2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
14 #include <linux/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 int ext4_resize_begin(struct super_block
*sb
)
23 if (!capable(CAP_SYS_RESOURCE
))
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
30 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
31 ext4_warning(sb
, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
36 if (test_and_set_bit_lock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
))
42 void ext4_resize_end(struct super_block
*sb
)
44 clear_bit_unlock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
);
45 smp_mb__after_clear_bit();
48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
49 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
51 static int verify_group_input(struct super_block
*sb
,
52 struct ext4_new_group_data
*input
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
55 struct ext4_super_block
*es
= sbi
->s_es
;
56 ext4_fsblk_t start
= ext4_blocks_count(es
);
57 ext4_fsblk_t end
= start
+ input
->blocks_count
;
58 ext4_group_t group
= input
->group
;
59 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
60 unsigned overhead
= ext4_bg_has_super(sb
, group
) ?
61 (1 + ext4_bg_num_gdb(sb
, group
) +
62 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
63 ext4_fsblk_t metaend
= start
+ overhead
;
64 struct buffer_head
*bh
= NULL
;
65 ext4_grpblk_t free_blocks_count
, offset
;
68 input
->free_blocks_count
= free_blocks_count
=
69 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
71 if (test_opt(sb
, DEBUG
))
72 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
73 "(%d free, %u reserved)\n",
74 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
75 "no-super", input
->group
, input
->blocks_count
,
76 free_blocks_count
, input
->reserved_blocks
);
78 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
79 if (group
!= sbi
->s_groups_count
)
80 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
81 input
->group
, sbi
->s_groups_count
);
83 ext4_warning(sb
, "Last group not full");
84 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
85 ext4_warning(sb
, "Reserved blocks too high (%u)",
86 input
->reserved_blocks
);
87 else if (free_blocks_count
< 0)
88 ext4_warning(sb
, "Bad blocks count %u",
90 else if (!(bh
= sb_bread(sb
, end
- 1)))
91 ext4_warning(sb
, "Cannot read last block (%llu)",
93 else if (outside(input
->block_bitmap
, start
, end
))
94 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
95 (unsigned long long)input
->block_bitmap
);
96 else if (outside(input
->inode_bitmap
, start
, end
))
97 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
98 (unsigned long long)input
->inode_bitmap
);
99 else if (outside(input
->inode_table
, start
, end
) ||
100 outside(itend
- 1, start
, end
))
101 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
102 (unsigned long long)input
->inode_table
, itend
- 1);
103 else if (input
->inode_bitmap
== input
->block_bitmap
)
104 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
105 (unsigned long long)input
->block_bitmap
);
106 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
107 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
109 (unsigned long long)input
->block_bitmap
,
110 (unsigned long long)input
->inode_table
, itend
- 1);
111 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
112 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
114 (unsigned long long)input
->inode_bitmap
,
115 (unsigned long long)input
->inode_table
, itend
- 1);
116 else if (inside(input
->block_bitmap
, start
, metaend
))
117 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 (unsigned long long)input
->block_bitmap
,
120 else if (inside(input
->inode_bitmap
, start
, metaend
))
121 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 (unsigned long long)input
->inode_bitmap
,
124 else if (inside(input
->inode_table
, start
, metaend
) ||
125 inside(itend
- 1, start
, metaend
))
126 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
128 (unsigned long long)input
->inode_table
,
129 itend
- 1, start
, metaend
- 1);
138 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
141 struct ext4_new_flex_group_data
{
142 struct ext4_new_group_data
*groups
; /* new_group_data for groups
144 __u16
*bg_flags
; /* block group flags of groups
146 ext4_group_t count
; /* number of groups in @groups
151 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
154 * Returns NULL on failure otherwise address of the allocated structure.
156 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
158 struct ext4_new_flex_group_data
*flex_gd
;
160 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
164 flex_gd
->count
= flexbg_size
;
166 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
167 flexbg_size
, GFP_NOFS
);
168 if (flex_gd
->groups
== NULL
)
171 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
172 if (flex_gd
->bg_flags
== NULL
)
178 kfree(flex_gd
->groups
);
185 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
187 kfree(flex_gd
->bg_flags
);
188 kfree(flex_gd
->groups
);
193 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
194 * and inode tables for a flex group.
196 * This function is used by 64bit-resize. Note that this function allocates
197 * group tables from the 1st group of groups contained by @flexgd, which may
198 * be a partial of a flex group.
200 * @sb: super block of fs to which the groups belongs
202 static void ext4_alloc_group_tables(struct super_block
*sb
,
203 struct ext4_new_flex_group_data
*flex_gd
,
206 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
207 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
208 ext4_fsblk_t start_blk
;
209 ext4_fsblk_t last_blk
;
210 ext4_group_t src_group
;
211 ext4_group_t bb_index
= 0;
212 ext4_group_t ib_index
= 0;
213 ext4_group_t it_index
= 0;
215 ext4_group_t last_group
;
218 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
220 src_group
= group_data
[0].group
;
221 last_group
= src_group
+ flex_gd
->count
- 1;
223 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
224 (last_group
& ~(flexbg_size
- 1))));
226 group
= group_data
[0].group
;
227 start_blk
= ext4_group_first_block_no(sb
, src_group
);
228 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
230 overhead
= ext4_bg_has_super(sb
, src_group
) ?
