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_atomic();
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
;
84 struct buffer_head
*bh
= NULL
;
85 ext4_grpblk_t free_blocks_count
, offset
;
88 if (group
!= sbi
->s_groups_count
) {
89 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
90 input
->group
, sbi
->s_groups_count
);
94 overhead
= ext4_group_overhead_blocks(sb
, group
);
95 metaend
= start
+ overhead
;
96 input
->free_blocks_count
= free_blocks_count
=
97 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
99 if (test_opt(sb
, DEBUG
))
100 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
101 "(%d free, %u reserved)\n",
102 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
103 "no-super", input
->group
, input
->blocks_count
,
104 free_blocks_count
, input
->reserved_blocks
);
106 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
108 ext4_warning(sb
, "Last group not full");
109 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
110 ext4_warning(sb
, "Reserved blocks too high (%u)",
111 input
->reserved_blocks
);
112 else if (free_blocks_count
< 0)
113 ext4_warning(sb
, "Bad blocks count %u",
114 input
->blocks_count
);
115 else if (!(bh
= sb_bread(sb
, end
- 1)))
116 ext4_warning(sb
, "Cannot read last block (%llu)",
118 else if (outside(input
->block_bitmap
, start
, end
))
119 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
120 (unsigned long long)input
->block_bitmap
);
121 else if (outside(input
->inode_bitmap
, start
, end
))
122 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
123 (unsigned long long)input
->inode_bitmap
);
124 else if (outside(input
->inode_table
, start
, end
) ||
125 outside(itend
- 1, start
, end
))
126 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
127 (unsigned long long)input
->inode_table
, itend
- 1);
128 else if (input
->inode_bitmap
== input
->block_bitmap
)
129 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
130 (unsigned long long)input
->block_bitmap
);
131 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
132 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
134 (unsigned long long)input
->block_bitmap
,
135 (unsigned long long)input
->inode_table
, itend
- 1);
136 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
137 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
139 (unsigned long long)input
->inode_bitmap
,
140 (unsigned long long)input
->inode_table
, itend
- 1);
141 else if (inside(input
->block_bitmap
, start
, metaend
))
142 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
143 (unsigned long long)input
->block_bitmap
,
145 else if (inside(input
->inode_bitmap
, start
, metaend
))
146 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
147 (unsigned long long)input
->inode_bitmap
,
149 else if (inside(input
->inode_table
, start
, metaend
) ||
150 inside(itend
- 1, start
, metaend
))
151 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
153 (unsigned long long)input
->inode_table
,
154 itend
- 1, start
, metaend
- 1);
163 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
166 struct ext4_new_flex_group_data
{
167 struct ext4_new_group_data
*groups
; /* new_group_data for groups
169 __u16
*bg_flags
; /* block group flags of groups
171 ext4_group_t count
; /* number of groups in @groups
176 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
179 * Returns NULL on failure otherwise address of the allocated structure.
181 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
183 struct ext4_new_flex_group_data
*flex_gd
;
185 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
189 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_flex_group_data
))
191 flex_gd
->count
= flexbg_size
;
193 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
194 flexbg_size
, GFP_NOFS
);
195 if (flex_gd
->groups
== NULL
)
198 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
199 if (flex_gd
->bg_flags
== NULL
)
205 kfree(flex_gd
->groups
);
212 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
214 kfree(flex_gd
->bg_flags
);
215 kfree(flex_gd
->groups
);
220 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
221 * and inode tables for a flex group.
223 * This function is used by 64bit-resize. Note that this function allocates
224 * group tables from the 1st group of groups contained by @flexgd, which may
225 * be a partial of a flex group.
