: the number of blocks resides in the resized fs
1849 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1851 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1852 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1853 struct ext4_super_block
*es
= sbi
->s_es
;
1854 struct buffer_head
*bh
;
1855 struct inode
*resize_inode
= NULL
;
1856 ext4_grpblk_t add
, offset
;
1857 unsigned long n_desc_blocks
;
1858 unsigned long o_desc_blocks
;
1859 ext4_group_t o_group
;
1860 ext4_group_t n_group
;
1861 ext4_fsblk_t o_blocks_count
;
1862 ext4_fsblk_t n_blocks_count_retry
= 0;
1863 unsigned long last_update_time
= 0;
1864 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1867 /* See if the device is actually as big as what was requested */
1868 bh
= sb_bread(sb
, n_blocks_count
- 1);
1870 ext4_warning(sb
, "can't read last block, resize aborted");
1876 o_blocks_count
= ext4_blocks_count(es
);
1878 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1879 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1881 if (n_blocks_count
< o_blocks_count
) {
1882 /* On-line shrinking not supported */
1883 ext4_warning(sb
, "can't shrink FS - resize aborted");
1887 if (n_blocks_count
== o_blocks_count
)
1888 /* Nothing need to do */
1891 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &offset
);
1892 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1894 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1895 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1897 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1899 if (EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
)) {
1901 ext4_error(sb
, "resize_inode and meta_bg enabled "
1905 if (n_desc_blocks
> o_desc_blocks
+
1906 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1907 n_blocks_count_retry
= n_blocks_count
;
1908 n_desc_blocks
= o_desc_blocks
+
1909 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1910 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1911 n_blocks_count
= n_group
* EXT4_BLOCKS_PER_GROUP(sb
);
1912 n_group
--; /* set to last group number */
1916 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1917 if (IS_ERR(resize_inode
)) {
1918 ext4_warning(sb
, "Error opening resize inode");
1919 return PTR_ERR(resize_inode
);
1923 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
1924 err
= ext4_convert_meta_bg(sb
, resize_inode
);
1929 resize_inode
= NULL
;
1931 if (n_blocks_count_retry
) {
1932 n_blocks_count
= n_blocks_count_retry
;
1933 n_blocks_count_retry
= 0;
1938 /* extend the last group */
1939 if (n_group
== o_group
)
1940 add
= n_blocks_count
- o_blocks_count
;
1942 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1944 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1949 if (ext4_blocks_count(es
) == n_blocks_count
)
1952 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
1956 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
1960 flex_gd
= alloc_flex_gd(flexbg_size
);
1961 if (flex_gd
== NULL
) {
1966 /* Add flex groups. Note that a regular group is a
1967 * flex group with 1 group.
1969 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
1971 if (jiffies
- last_update_time
> HZ
* 10) {
1972 if (last_update_time
)
1973 ext4_msg(sb
, KERN_INFO
,
1974 "resized to %llu blocks",
1975 ext4_blocks_count(es
));
1976 last_update_time
= jiffies
;
1978 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
1980 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
1985 if (!err
&& n_blocks_count_retry
) {
1986 n_blocks_count
= n_blocks_count_retry
;
1987 n_blocks_count_retry
= 0;
1988 free_flex_gd(flex_gd
);
1995 free_flex_gd(flex_gd
);
1996 if (resize_inode
!= NULL
)
1998 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu", n_blocks_count
);