231 (1 + ext4_bg_num_gdb(sb
, src_group
) +
232 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
234 start_blk
+= overhead
;
236 BUG_ON(src_group
>= group_data
[0].group
+ flex_gd
->count
);
237 /* We collect contiguous blocks as much as possible. */
239 for (; src_group
<= last_group
; src_group
++)
240 if (!ext4_bg_has_super(sb
, src_group
))
241 last_blk
+= group_data
[src_group
- group
].blocks_count
;
245 /* Allocate block bitmaps */
246 for (; bb_index
< flex_gd
->count
; bb_index
++) {
247 if (start_blk
>= last_blk
)
249 group_data
[bb_index
].block_bitmap
= start_blk
++;
250 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
251 group
-= group_data
[0].group
;
252 group_data
[group
].free_blocks_count
--;
254 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
257 /* Allocate inode bitmaps */
258 for (; ib_index
< flex_gd
->count
; ib_index
++) {
259 if (start_blk
>= last_blk
)
261 group_data
[ib_index
].inode_bitmap
= start_blk
++;
262 ext4_get_group_no_and_offset(sb
, start_blk
- 1, &group
, NULL
);
263 group
-= group_data
[0].group
;
264 group_data
[group
].free_blocks_count
--;
266 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
269 /* Allocate inode tables */
270 for (; it_index
< flex_gd
->count
; it_index
++) {
271 if (start_blk
+ EXT4_SB(sb
)->s_itb_per_group
> last_blk
)
273 group_data
[it_index
].inode_table
= start_blk
;
274 ext4_get_group_no_and_offset(sb
, start_blk
, &group
, NULL
);
275 group
-= group_data
[0].group
;
276 group_data
[group
].free_blocks_count
-=
277 EXT4_SB(sb
)->s_itb_per_group
;
279 flex_gd
->bg_flags
[group
] &= ~EXT4_BG_BLOCK_UNINIT
;
281 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
284 if (test_opt(sb
, DEBUG
)) {
286 group
= group_data
[0].group
;
288 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
289 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
292 for (i
= 0; i
< flex_gd
->count
; i
++) {
293 printk(KERN_DEBUG
"adding %s group %u: %u "
294 "blocks (%d free)\n",
295 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
296 "no-super", group
+ i
,
297 group_data
[i
].blocks_count
,
298 group_data
[i
].free_blocks_count
);
303 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
306 struct buffer_head
*bh
;
309 bh
= sb_getblk(sb
, blk
);
311 return ERR_PTR(-EIO
);
312 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
316 memset(bh
->b_data
, 0, sb
->s_blocksize
);
317 set_buffer_uptodate(bh
);
324 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
325 * If that fails, restart the transaction & regain write access for the
326 * buffer head which is used for block_bitmap modifications.
328 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
332 if (ext4_handle_has_enough_credits(handle
, thresh
))
335 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
339 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
348 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
350 * Helper function for ext4_setup_new_group_blocks() which set .
353 * @handle: journal handle
354 * @flex_gd: flex group data
356 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
357 struct ext4_new_flex_group_data
*flex_gd
,
358 ext4_fsblk_t block
, ext4_group_t count
)
362 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
363 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
365 struct buffer_head
*bh
;
369 ext4_get_group_no_and_offset(sb
, block
, &group
, NULL
);
370 start
= ext4_group_first_block_no(sb
, group
);
371 group
-= flex_gd
->groups
[0].group
;
373 count2
= sb
->s_blocksize
* 8 - (block
- start
);
377 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
378 BUG_ON(flex_gd
->count
> 1);
382 err
= extend_or_restart_transaction(handle
, 1);
386 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
390 err
= ext4_journal_get_write_access(handle
, bh
);
393 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
394 block
- start
, count2
);
395 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
397 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
407 * Set up the block and inode bitmaps, and the inode table for the new groups.
408 * This doesn't need to be part of the main transaction, since we are only
409 * changing blocks outside the actual filesystem. We still do journaling to
410 * ensure the recovery is correct in case of a failure just after resize.
411 * If any part of this fails, we simply abort the resize.
413 * setup_new_flex_group_blocks handles a flex group as follow:
414 * 1. copy super block and GDT, and initialize group tables if necessary.
415 * In this step, we only set bits in blocks bitmaps for blocks taken by
416 * super block and GDT.
417 * 2. allocate group tables in block bitmaps, that is, set bits in block
418 * bitmap for blocks taken by group tables.
420 static int setup_new_flex_group_blocks(struct super_block
*sb
,
421 struct ext4_new_flex_group_data
*flex_gd
)
423 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
426 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
427 struct ext4_super_block
*es
= sbi
->s_es
;
428 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
429 __u16
*bg_flags
= flex_gd
->bg_flags
;
431 ext4_group_t group
, count
;
432 struct buffer_head
*bh
= NULL
;
433 int reserved_gdb
, i
, j
, err
= 0, err2
;
435 BUG_ON(!flex_gd
->count
|| !group_data
||
436 group_data
[0].group
!= sbi
->s_groups_count
);
438 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
440 /* This transaction may be extended/restarted along the way */
441 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
443 return PTR_ERR(handle
);
445 group
= group_data
[0].group
;
446 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
447 unsigned long gdblocks
;
449 gdblocks
= ext4_bg_num_gdb(sb
, group
);
450 start
= ext4_group_first_block_no(sb
, group
);
452 /* Copy all of the GDT blocks into the backup in this group */
453 for (j
= 0, block
= start
+ 1; j
< gdblocks
; j
++, block
++) {
454 struct buffer_head
*gdb
;
456 ext4_debug("update backup group %#04llx\n", block
);
457 err
= extend_or_restart_transaction(handle
, 1);
461 gdb
= sb_getblk(sb
, block
);
467 err
= ext4_journal_get_write_access(handle
, gdb
);
472 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
474 set_buffer_uptodate(gdb
);
476 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
484 /* Zero out all of the reserved backup group descriptor
487 if (ext4_bg_has_super(sb
, group
)) {
488 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
489 reserved_gdb
, GFP_NOFS
);
494 /* Initialize group tables of the grop @group */
495 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
498 /* Zero out all of the inode table blocks */
499 block
= group_data
[i
].inode_table
;
500 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
501 block
, sbi
->s_itb_per_group
);
502 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
508 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
511 /* Initialize block bitmap of the @group */
512 block
= group_data
[i
].block_bitmap
;
513 err
= extend_or_restart_transaction(handle
, 1);
517 bh
= bclean(handle
, sb
, block
);
522 if (ext4_bg_has_super(sb
, group
)) {
523 ext4_debug("mark backup superblock %#04llx (+0)\n",
525 ext4_set_bits(bh
->b_data
, 0, gdblocks
+ reserved_gdb
+
528 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
529 sb
->s_blocksize
* 8, bh
->b_data
);
530 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
536 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
539 /* Initialize inode bitmap of the @group */
540 block
= group_data
[i
].inode_bitmap
;
541 err
= extend_or_restart_transaction(handle
, 1);
544 /* Mark unused entries in inode bitmap used */
545 bh
= bclean(handle
, sb
, block
);
551 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
552 sb
->s_blocksize
* 8, bh
->b_data
);
553 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
560 /* Mark group tables in block bitmap */
561 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
562 count
= group_table_count
[j
];
563 start
= (&group_data
[0].block_bitmap
)[j
];
565 for (i
= 1; i
< flex_gd
->count
; i
++) {
566 block
+= group_table_count
[j
];
567 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
568 count
+= group_table_count
[j
];
571 err
= set_flexbg_block_bitmap(sb
, handle
,
572 flex_gd
, start
, count
);
575 count
= group_table_count
[j
];
576 start
= group_data
[i
].block_bitmap
;
581 err
= set_flexbg_block_bitmap(sb
, handle
,
582 flex_gd
, start
, count
);
590 err2
= ext4_journal_stop(handle
);
598 * Set up the block and inode bitmaps, and the inode table for the new group.