227 * @sb: super block of fs to which the groups belongs
229 * Returns 0 on a successful allocation of the metadata blocks in the
232 static int ext4_alloc_group_tables(struct super_block
*sb
,
233 struct ext4_new_flex_group_data
*flex_gd
,
236 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
237 ext4_fsblk_t start_blk
;
238 ext4_fsblk_t last_blk
;
239 ext4_group_t src_group
;
240 ext4_group_t bb_index
= 0;
241 ext4_group_t ib_index
= 0;
242 ext4_group_t it_index
= 0;
244 ext4_group_t last_group
;
246 __u16 uninit_mask
= (flexbg_size
> 1) ? ~EXT4_BG_BLOCK_UNINIT
: ~0;
248 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
250 src_group
= group_data
[0].group
;
251 last_group
= src_group
+ flex_gd
->count
- 1;
253 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
254 (last_group
& ~(flexbg_size
- 1))));
256 group
= group_data
[0].group
;
257 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
259 start_blk
= ext4_group_first_block_no(sb
, src_group
);
260 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
262 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
264 start_blk
+= overhead
;
266 /* We collect contiguous blocks as much as possible. */
268 for (; src_group
<= last_group
; src_group
++) {
269 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
271 last_blk
+= group_data
[src_group
- group
].blocks_count
;
276 /* Allocate block bitmaps */
277 for (; bb_index
< flex_gd
->count
; bb_index
++) {
278 if (start_blk
>= last_blk
)
280 group_data
[bb_index
].block_bitmap
= start_blk
++;
281 group
= ext4_get_group_number(sb
, start_blk
- 1);
282 group
-= group_data
[0].group
;
283 group_data
[group
].free_blocks_count
--;
284 flex_gd
->bg_flags
[group
] &= uninit_mask
;
287 /* Allocate inode bitmaps */
288 for (; ib_index
< flex_gd
->count
; ib_index
++) {
289 if (start_blk
>= last_blk
)
291 group_data
[ib_index
].inode_bitmap
= start_blk
++;
292 group
= ext4_get_group_number(sb
, start_blk
- 1);
293 group
-= group_data
[0].group
;
294 group_data
[group
].free_blocks_count
--;
295 flex_gd
->bg_flags
[group
] &= uninit_mask
;
298 /* Allocate inode tables */
299 for (; it_index
< flex_gd
->count
; it_index
++) {
300 unsigned int itb
= EXT4_SB(sb
)->s_itb_per_group
;
301 ext4_fsblk_t next_group_start
;
303 if (start_blk
+ itb
> last_blk
)
305 group_data
[it_index
].inode_table
= start_blk
;
306 group
= ext4_get_group_number(sb
, start_blk
);
307 next_group_start
= ext4_group_first_block_no(sb
, group
+ 1);
308 group
-= group_data
[0].group
;
310 if (start_blk
+ itb
> next_group_start
) {
311 flex_gd
->bg_flags
[group
+ 1] &= uninit_mask
;
312 overhead
= start_blk
+ itb
- next_group_start
;
313 group_data
[group
+ 1].free_blocks_count
-= overhead
;
317 group_data
[group
].free_blocks_count
-= itb
;
318 flex_gd
->bg_flags
[group
] &= uninit_mask
;
319 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
322 if (test_opt(sb
, DEBUG
)) {
324 group
= group_data
[0].group
;
326 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
327 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
330 for (i
= 0; i
< flex_gd
->count
; i
++) {
331 printk(KERN_DEBUG
"adding %s group %u: %u "
332 "blocks (%d free)\n",
333 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
334 "no-super", group
+ i
,
335 group_data
[i
].blocks_count
,
336 group_data
[i
].free_blocks_count
);
342 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
345 struct buffer_head
*bh
;
348 bh
= sb_getblk(sb
, blk
);
350 return ERR_PTR(-ENOMEM
);
351 BUFFER_TRACE(bh
, "get_write_access");
352 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
356 memset(bh
->b_data
, 0, sb
->s_blocksize
);
357 set_buffer_uptodate(bh
);
364 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
365 * If that fails, restart the transaction & regain write access for the
366 * buffer head which is used for block_bitmap modifications.
368 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
372 if (ext4_handle_has_enough_credits(handle
, thresh
))
375 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
379 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
388 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
390 * Helper function for ext4_setup_new_group_blocks() which set .
393 * @handle: journal handle
394 * @flex_gd: flex group data
396 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
397 struct ext4_new_flex_group_data
*flex_gd
,
398 ext4_fsblk_t block
, ext4_group_t count
)
402 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
403 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
405 struct buffer_head
*bh
;
409 group
= ext4_get_group_number(sb
, block
);
410 start
= ext4_group_first_block_no(sb
, group
);
411 group
-= flex_gd
->groups
[0].group
;
413 count2
= EXT4_BLOCKS_PER_GROUP(sb
) - (block
- start
);
417 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
418 BUG_ON(flex_gd
->count
> 1);
422 err
= extend_or_restart_transaction(handle
, 1);
426 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
430 BUFFER_TRACE(bh
, "get_write_access");
431 err
= ext4_journal_get_write_access(handle
, bh
);
434 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
435 block
- start
, count2
);
436 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
438 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
448 * Set up the block and inode bitmaps, and the inode table for the new groups.
449 * This doesn't need to be part of the main transaction, since we are only
450 * changing blocks outside the actual filesystem. We still do journaling to
451 * ensure the recovery is correct in case of a failure just after resize.
452 * If any part of this fails, we simply abort the resize.
454 * setup_new_flex_group_blocks handles a flex group as follow:
455 * 1. copy super block and GDT, and initialize group tables if necessary.
456 * In this step, we only set bits in blocks bitmaps for blocks taken by
457 * super block and GDT.
458 * 2. allocate group tables in block bitmaps, that is, set bits in block
459 * bitmap for blocks taken by group tables.