599 * This doesn't need to be part of the main transaction, since we are only
600 * changing blocks outside the actual filesystem. We still do journaling to
601 * ensure the recovery is correct in case of a failure just after resize.
602 * If any part of this fails, we simply abort the resize.
604 static int setup_new_group_blocks(struct super_block
*sb
,
605 struct ext4_new_group_data
*input
)
607 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
608 ext4_fsblk_t start
= ext4_group_first_block_no(sb
, input
->group
);
609 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
610 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
) : 0;
611 unsigned long gdblocks
= ext4_bg_num_gdb(sb
, input
->group
);
612 struct buffer_head
*bh
;
619 /* This transaction may be extended/restarted along the way */
620 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
623 return PTR_ERR(handle
);
625 BUG_ON(input
->group
!= sbi
->s_groups_count
);
627 /* Copy all of the GDT blocks into the backup in this group */
628 for (i
= 0, bit
= 1, block
= start
+ 1;
629 i
< gdblocks
; i
++, block
++, bit
++) {
630 struct buffer_head
*gdb
;
632 ext4_debug("update backup group %#04llx (+%d)\n", block
, bit
);
633 err
= extend_or_restart_transaction(handle
, 1);
637 gdb
= sb_getblk(sb
, block
);
642 if ((err
= ext4_journal_get_write_access(handle
, gdb
))) {
646 memcpy(gdb
->b_data
, sbi
->s_group_desc
[i
]->b_data
, gdb
->b_size
);
647 set_buffer_uptodate(gdb
);
648 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
656 /* Zero out all of the reserved backup group descriptor table blocks */
657 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
658 block
, sbi
->s_itb_per_group
);
659 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1, reserved_gdb
,
664 err
= extend_or_restart_transaction(handle
, 2);
668 bh
= bclean(handle
, sb
, input
->block_bitmap
);
674 if (ext4_bg_has_super(sb
, input
->group
)) {
675 ext4_debug("mark backup group tables %#04llx (+0)\n", start
);
676 ext4_set_bits(bh
->b_data
, 0, gdblocks
+ reserved_gdb
+ 1);
679 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input
->block_bitmap
,
680 input
->block_bitmap
- start
);
681 ext4_set_bit(input
->block_bitmap
- start
, bh
->b_data
);
682 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input
->inode_bitmap
,
683 input
->inode_bitmap
- start
);
684 ext4_set_bit(input
->inode_bitmap
- start
, bh
->b_data
);
686 /* Zero out all of the inode table blocks */
687 block
= input
->inode_table
;
688 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
689 block
, sbi
->s_itb_per_group
);
690 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
, GFP_NOFS
);
693 ext4_set_bits(bh
->b_data
, input
->inode_table
- start
,
694 sbi
->s_itb_per_group
);
697 ext4_mark_bitmap_end(input
->blocks_count
, sb
->s_blocksize
* 8,
699 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
701 ext4_std_error(sb
, err
);
705 /* Mark unused entries in inode bitmap used */
706 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
707 input
->inode_bitmap
, input
->inode_bitmap
- start
);
708 if (IS_ERR(bh
= bclean(handle
, sb
, input
->inode_bitmap
))) {
713 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
715 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
717 ext4_std_error(sb
, err
);
722 if ((err2
= ext4_journal_stop(handle
)) && !err
)
729 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
730 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
731 * calling this for the first time. In a sparse filesystem it will be the
732 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
733 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
735 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
736 unsigned *five
, unsigned *seven
)
738 unsigned *min
= three
;
742 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
743 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
765 * Check that all of the backup GDT blocks are held in the primary GDT block.
766 * It is assumed that they are stored in group order. Returns the number of
767 * groups in current filesystem that have BACKUPS, or -ve error code.
769 static int verify_reserved_gdb(struct super_block
*sb
,
771 struct buffer_head
*primary
)
773 const ext4_fsblk_t blk
= primary
->b_blocknr
;
778 __le32
*p
= (__le32
*)primary
->b_data
;
781 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
782 if (le32_to_cpu(*p
++) !=
783 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
784 ext4_warning(sb
, "reserved GDT %llu"
785 " missing grp %d (%llu)",
788 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
792 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
800 * Called when we need to bring a reserved group descriptor table block into
801 * use from the resize inode. The primary copy of the new GDT block currently
802 * is an indirect block (under the double indirect block in the resize inode).
803 * The new backup GDT blocks will be stored as leaf blocks in this indirect
804 * block, in group order. Even though we know all the block numbers we need,
805 * we check to ensure that the resize inode has actually reserved these blocks.
807 * Don't need to update the block bitmaps because the blocks are still in use.
809 * We get all of the error cases out of the way, so that we are sure to not
810 * fail once we start modifying the data on disk, because JBD has no rollback.
812 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
815 struct super_block
*sb
= inode
->i_sb
;
816 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
817 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
818 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
819 struct buffer_head
**o_group_desc
, **n_group_desc
;
820 struct buffer_head
*dind
;
821 struct buffer_head
*gdb_bh
;
823 struct ext4_iloc iloc
;
827 if (test_opt(sb
, DEBUG
))
829 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
833 * If we are not using the primary superblock/GDT copy don't resize,
834 * because the user tools have no way of handling this. Probably a
835 * bad time to do it anyways.
837 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
838 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
839 ext4_warning(sb
, "won't resize using backup superblock at %llu",
840 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
844 gdb_bh
= sb_bread(sb
, gdblock
);
848 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
854 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
855 dind
= sb_bread(sb
, le32_to_cpu(*data
));
861 data
= (__le32
*)dind
->b_data
;
862 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
863 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
869 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
873 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
877 err
= ext4_journal_get_write_access(handle
, dind
);
879 ext4_std_error(sb
, err
);
881 /* ext4_reserve_inode_write() gets a reference on the iloc */
882 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
886 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
887 sizeof(struct buffer_head
*),
891 ext4_warning(sb
, "not enough memory for %lu groups",
897 * Finally, we have all of the possible failures behind us...
899 * Remove new GDT block from inode double-indirect block and clear out
900 * the new GDT block for use (which also "frees" the backup GDT blocks
901 * from the reserved inode). We don't need to change the bitmaps for
902 * these blocks, because they are marked as in-use from being in the
903 * reserved inode, and will become GDT blocks (primary and backup).
905 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
906 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
908 ext4_std_error(sb
, err
);
911 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
912 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
913 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
914 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
916 ext4_std_error(sb
, err
);
921 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
922 memcpy(n_group_desc
, o_group_desc
,
923 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
924 n_group_desc
[gdb_num
] = gdb_bh
;
925 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
926 EXT4_SB(sb
)->s_gdb_count
++;
927 ext4_kvfree(o_group_desc
);
929 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
930 err
= ext4_handle_dirty_metadata(handle
, NULL
, EXT4_SB(sb
)->s_sbh
);
932 ext4_std_error(sb
, err
);
937 ext4_kvfree(n_group_desc
);
938 /* ext4_handle_release_buffer(handle, iloc.bh); */
941 /* ext4_handle_release_buffer(handle, dind); */
943 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
949 ext4_debug("leaving with error %d\n", err
);
954 * Called when we are adding a new group which has a backup copy of each of
955 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
956 * We need to add these reserved backup GDT blocks to the resize inode, so
957 * that they are kept for future resizing and not allocated to files.
959 * Each reserved backup GDT block will go into a different indirect block.
960 * The indirect blocks are actually the primary reserved GDT blocks,
961 * so we know in advance what their block numbers are. We only get the
962 * double-indirect block to verify it is pointing to the primary reserved
963 * GDT blocks so we don't overwrite a data block by accident. The reserved
964 * backup GDT blocks are stored in their reserved primary GDT block.
966 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
969 struct super_block
*sb
= inode
->i_sb
;
970 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
971 struct buffer_head
**primary
;
972 struct buffer_head
*dind
;
973 struct ext4_iloc iloc
;
980 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
984 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
985 dind
= sb_bread(sb
, le32_to_cpu(*data
));
991 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
992 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
993 EXT4_ADDR_PER_BLOCK(sb
));
994 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
996 /* Get each reserved primary GDT block and verify it holds backups */
997 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
998 if (le32_to_cpu(*data
) != blk
) {
999 ext4_warning(sb
, "reserved block %llu"
1000 " not at offset %ld",
1002 (long)(data
- (__le32
*)dind
->b_data
));
1006 primary
[res
] = sb_bread(sb
, blk
);
1007 if (!primary
[res
]) {
1011 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
1012 if (gdbackups
< 0) {
1013 brelse(primary
[res
]);
1018 data
= (__le32
*)dind
->b_data
;
1021 for (i
= 0; i
< reserved_gdb
; i
++) {
1022 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
1025 for (j = 0; j < i; j++)
1026 ext4_handle_release_buffer(handle, primary[j]);
1032 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
1036 * Finally we can add each of the reserved backup GDT blocks from
1037 * the new group to its reserved primary GDT block.
1039 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1040 for (i
= 0; i
< reserved_gdb
; i
++) {
1042 data
= (__le32
*)primary
[i
]->b_data
;
1043 /* printk("reserving backup %lu[%u] = %lu\n",
1044 primary[i]->b_blocknr, gdbackups,
1045 blk + primary[i]->b_blocknr); */
1046 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1047 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1051 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
1052 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1056 brelse(primary
[res
]);
1066 * Update the backup copies of the ext4 metadata. These don't need to be part
1067 * of the main resize transaction, because e2fsck will re-write them if there
1068 * is a problem (basically only OOM will cause a problem). However, we
1069 * _should_ update the backups if possible, in case the primary gets trashed
1070 * for some reason and we need to run e2fsck from a backup superblock. The
1071 * important part is that the new block and inode counts are in the backup
1072 * superblocks, and the location of the new group metadata in the GDT backups.
1074 * We do not need take the s_resize_lock for this, because these
1075 * blocks are not otherwise touched by the filesystem code when it is
1076 * mounted. We don't need to worry about last changing from
1077 * sbi->s_groups_count, because the worst that can happen is that we
1078 * do not copy the full number of backups at this time. The resize
1079 * which changed s_groups_count will backup again.