461 static int setup_new_flex_group_blocks(struct super_block
*sb
,
462 struct ext4_new_flex_group_data
*flex_gd
)
464 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
467 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
468 struct ext4_super_block
*es
= sbi
->s_es
;
469 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
470 __u16
*bg_flags
= flex_gd
->bg_flags
;
472 ext4_group_t group
, count
;
473 struct buffer_head
*bh
= NULL
;
474 int reserved_gdb
, i
, j
, err
= 0, err2
;
477 BUG_ON(!flex_gd
->count
|| !group_data
||
478 group_data
[0].group
!= sbi
->s_groups_count
);
480 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
481 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
483 /* This transaction may be extended/restarted along the way */
484 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
486 return PTR_ERR(handle
);
488 group
= group_data
[0].group
;
489 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
490 unsigned long gdblocks
;
491 ext4_grpblk_t overhead
;
493 gdblocks
= ext4_bg_num_gdb(sb
, group
);
494 start
= ext4_group_first_block_no(sb
, group
);
496 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
500 ext4_group_t first_group
;
501 first_group
= ext4_meta_bg_first_group(sb
, group
);
502 if (first_group
!= group
+ 1 &&
503 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
507 block
= start
+ ext4_bg_has_super(sb
, group
);
508 /* Copy all of the GDT blocks into the backup in this group */
509 for (j
= 0; j
< gdblocks
; j
++, block
++) {
510 struct buffer_head
*gdb
;
512 ext4_debug("update backup group %#04llx\n", block
);
513 err
= extend_or_restart_transaction(handle
, 1);
517 gdb
= sb_getblk(sb
, block
);
518 if (unlikely(!gdb
)) {
523 BUFFER_TRACE(gdb
, "get_write_access");
524 err
= ext4_journal_get_write_access(handle
, gdb
);
529 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
531 set_buffer_uptodate(gdb
);
533 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
541 /* Zero out all of the reserved backup group descriptor
544 if (ext4_bg_has_super(sb
, group
)) {
545 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
546 reserved_gdb
, GFP_NOFS
);
552 /* Initialize group tables of the grop @group */
553 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
556 /* Zero out all of the inode table blocks */
557 block
= group_data
[i
].inode_table
;
558 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
559 block
, sbi
->s_itb_per_group
);
560 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
566 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
569 /* Initialize block bitmap of the @group */
570 block
= group_data
[i
].block_bitmap
;
571 err
= extend_or_restart_transaction(handle
, 1);
575 bh
= bclean(handle
, sb
, block
);
580 overhead
= ext4_group_overhead_blocks(sb
, group
);
582 ext4_debug("mark backup superblock %#04llx (+0)\n",
584 ext4_set_bits(bh
->b_data
, 0, overhead
);
586 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
587 sb
->s_blocksize
* 8, bh
->b_data
);
588 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
594 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
597 /* Initialize inode bitmap of the @group */
598 block
= group_data
[i
].inode_bitmap
;
599 err
= extend_or_restart_transaction(handle
, 1);
602 /* Mark unused entries in inode bitmap used */
603 bh
= bclean(handle
, sb
, block
);
609 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
610 sb
->s_blocksize
* 8, bh
->b_data
);
611 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
618 /* Mark group tables in block bitmap */
619 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
620 count
= group_table_count
[j
];
621 start
= (&group_data
[0].block_bitmap
)[j
];
623 for (i
= 1; i
< flex_gd
->count
; i
++) {
624 block
+= group_table_count
[j
];
625 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
626 count
+= group_table_count
[j
];
629 err
= set_flexbg_block_bitmap(sb
, handle
,
630 flex_gd
, start
, count
);
633 count
= group_table_count
[j
];
634 start
= (&group_data
[i
].block_bitmap
)[j
];
639 err
= set_flexbg_block_bitmap(sb
, handle
,
640 flex_gd
, start
, count
);
648 err2
= ext4_journal_stop(handle
);
656 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
657 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
658 * calling this for the first time. In a sparse filesystem it will be the
659 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
660 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
662 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
663 unsigned *five
, unsigned *seven
)
665 unsigned *min
= three
;
669 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
670 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
692 * Check that all of the backup GDT blocks are held in the primary GDT block.
693 * It is assumed that they are stored in group order. Returns the number of
694 * groups in current filesystem that have BACKUPS, or -ve error code.
696 static int verify_reserved_gdb(struct super_block
*sb
,
698 struct buffer_head
*primary
)
700 const ext4_fsblk_t blk
= primary
->b_blocknr
;
705 __le32
*p
= (__le32
*)primary
->b_data
;
708 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
709 if (le32_to_cpu(*p
++) !=
710 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
711 ext4_warning(sb
, "reserved GDT %llu"
712 " missing grp %d (%llu)",
715 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
719 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
727 * Called when we need to bring a reserved group descriptor table block into
728 * use from the resize inode. The primary copy of the new GDT block currently
729 * is an indirect block (under the double indirect block in the resize inode).
730 * The new backup GDT blocks will be stored as leaf blocks in this indirect
731 * block, in group order. Even though we know all the block numbers we need,
732 * we check to ensure that the resize inode has actually reserved these blocks.
734 * Don't need to update the block bitmaps because the blocks are still in use.
736 * We get all of the error cases out of the way, so that we are sure to not
737 * fail once we start modifying the data on disk, because JBD has no rollback.