1081 static void update_backups(struct super_block
*sb
,
1082 int blk_off
, char *data
, int size
)
1084 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1085 const ext4_group_t last
= sbi
->s_groups_count
;
1086 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1091 int rest
= sb
->s_blocksize
- size
;
1095 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
1096 if (IS_ERR(handle
)) {
1098 err
= PTR_ERR(handle
);
1102 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
1103 struct buffer_head
*bh
;
1105 /* Out of journal space, and can't get more - abort - so sad */
1106 if (ext4_handle_valid(handle
) &&
1107 handle
->h_buffer_credits
== 0 &&
1108 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1109 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1112 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
1117 ext4_debug("update metadata backup %#04lx\n",
1118 (unsigned long)bh
->b_blocknr
);
1119 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
1122 memcpy(bh
->b_data
, data
, size
);
1124 memset(bh
->b_data
+ size
, 0, rest
);
1125 set_buffer_uptodate(bh
);
1127 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1129 ext4_std_error(sb
, err
);
1132 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1136 * Ugh! Need to have e2fsck write the backup copies. It is too
1137 * late to revert the resize, we shouldn't fail just because of
1138 * the backup copies (they are only needed in case of corruption).
1140 * However, if we got here we have a journal problem too, so we
1141 * can't really start a transaction to mark the superblock.
1142 * Chicken out and just set the flag on the hope it will be written
1143 * to disk, and if not - we will simply wait until next fsck.
1147 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1148 "forcing fsck on next reboot", group
, err
);
1149 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1150 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1151 mark_buffer_dirty(sbi
->s_sbh
);
1156 * ext4_add_new_descs() adds @count group descriptor of groups
1157 * starting at @group
1159 * @handle: journal handle
1161 * @group: the group no. of the first group desc to be added
1162 * @resize_inode: the resize inode
1163 * @count: number of group descriptors to be added
1165 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1166 ext4_group_t group
, struct inode
*resize_inode
,
1169 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1170 struct ext4_super_block
*es
= sbi
->s_es
;
1171 struct buffer_head
*gdb_bh
;
1172 int i
, gdb_off
, gdb_num
, err
= 0;
1174 for (i
= 0; i
< count
; i
++, group
++) {
1175 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1176 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1178 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1179 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1182 * We will only either add reserved group blocks to a backup group
1183 * or remove reserved blocks for the first group in a new group block.
1184 * Doing both would be mean more complex code, and sane people don't
1185 * use non-sparse filesystems anymore. This is already checked above.
1188 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1189 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1191 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1192 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1194 err
= add_new_gdb(handle
, resize_inode
, group
);
1202 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1204 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1205 struct ext4_new_flex_group_data
*flex_gd
)
1207 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1208 struct ext4_group_desc
*gdp
;
1209 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1210 struct buffer_head
*gdb_bh
;
1212 __u16
*bg_flags
= flex_gd
->bg_flags
;
1213 int i
, gdb_off
, gdb_num
, err
= 0;
1216 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1217 group
= group_data
->group
;
1219 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1220 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1223 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1225 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1226 /* Update group descriptor block for new group */
1227 gdp
= (struct ext4_group_desc
*)((char *)gdb_bh
->b_data
+
1228 gdb_off
* EXT4_DESC_SIZE(sb
));
1230 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1231 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1232 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1233 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1234 ext4_free_group_clusters_set(sb
, gdp
,
1235 EXT4_B2C(sbi
, group_data
->free_blocks_count
));
1236 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1237 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1238 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, group
, gdp
);
1240 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1241 if (unlikely(err
)) {
1242 ext4_std_error(sb
, err
);
1247 * We can allocate memory for mb_alloc based on the new group
1250 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1258 * ext4_update_super() updates the super block so that the newly added
1259 * groups can be seen by the filesystem.
1262 * @flex_gd: new added groups
1264 static void ext4_update_super(struct super_block
*sb
,
1265 struct ext4_new_flex_group_data
*flex_gd
)
1267 ext4_fsblk_t blocks_count
= 0;
1268 ext4_fsblk_t free_blocks
= 0;
1269 ext4_fsblk_t reserved_blocks
= 0;
1270 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1271 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1272 struct ext4_super_block
*es
= sbi
->s_es
;
1275 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1277 * Make the new blocks and inodes valid next. We do this before
1278 * increasing the group count so that once the group is enabled,
1279 * all of its blocks and inodes are already valid.
1281 * We always allocate group-by-group, then block-by-block or
1282 * inode-by-inode within a group, so enabling these
1283 * blocks/inodes before the group is live won't actually let us
1284 * allocate the new space yet.
1286 for (i
= 0; i
< flex_gd
->count
; i
++) {
1287 blocks_count
+= group_data
[i
].blocks_count
;
1288 free_blocks
+= group_data
[i
].free_blocks_count
;
1291 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1292 do_div(reserved_blocks
, ext4_blocks_count(es
));
1293 reserved_blocks
*= blocks_count
;
1294 do_div(reserved_blocks
, 100);
1296 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1297 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1301 * We need to protect s_groups_count against other CPUs seeing
1302 * inconsistent state in the superblock.
1304 * The precise rules we use are:
1306 * * Writers must perform a smp_wmb() after updating all
1307 * dependent data and before modifying the groups count
1309 * * Readers must perform an smp_rmb() after reading the groups
1310 * count and before reading any dependent data.
1312 * NB. These rules can be relaxed when checking the group count
1313 * while freeing data, as we can only allocate from a block
1314 * group after serialising against the group count, and we can
1315 * only then free after serialising in turn against that
1320 /* Update the global fs size fields */
1321 sbi
->s_groups_count
+= flex_gd
->count
;
1323 /* Update the reserved block counts only once the new group is
1325 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1328 /* Update the free space counts */
1329 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1330 EXT4_B2C(sbi
, free_blocks
));
1331 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1332 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1334 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
1335 EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1336 sbi
->s_log_groups_per_flex
) {
1337 ext4_group_t flex_group
;
1338 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1339 atomic_add(EXT4_B2C(sbi
, free_blocks
),
1340 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1341 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1342 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1345 if (test_opt(sb
, DEBUG
))
1346 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1347 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1348 blocks_count
, free_blocks
, reserved_blocks
);
1351 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1352 * _before_ we start modifying the filesystem, because we cannot abort the
1353 * transaction and not have it write the data to disk.
1355 static int ext4_flex_group_add(struct super_block
*sb
,
1356 struct inode
*resize_inode
,
1357 struct ext4_new_flex_group_data
*flex_gd
)
1359 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1360 struct ext4_super_block
*es
= sbi
->s_es
;
1361 ext4_fsblk_t o_blocks_count
;
1365 unsigned reserved_gdb
;
1366 int err
= 0, err2
= 0, credit
;
1368 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1370 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1371 o_blocks_count
= ext4_blocks_count(es
);
1372 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1375 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1379 * We will always be modifying at least the superblock and GDT
1380 * block. If we are adding a group past the last current GDT block,
1381 * we will also modify the inode and the dindirect block. If we
1382 * are adding a group with superblock/GDT backups we will also
1383 * modify each of the reserved GDT dindirect blocks.
1385 credit
= flex_gd
->count
* 4 + reserved_gdb
;
1386 handle
= ext4_journal_start_sb(sb
, credit
);
1387 if (IS_ERR(handle
)) {
1388 err
= PTR_ERR(handle
);
1392 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1396 group
= flex_gd
->groups
[0].group
;
1397 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1398 err
= ext4_add_new_descs(handle
, sb
, group
,
1399 resize_inode
, flex_gd
->count
);
1403 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1407 ext4_update_super(sb
, flex_gd
);
1409 err
= ext4_handle_dirty_super(handle
, sb
);
1412 err2
= ext4_journal_stop(handle
);
1418 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1419 sizeof(struct ext4_super_block
));
1420 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
1421 struct buffer_head
*gdb_bh
;
1423 gdb_num
= group
/ EXT4_BLOCKS_PER_GROUP(sb
);
1424 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1425 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1433 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1434 struct ext4_new_flex_group_data
*flex_gd
,
1435 ext4_fsblk_t n_blocks_count
,
1436 unsigned long flexbg_size
)
1438 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1439 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1440 ext4_fsblk_t o_blocks_count
;
1441 ext4_group_t n_group
;
1443 ext4_group_t last_group
;
1445 ext4_grpblk_t blocks_per_group
;
1448 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1450 o_blocks_count
= ext4_blocks_count(es
);
1452 if (o_blocks_count
== n_blocks_count
)
1455 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1457 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1459 last_group
= group
| (flexbg_size
- 1);
1460 if (last_group
> n_group
)
1461 last_group
= n_group
;
1463 flex_gd
->count
= last_group
- group
+ 1;
1465 for (i
= 0; i
< flex_gd
->count
; i
++) {
1468 group_data
[i
].group
= group
+ i
;
1469 group_data
[i
].blocks_count
= blocks_per_group
;
1470 overhead
= ext4_bg_has_super(sb
, group
+ i
) ?
1471 (1 + ext4_bg_num_gdb(sb
, group
+ i
) +
1472 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
1473 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1474 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1475 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
1476 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1477 EXT4_BG_INODE_UNINIT
;
1479 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1482 if (last_group
== n_group
&&
1483 EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1484 EXT4_FEATURE_RO_COMPAT_GDT_CSUM
))
1485 /* We need to initialize block bitmap of last group. */
1486 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1488 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1489 group_data
[i
- 1].blocks_count
= last
+ 1;
1490 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1497 /* Add group descriptor data to an existing or new group descriptor block.
1498 * Ensure we handle all possible error conditions _before_ we start modifying
1499 * the filesystem, because we cannot abort the transaction and not have it
1500 * write the data to disk.
1502 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1503 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1505 * We only need to hold the superblock lock while we are actually adding
1506 * in the new group's counts to the superblock. Prior to that we have
1507 * not really "added" the group at all. We re-check that we are still
1508 * adding in the last group in case things have changed since verifying.
1510 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1512 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1513 struct ext4_super_block
*es
= sbi
->s_es
;
1514 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1515 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1516 struct buffer_head
*primary
= NULL
;
1517 struct ext4_group_desc
*gdp
;
1518 struct inode
*inode
= NULL
;
1520 int gdb_off
, gdb_num
;
1523 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
1524 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1526 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1527 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
1528 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1532 if (ext4_blocks_count(es
) + input
->blocks_count
<
1533 ext4_blocks_count(es
)) {
1534 ext4_warning(sb
, "blocks_count overflow");
1538 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1539 le32_to_cpu(es
->s_inodes_count
)) {
1540 ext4_warning(sb
, "inodes_count overflow");
1544 if (reserved_gdb
|| gdb_off
== 0) {
1545 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
1546 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
1547 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1549 "No reserved GDT blocks, can't resize");
1552 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1553 if (IS_ERR(inode
)) {
1554 ext4_warning(sb
, "Error opening resize inode");
1555 return PTR_ERR(inode
);
1560 if ((err
= verify_group_input(sb
, input
)))
1563 if ((err
= setup_new_group_blocks(sb
, input
)))
1567 * We will always be modifying at least the superblock and a GDT
1568 * block. If we are adding a group past the last current GDT block,
1569 * we will also modify the inode and the dindirect block. If we
1570 * are adding a group with superblock/GDT backups we will also
1571 * modify each of the reserved GDT dindirect blocks.
1573 handle
= ext4_journal_start_sb(sb
,
1574 ext4_bg_has_super(sb
, input
->group
) ?
1575 3 + reserved_gdb
: 4);
1576 if (IS_ERR(handle
)) {
1577 err
= PTR_ERR(handle
);
1581 if ((err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
)))
1585 * We will only either add reserved group blocks to a backup group
1586 * or remove reserved blocks for the first group in a new group block.