739 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
742 struct super_block
*sb
= inode
->i_sb
;
743 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
744 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
745 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
746 struct buffer_head
**o_group_desc
, **n_group_desc
;
747 struct buffer_head
*dind
;
748 struct buffer_head
*gdb_bh
;
750 struct ext4_iloc iloc
;
754 if (test_opt(sb
, DEBUG
))
756 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
760 * If we are not using the primary superblock/GDT copy don't resize,
761 * because the user tools have no way of handling this. Probably a
762 * bad time to do it anyways.
764 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
765 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
766 ext4_warning(sb
, "won't resize using backup superblock at %llu",
767 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
771 gdb_bh
= sb_bread(sb
, gdblock
);
775 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
781 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
782 dind
= sb_bread(sb
, le32_to_cpu(*data
));
788 data
= (__le32
*)dind
->b_data
;
789 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
790 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
796 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
797 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
801 BUFFER_TRACE(gdb_bh
, "get_write_access");
802 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
806 BUFFER_TRACE(dind
, "get_write_access");
807 err
= ext4_journal_get_write_access(handle
, dind
);
809 ext4_std_error(sb
, err
);
811 /* ext4_reserve_inode_write() gets a reference on the iloc */
812 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
816 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
817 sizeof(struct buffer_head
*),
821 ext4_warning(sb
, "not enough memory for %lu groups",
827 * Finally, we have all of the possible failures behind us...
829 * Remove new GDT block from inode double-indirect block and clear out
830 * the new GDT block for use (which also "frees" the backup GDT blocks
831 * from the reserved inode). We don't need to change the bitmaps for
832 * these blocks, because they are marked as in-use from being in the
833 * reserved inode, and will become GDT blocks (primary and backup).
835 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
836 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
838 ext4_std_error(sb
, err
);
841 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
842 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
843 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
844 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
846 ext4_std_error(sb
, err
);
851 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
852 memcpy(n_group_desc
, o_group_desc
,
853 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
854 n_group_desc
[gdb_num
] = gdb_bh
;
855 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
856 EXT4_SB(sb
)->s_gdb_count
++;
857 ext4_kvfree(o_group_desc
);
859 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
860 err
= ext4_handle_dirty_super(handle
, sb
);
862 ext4_std_error(sb
, err
);
867 ext4_kvfree(n_group_desc
);
874 ext4_debug("leaving with error %d\n", err
);
879 * add_new_gdb_meta_bg is the sister of add_new_gdb.
881 static int add_new_gdb_meta_bg(struct super_block
*sb
,
882 handle_t
*handle
, ext4_group_t group
) {
883 ext4_fsblk_t gdblock
;
884 struct buffer_head
*gdb_bh
;
885 struct buffer_head
**o_group_desc
, **n_group_desc
;
886 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
889 gdblock
= ext4_meta_bg_first_block_no(sb
, group
) +
890 ext4_bg_has_super(sb
, group
);
891 gdb_bh
= sb_bread(sb
, gdblock
);
894 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
895 sizeof(struct buffer_head
*),
899 ext4_warning(sb
, "not enough memory for %lu groups",
904 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
905 memcpy(n_group_desc
, o_group_desc
,
906 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
907 n_group_desc
[gdb_num
] = gdb_bh
;
908 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
909 EXT4_SB(sb
)->s_gdb_count
++;
910 ext4_kvfree(o_group_desc
);
911 BUFFER_TRACE(gdb_bh
, "get_write_access");
912 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
919 * Called when we are adding a new group which has a backup copy of each of
920 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
921 * We need to add these reserved backup GDT blocks to the resize inode, so
922 * that they are kept for future resizing and not allocated to files.
924 * Each reserved backup GDT block will go into a different indirect block.
925 * The indirect blocks are actually the primary reserved GDT blocks,
926 * so we know in advance what their block numbers are. We only get the
927 * double-indirect block to verify it is pointing to the primary reserved
928 * GDT blocks so we don't overwrite a data block by accident. The reserved
929 * backup GDT blocks are stored in their reserved primary GDT block.
931 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
934 struct super_block
*sb
= inode
->i_sb
;
935 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
936 struct buffer_head
**primary
;
937 struct buffer_head
*dind
;
938 struct ext4_iloc iloc
;
945 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
949 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
950 dind
= sb_bread(sb
, le32_to_cpu(*data
));
956 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
957 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
958 EXT4_ADDR_PER_BLOCK(sb
));
959 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
961 /* Get each reserved primary GDT block and verify it holds backups */
962 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
963 if (le32_to_cpu(*data
) != blk
) {
964 ext4_warning(sb
, "reserved block %llu"
965 " not at offset %ld",
967 (long)(data
- (__le32
*)dind
->b_data
));
971 primary
[res
] = sb_bread(sb
, blk
);
976 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
978 brelse(primary
[res
]);
983 data
= (__le32
*)dind
->b_data
;
986 for (i
= 0; i
< reserved_gdb
; i
++) {
987 BUFFER_TRACE(primary
[i
], "get_write_access");
988 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
])))
992 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
996 * Finally we can add each of the reserved backup GDT blocks from
997 * the new group to its reserved primary GDT block.