1587 * Doing both would be mean more complex code, and sane people don't
1588 * use non-sparse filesystems anymore. This is already checked above.
1591 primary
= sbi
->s_group_desc
[gdb_num
];
1592 if ((err
= ext4_journal_get_write_access(handle
, primary
)))
1595 if (reserved_gdb
&& ext4_bg_num_gdb(sb
, input
->group
)) {
1596 err
= reserve_backup_gdb(handle
, inode
, input
->group
);
1602 * Note that we can access new group descriptor block safely
1603 * only if add_new_gdb() succeeds.
1605 err
= add_new_gdb(handle
, inode
, input
->group
);
1608 primary
= sbi
->s_group_desc
[gdb_num
];
1612 * OK, now we've set up the new group. Time to make it active.
1614 * so we have to be safe wrt. concurrent accesses the group
1615 * data. So we need to be careful to set all of the relevant
1616 * group descriptor data etc. *before* we enable the group.
1618 * The key field here is sbi->s_groups_count: as long as
1619 * that retains its old value, nobody is going to access the new
1622 * So first we update all the descriptor metadata for the new
1623 * group; then we update the total disk blocks count; then we
1624 * update the groups count to enable the group; then finally we
1625 * update the free space counts so that the system can start
1626 * using the new disk blocks.
1629 /* Update group descriptor block for new group */
1630 gdp
= (struct ext4_group_desc
*)((char *)primary
->b_data
+
1631 gdb_off
* EXT4_DESC_SIZE(sb
));
1633 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1634 ext4_block_bitmap_set(sb
, gdp
, input
->block_bitmap
); /* LV FIXME */
1635 ext4_inode_bitmap_set(sb
, gdp
, input
->inode_bitmap
); /* LV FIXME */
1636 ext4_inode_table_set(sb
, gdp
, input
->inode_table
); /* LV FIXME */
1637 ext4_free_group_clusters_set(sb
, gdp
, input
->free_blocks_count
);
1638 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1639 gdp
->bg_flags
= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
1640 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, input
->group
, gdp
);
1643 * We can allocate memory for mb_alloc based on the new group
1646 err
= ext4_mb_add_groupinfo(sb
, input
->group
, gdp
);
1651 * Make the new blocks and inodes valid next. We do this before
1652 * increasing the group count so that once the group is enabled,
1653 * all of its blocks and inodes are already valid.
1655 * We always allocate group-by-group, then block-by-block or
1656 * inode-by-inode within a group, so enabling these
1657 * blocks/inodes before the group is live won't actually let us
1658 * allocate the new space yet.
1660 ext4_blocks_count_set(es
, ext4_blocks_count(es
) +
1661 input
->blocks_count
);
1662 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
));
1665 * We need to protect s_groups_count against other CPUs seeing
1666 * inconsistent state in the superblock.
1668 * The precise rules we use are:
1670 * * Writers must perform a smp_wmb() after updating all dependent
1671 * data and before modifying the groups count
1673 * * Readers must perform an smp_rmb() after reading the groups count
1674 * and before reading any dependent data.
1676 * NB. These rules can be relaxed when checking the group count
1677 * while freeing data, as we can only allocate from a block
1678 * group after serialising against the group count, and we can
1679 * only then free after serialising in turn against that
1684 /* Update the global fs size fields */
1685 sbi
->s_groups_count
++;
1687 err
= ext4_handle_dirty_metadata(handle
, NULL
, primary
);
1688 if (unlikely(err
)) {
1689 ext4_std_error(sb
, err
);
1693 /* Update the reserved block counts only once the new group is
1695 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1696 input
->reserved_blocks
);
1698 /* Update the free space counts */
1699 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1700 EXT4_B2C(sbi
, input
->free_blocks_count
));
1701 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1702 EXT4_INODES_PER_GROUP(sb
));
1704 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1705 sbi
->s_log_groups_per_flex
) {
1706 ext4_group_t flex_group
;
1707 flex_group
= ext4_flex_group(sbi
, input
->group
);
1708 atomic_add(EXT4_B2C(sbi
, input
->free_blocks_count
),
1709 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1710 atomic_add(EXT4_INODES_PER_GROUP(sb
),
1711 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1714 ext4_handle_dirty_super(handle
, sb
);
1717 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1719 if (!err
&& primary
) {
1720 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1721 sizeof(struct ext4_super_block
));
1722 update_backups(sb
, primary
->b_blocknr
, primary
->b_data
,
1728 } /* ext4_group_add */
1731 * extend a group without checking assuming that checking has been done.
1733 static int ext4_group_extend_no_check(struct super_block
*sb
,
1734 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1736 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1740 /* We will update the superblock, one block bitmap, and
1741 * one group descriptor via ext4_group_add_blocks().
1743 handle
= ext4_journal_start_sb(sb
, 3);
1744 if (IS_ERR(handle
)) {
1745 err
= PTR_ERR(handle
);
1746 ext4_warning(sb
, "error %d on journal start", err
);
1750 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1752 ext4_warning(sb
, "error %d on journal write access", err
);
1756 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1757 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1758 o_blocks_count
+ add
);
1759 /* We add the blocks to the bitmap and set the group need init bit */
1760 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1763 ext4_handle_dirty_super(handle
, sb
);
1764 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1765 o_blocks_count
+ add
);
1767 err2
= ext4_journal_stop(handle
);
1772 if (test_opt(sb
, DEBUG
))
1773 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1774 "blocks\n", ext4_blocks_count(es
));
1775 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1776 sizeof(struct ext4_super_block
));
1782 * Extend the filesystem to the new number of blocks specified. This entry
1783 * point is only used to extend the current filesystem to the end of the last
1784 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1785 * for emergencies (because it has no dependencies on reserved blocks).
1787 * If we _really_ wanted, we could use default values to call ext4_group_add()
1788 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1789 * GDT blocks are reserved to grow to the desired size.