999 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1000 for (i
= 0; i
< reserved_gdb
; i
++) {
1002 data
= (__le32
*)primary
[i
]->b_data
;
1003 /* printk("reserving backup %lu[%u] = %lu\n",
1004 primary[i]->b_blocknr, gdbackups,
1005 blk + primary[i]->b_blocknr); */
1006 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1007 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1011 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
1012 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1016 brelse(primary
[res
]);
1026 * Update the backup copies of the ext4 metadata. These don't need to be part
1027 * of the main resize transaction, because e2fsck will re-write them if there
1028 * is a problem (basically only OOM will cause a problem). However, we
1029 * _should_ update the backups if possible, in case the primary gets trashed
1030 * for some reason and we need to run e2fsck from a backup superblock. The
1031 * important part is that the new block and inode counts are in the backup
1032 * superblocks, and the location of the new group metadata in the GDT backups.
1034 * We do not need take the s_resize_lock for this, because these
1035 * blocks are not otherwise touched by the filesystem code when it is
1036 * mounted. We don't need to worry about last changing from
1037 * sbi->s_groups_count, because the worst that can happen is that we
1038 * do not copy the full number of backups at this time. The resize
1039 * which changed s_groups_count will backup again.
1041 static void update_backups(struct super_block
*sb
, int blk_off
, char *data
,
1042 int size
, int meta_bg
)
1044 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1046 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1050 ext4_group_t group
= 0;
1051 int rest
= sb
->s_blocksize
- size
;
1055 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
1056 if (IS_ERR(handle
)) {
1058 err
= PTR_ERR(handle
);
1063 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1064 last
= sbi
->s_groups_count
;
1066 group
= ext4_meta_bg_first_group(sb
, group
) + 1;
1067 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1070 while (group
< sbi
->s_groups_count
) {
1071 struct buffer_head
*bh
;
1072 ext4_fsblk_t backup_block
;
1074 /* Out of journal space, and can't get more - abort - so sad */
1075 if (ext4_handle_valid(handle
) &&
1076 handle
->h_buffer_credits
== 0 &&
1077 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1078 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1082 backup_block
= group
* bpg
+ blk_off
;
1084 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1085 ext4_bg_has_super(sb
, group
));
1087 bh
= sb_getblk(sb
, backup_block
);
1088 if (unlikely(!bh
)) {
1092 ext4_debug("update metadata backup %llu(+%llu)\n",
1093 backup_block
, backup_block
-
1094 ext4_group_first_block_no(sb
, group
));
1095 BUFFER_TRACE(bh
, "get_write_access");
1096 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
1099 memcpy(bh
->b_data
, data
, size
);
1101 memset(bh
->b_data
+ size
, 0, rest
);
1102 set_buffer_uptodate(bh
);
1104 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1106 ext4_std_error(sb
, err
);
1110 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1111 else if (group
== last
)
1116 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1120 * Ugh! Need to have e2fsck write the backup copies. It is too
1121 * late to revert the resize, we shouldn't fail just because of
1122 * the backup copies (they are only needed in case of corruption).
1124 * However, if we got here we have a journal problem too, so we
1125 * can't really start a transaction to mark the superblock.
1126 * Chicken out and just set the flag on the hope it will be written
1127 * to disk, and if not - we will simply wait until next fsck.
1131 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1132 "forcing fsck on next reboot", group
, err
);
1133 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1134 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1135 mark_buffer_dirty(sbi
->s_sbh
);
1140 * ext4_add_new_descs() adds @count group descriptor of groups
1141 * starting at @group
1143 * @handle: journal handle
1145 * @group: the group no. of the first group desc to be added
1146 * @resize_inode: the resize inode
1147 * @count: number of group descriptors to be added
1149 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1150 ext4_group_t group
, struct inode
*resize_inode
,
1153 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1154 struct ext4_super_block
*es
= sbi
->s_es
;
1155 struct buffer_head
*gdb_bh
;
1156 int i
, gdb_off
, gdb_num
, err
= 0;
1159 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1160 for (i
= 0; i
< count
; i
++, group
++) {
1161 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1162 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1164 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1165 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1168 * We will only either add reserved group blocks to a backup group
1169 * or remove reserved blocks for the first group in a new group block.
1170 * Doing both would be mean more complex code, and sane people don't
1171 * use non-sparse filesystems anymore. This is already checked above.
1174 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1175 BUFFER_TRACE(gdb_bh
, "get_write_access");
1176 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1178 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1179 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1180 } else if (meta_bg
!= 0) {
1181 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1183 err
= add_new_gdb(handle
, resize_inode
, group
);
1191 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1193 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1196 if (!bh_uptodate_or_lock(bh
)) {
1197 if (bh_submit_read(bh
) < 0) {
1206 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1208 struct ext4_group_desc
*gdp
,
1209 struct ext4_new_group_data
*group_data
)
1211 struct buffer_head
*bh
;
1213 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1214 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
1217 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1220 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1221 EXT4_INODES_PER_GROUP(sb
) / 8);
1224 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1227 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1234 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1236 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1237 struct ext4_new_flex_group_data
*flex_gd
)
1239 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1240 struct ext4_group_desc
*gdp
;
1241 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1242 struct buffer_head
*gdb_bh
;
1244 __u16
*bg_flags
= flex_gd
->bg_flags
;
1245 int i
, gdb_off
, gdb_num
, err
= 0;
1248 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1249 group
= group_data
->group
;
1251 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1252 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1255 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1257 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1258 /* Update group descriptor block for new group */
1259 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1260 gdb_off
* EXT4_DESC_SIZE(sb
));
1262 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1263 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1264 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1265 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1267 ext4_std_error(sb
, err
);
1271 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1272 ext4_free_group_clusters_set(sb
, gdp
,
1273 EXT4_NUM_B2C(sbi
, group_data
->free_blocks_count
));
1274 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1275 if (ext4_has_group_desc_csum(sb
))
1276 ext4_itable_unused_set(sb
, gdp
,
1277 EXT4_INODES_PER_GROUP(sb
));
1278 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1279 ext4_group_desc_csum_set(sb
, group
, gdp
);
1281 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1282 if (unlikely(err
)) {
1283 ext4_std_error(sb
, err
);
1288 * We can allocate memory for mb_alloc based on the new group
1291 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1299 * ext4_update_super() updates the super block so that the newly added
1300 * groups can be seen by the filesystem.
1303 * @flex_gd: new added groups
1305 static void ext4_update_super(struct super_block
*sb
,
1306 struct ext4_new_flex_group_data
*flex_gd
)
1308 ext4_fsblk_t blocks_count
= 0;
1309 ext4_fsblk_t free_blocks
= 0;
1310 ext4_fsblk_t reserved_blocks
= 0;
1311 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1312 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1313 struct ext4_super_block
*es
= sbi
->s_es
;
1316 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1318 * Make the new blocks and inodes valid next. We do this before
1319 * increasing the group count so that once the group is enabled,
1320 * all of its blocks and inodes are already valid.
1322 * We always allocate group-by-group, then block-by-block or
1323 * inode-by-inode within a group, so enabling these
1324 * blocks/inodes before the group is live won't actually let us
1325 * allocate the new space yet.
1327 for (i
= 0; i
< flex_gd
->count
; i
++) {
1328 blocks_count
+= group_data
[i
].blocks_count
;
1329 free_blocks
+= group_data
[i
].free_blocks_count
;
1332 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1333 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1334 reserved_blocks
*= blocks_count
;
1335 do_div(reserved_blocks
, 100);
1337 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1338 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1339 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1341 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1344 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1346 * We need to protect s_groups_count against other CPUs seeing
1347 * inconsistent state in the superblock.
1349 * The precise rules we use are:
1351 * * Writers must perform a smp_wmb() after updating all
1352 * dependent data and before modifying the groups count
1354 * * Readers must perform an smp_rmb() after reading the groups
1355 * count and before reading any dependent data.
1357 * NB. These rules can be relaxed when checking the group count
1358 * while freeing data, as we can only allocate from a block
1359 * group after serialising against the group count, and we can
1360 * only then free after serialising in turn against that
1365 /* Update the global fs size fields */
1366 sbi
->s_groups_count
+= flex_gd
->count
;
1367 sbi
->s_blockfile_groups
= min_t(ext4_group_t
, sbi
->s_groups_count
,
1368 (EXT4_MAX_BLOCK_FILE_PHYS
/ EXT4_BLOCKS_PER_GROUP(sb
)));
1370 /* Update the reserved block counts only once the new group is
1372 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1375 /* Update the free space counts */
1376 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1377 EXT4_NUM_B2C(sbi
, free_blocks
));
1378 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1379 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1381 ext4_debug("free blocks count %llu",
1382 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1383 if (EXT4_HAS_INCOMPAT_FEATURE(sb
,
1384 EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
1385 sbi
->s_log_groups_per_flex
) {
1386 ext4_group_t flex_group
;
1387 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1388 atomic64_add(EXT4_NUM_B2C(sbi
, free_blocks
),
1389 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1390 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1391 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1395 * Update the fs overhead information
1397 ext4_calculate_overhead(sb
);
1399 if (test_opt(sb
, DEBUG
))
1400 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1401 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1402 blocks_count
, free_blocks
, reserved_blocks
);
1405 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1406 * _before_ we start modifying the filesystem, because we cannot abort the
1407 * transaction and not have it write the data to disk.
1409 static int ext4_flex_group_add(struct super_block
*sb
,
1410 struct inode
*resize_inode
,
1411 struct ext4_new_flex_group_data
*flex_gd
)
1413 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1414 struct ext4_super_block
*es
= sbi
->s_es
;
1415 ext4_fsblk_t o_blocks_count
;
1419 unsigned reserved_gdb
;
1420 int err
= 0, err2
= 0, credit
;
1422 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1424 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1425 o_blocks_count
= ext4_blocks_count(es
);
1426 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1429 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1433 * We will always be modifying at least the superblock and GDT
1434 * block. If we are adding a group past the last current GDT block,
1435 * we will also modify the inode and the dindirect block. If we
1436 * are adding a group with superblock/GDT backups we will also
1437 * modify each of the reserved GDT dindirect blocks.