1791 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1792 ext4_fsblk_t n_blocks_count
)
1794 ext4_fsblk_t o_blocks_count
;
1797 struct buffer_head
*bh
;
1802 o_blocks_count
= ext4_blocks_count(es
);
1804 if (test_opt(sb
, DEBUG
))
1805 printk(KERN_DEBUG
"EXT4-fs: extending last group from %llu to %llu blocks\n",
1806 o_blocks_count
, n_blocks_count
);
1808 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1811 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1812 printk(KERN_ERR
"EXT4-fs: filesystem on %s:"
1813 " too large to resize to %llu blocks safely\n",
1814 sb
->s_id
, n_blocks_count
);
1815 if (sizeof(sector_t
) < 8)
1816 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1820 if (n_blocks_count
< o_blocks_count
) {
1821 ext4_warning(sb
, "can't shrink FS - resize aborted");
1825 /* Handle the remaining blocks in the last group only. */
1826 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1829 ext4_warning(sb
, "need to use ext2online to resize further");
1833 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1835 if (o_blocks_count
+ add
< o_blocks_count
) {
1836 ext4_warning(sb
, "blocks_count overflow");
1840 if (o_blocks_count
+ add
> n_blocks_count
)
1841 add
= n_blocks_count
- o_blocks_count
;
1843 if (o_blocks_count
+ add
< n_blocks_count
)
1844 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1845 o_blocks_count
+ add
, add
);
1847 /* See if the device is actually as big as what was requested */
1848 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1850 ext4_warning(sb
, "can't read last block, resize aborted");
1855 /* We will update the superblock, one block bitmap, and
1856 * one group descriptor via ext4_free_blocks().
1858 handle
= ext4_journal_start_sb(sb
, 3);
1859 if (IS_ERR(handle
)) {
1860 err
= PTR_ERR(handle
);
1861 ext4_warning(sb
, "error %d on journal start", err
);
1865 if ((err
= ext4_journal_get_write_access(handle
,
1866 EXT4_SB(sb
)->s_sbh
))) {
1867 ext4_warning(sb
, "error %d on journal write access", err
);
1868 ext4_journal_stop(handle
);
1871 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1872 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1873 o_blocks_count
+ add
);
1874 /* We add the blocks to the bitmap and set the group need init bit */
1875 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1876 ext4_handle_dirty_super(handle
, sb
);
1877 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1878 o_blocks_count
+ add
);
1879 err2
= ext4_journal_stop(handle
);
1886 if (test_opt(sb
, DEBUG
))
1887 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu blocks\n",
1888 ext4_blocks_count(es
));
1889 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1890 sizeof(struct ext4_super_block
));
1893 } /* ext4_group_extend */
1896 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1898 * @sb: super block of the fs to be resized
1899 * @n_blocks_count: the number of blocks resides in the resized fs
1901 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1903 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1904 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1905 struct ext4_super_block
*es
= sbi
->s_es
;
1906 struct buffer_head
*bh
;
1907 struct inode
*resize_inode
;
1908 ext4_fsblk_t o_blocks_count
;
1909 ext4_group_t o_group
;
1910 ext4_group_t n_group
;
1911 ext4_grpblk_t offset
;
1912 unsigned long n_desc_blocks
;
1913 unsigned long o_desc_blocks
;
1914 unsigned long desc_blocks
;
1915 int err
= 0, flexbg_size
= 1;
1917 o_blocks_count
= ext4_blocks_count(es
);
1919 if (test_opt(sb
, DEBUG
))
1920 printk(KERN_DEBUG
"EXT4-fs: resizing filesystem from %llu "
1921 "upto %llu blocks\n", o_blocks_count
, n_blocks_count
);
1923 if (n_blocks_count
< o_blocks_count
) {
1924 /* On-line shrinking not supported */
1925 ext4_warning(sb
, "can't shrink FS - resize aborted");
1929 if (n_blocks_count
== o_blocks_count
)
1930 /* Nothing need to do */
1933 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &offset
);
1934 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &o_group
, &offset
);
1936 n_desc_blocks
= (n_group
+ EXT4_DESC_PER_BLOCK(sb
)) /
1937 EXT4_DESC_PER_BLOCK(sb
);
1938 o_desc_blocks
= (sbi
->s_groups_count
+ EXT4_DESC_PER_BLOCK(sb
) - 1) /
1939 EXT4_DESC_PER_BLOCK(sb
);
1940 desc_blocks
= n_desc_blocks
- o_desc_blocks
;
1943 (!EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
) ||
1944 le16_to_cpu(es
->s_reserved_gdt_blocks
) < desc_blocks
)) {
1945 ext4_warning(sb
, "No reserved GDT blocks, can't resize");
1949 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1950 if (IS_ERR(resize_inode
)) {
1951 ext4_warning(sb
, "Error opening resize inode");
1952 return PTR_ERR(resize_inode
);
1955 /* See if the device is actually as big as what was requested */
1956 bh
= sb_bread(sb
, n_blocks_count
- 1);
1958 ext4_warning(sb
, "can't read last block, resize aborted");
1964 /* extend the last group */
1966 add
= EXT4_BLOCKS_PER_GROUP(sb
) - offset
;
1967 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1972 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1973 es
->s_log_groups_per_flex
)
1974 flexbg_size
= 1 << es
->s_log_groups_per_flex
;
1976 o_blocks_count
= ext4_blocks_count(es
);
1977 if (o_blocks_count
== n_blocks_count
)
1980 flex_gd
= alloc_flex_gd(flexbg_size
);
1981 if (flex_gd
== NULL
) {
1986 /* Add flex groups. Note that a regular group is a
1987 * flex group with 1 group.
1989 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
1991 ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
);
1992 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
1999 free_flex_gd(flex_gd
);
2002 if (test_opt(sb
, DEBUG
))
2003 printk(KERN_DEBUG
"EXT4-fs: resized filesystem from %llu "
2004 "upto %llu blocks\n", o_blocks_count
, n_blocks_count
);