1439 credit
= flex_gd
->count
* 4 + reserved_gdb
;
1440 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credit
);
1441 if (IS_ERR(handle
)) {
1442 err
= PTR_ERR(handle
);
1446 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1447 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1451 group
= flex_gd
->groups
[0].group
;
1452 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1453 err
= ext4_add_new_descs(handle
, sb
, group
,
1454 resize_inode
, flex_gd
->count
);
1458 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1462 ext4_update_super(sb
, flex_gd
);
1464 err
= ext4_handle_dirty_super(handle
, sb
);
1467 err2
= ext4_journal_stop(handle
);
1472 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1473 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1474 EXT4_DESC_PER_BLOCK(sb
));
1475 int meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
,
1476 EXT4_FEATURE_INCOMPAT_META_BG
);
1477 sector_t old_gdb
= 0;
1479 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1480 sizeof(struct ext4_super_block
), 0);
1481 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1482 struct buffer_head
*gdb_bh
;
1484 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1485 if (old_gdb
== gdb_bh
->b_blocknr
)
1487 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1488 gdb_bh
->b_size
, meta_bg
);
1489 old_gdb
= gdb_bh
->b_blocknr
;
1496 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1497 struct ext4_new_flex_group_data
*flex_gd
,
1498 ext4_fsblk_t n_blocks_count
,
1499 unsigned long flexbg_size
)
1501 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1502 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1503 ext4_fsblk_t o_blocks_count
;
1504 ext4_group_t n_group
;
1506 ext4_group_t last_group
;
1508 ext4_grpblk_t blocks_per_group
;
1511 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1513 o_blocks_count
= ext4_blocks_count(es
);
1515 if (o_blocks_count
== n_blocks_count
)
1518 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1520 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1522 last_group
= group
| (flexbg_size
- 1);
1523 if (last_group
> n_group
)
1524 last_group
= n_group
;
1526 flex_gd
->count
= last_group
- group
+ 1;
1528 for (i
= 0; i
< flex_gd
->count
; i
++) {
1531 group_data
[i
].group
= group
+ i
;
1532 group_data
[i
].blocks_count
= blocks_per_group
;
1533 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1534 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1535 if (ext4_has_group_desc_csum(sb
)) {
1536 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1537 EXT4_BG_INODE_UNINIT
;
1538 if (!test_opt(sb
, INIT_INODE_TABLE
))
1539 flex_gd
->bg_flags
[i
] |= EXT4_BG_INODE_ZEROED
;
1541 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1544 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1545 /* We need to initialize block bitmap of last group. */
1546 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1548 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1549 group_data
[i
- 1].blocks_count
= last
+ 1;
1550 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1557 /* Add group descriptor data to an existing or new group descriptor block.
1558 * Ensure we handle all possible error conditions _before_ we start modifying
1559 * the filesystem, because we cannot abort the transaction and not have it
1560 * write the data to disk.
1562 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1563 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1565 * We only need to hold the superblock lock while we are actually adding
1566 * in the new group's counts to the superblock. Prior to that we have
1567 * not really "added" the group at all. We re-check that we are still
1568 * adding in the last group in case things have changed since verifying.
1570 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1572 struct ext4_new_flex_group_data flex_gd
;
1573 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1574 struct ext4_super_block
*es
= sbi
->s_es
;
1575 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1576 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1577 struct inode
*inode
= NULL
;
1582 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1584 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
1585 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
1586 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1590 if (ext4_blocks_count(es
) + input
->blocks_count
<
1591 ext4_blocks_count(es
)) {
1592 ext4_warning(sb
, "blocks_count overflow");
1596 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1597 le32_to_cpu(es
->s_inodes_count
)) {
1598 ext4_warning(sb
, "inodes_count overflow");
1602 if (reserved_gdb
|| gdb_off
== 0) {
1603 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
1604 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
1605 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1607 "No reserved GDT blocks, can't resize");
1610 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1611 if (IS_ERR(inode
)) {
1612 ext4_warning(sb
, "Error opening resize inode");
1613 return PTR_ERR(inode
);
1618 err
= verify_group_input(sb
, input
);
1622 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1626 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1631 flex_gd
.groups
= input
;
1632 flex_gd
.bg_flags
= &bg_flags
;
1633 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1637 } /* ext4_group_add */
1640 * extend a group without checking assuming that checking has been done.
1642 static int ext4_group_extend_no_check(struct super_block
*sb
,
1643 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1645 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1649 /* We will update the superblock, one block bitmap, and
1650 * one group descriptor via ext4_group_add_blocks().
1652 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, 3);
1653 if (IS_ERR(handle
)) {
1654 err
= PTR_ERR(handle
);
1655 ext4_warning(sb
, "error %d on journal start", err
);
1659 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1660 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1662 ext4_warning(sb
, "error %d on journal write access", err
);
1666 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1667 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1668 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1669 o_blocks_count
+ add
);
1670 /* We add the blocks to the bitmap and set the group need init bit */
1671 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1674 ext4_handle_dirty_super(handle
, sb
);
1675 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1676 o_blocks_count
+ add
);
1678 err2
= ext4_journal_stop(handle
);
1683 if (test_opt(sb
, DEBUG
))
1684 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1685 "blocks\n", ext4_blocks_count(es
));
1686 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
,
1687 (char *)es
, sizeof(struct ext4_super_block
), 0);
1693 * Extend the filesystem to the new number of blocks specified. This entry
1694 * point is only used to extend the current filesystem to the end of the last
1695 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1696 * for emergencies (because it has no dependencies on reserved blocks).
1698 * If we _really_ wanted, we could use default values to call ext4_group_add()
1699 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1700 * GDT blocks are reserved to grow to the desired size.
1702 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1703 ext4_fsblk_t n_blocks_count
)
1705 ext4_fsblk_t o_blocks_count
;
1708 struct buffer_head
*bh
;
1712 o_blocks_count
= ext4_blocks_count(es
);
1714 if (test_opt(sb
, DEBUG
))
1715 ext4_msg(sb
, KERN_DEBUG
,
1716 "extending last group from %llu to %llu blocks",
1717 o_blocks_count
, n_blocks_count
);
1719 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1722 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1723 ext4_msg(sb
, KERN_ERR
,
1724 "filesystem too large to resize to %llu blocks safely",
1726 if (sizeof(sector_t
) < 8)
1727 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1731 if (n_blocks_count
< o_blocks_count
) {
1732 ext4_warning(sb
, "can't shrink FS - resize aborted");
1736 /* Handle the remaining blocks in the last group only. */
1737 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1740 ext4_warning(sb
, "need to use ext2online to resize further");
1744 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1746 if (o_blocks_count
+ add
< o_blocks_count
) {
1747 ext4_warning(sb
, "blocks_count overflow");
1751 if (o_blocks_count
+ add
> n_blocks_count
)
1752 add
= n_blocks_count
- o_blocks_count
;
1754 if (o_blocks_count
+ add
< n_blocks_count
)
1755 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1756 o_blocks_count
+ add
, add
);
1758 /* See if the device is actually as big as what was requested */
1759 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1761 ext4_warning(sb
, "can't read last block, resize aborted");
1766 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1768 } /* ext4_group_extend */
1771 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1773 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1777 * Release the resize inode and drop the resize_inode feature if there
1778 * are no more reserved gdt blocks, and then convert the file system
1781 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1784 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1785 struct ext4_super_block
*es
= sbi
->s_es
;
1786 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1788 int i
, ret
, err
= 0;
1791 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1793 if (es
->s_reserved_gdt_blocks
) {
1794 ext4_error(sb
, "Unexpected non-zero "
1795 "s_reserved_gdt_blocks");
1799 /* Do a quick sanity check of the resize inode */
1800 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
- 9))
1801 goto invalid_resize_inode
;
1802 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1803 if (i
== EXT4_DIND_BLOCK
) {
1807 goto invalid_resize_inode
;
1810 goto invalid_resize_inode
;
1812 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1815 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credits
);
1817 return PTR_ERR(handle
);
1819 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1820 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1824 EXT4_CLEAR_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
);
1825 EXT4_SET_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1826 sbi
->s_es
->s_first_meta_bg
=
1827 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1829 err
= ext4_handle_dirty_super(handle
, sb
);
1831 ext4_std_error(sb
, err
);
1836 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1837 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1838 EXT4_FREE_BLOCKS_METADATA
|
1839 EXT4_FREE_BLOCKS_FORGET
);
1840 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1841 inode
->i_blocks
= 0;
1843 err
= ext4_mark_inode_dirty(handle
, inode
);
1845 ext4_std_error(sb
, err
);
1849 ret
= ext4_journal_stop(handle
);
1854 invalid_resize_inode
:
1855 ext4_error(sb
, "corrupted/inconsistent resize inode");
1860 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1862 * @sb: super block of the fs to be resized
1863 * @n_blocks_count: the number of blocks resides in the resized fs
1865 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1867 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1868 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1869 struct ext4_super_block
*es
= sbi
->s_es
;
1870 struct buffer_head
*bh
;
1871 struct inode
*resize_inode
= NULL
;
1872 ext4_grpblk_t add
, offset
;
1873 unsigned long n_desc_blocks
;
1874 unsigned long o_desc_blocks
;
1875 ext4_group_t o_group
;
1876 ext4_group_t n_group
;
1877 ext4_fsblk_t o_blocks_count
;
1878 ext4_fsblk_t n_blocks_count_retry
= 0;
1879 unsigned long last_update_time
= 0;
1880 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1883 /* See if the device is actually as big as what was requested */
1884 bh
= sb_bread(sb
, n_blocks_count
- 1);
1886 ext4_warning(sb
, "can't read last block, resize aborted");
1892 o_blocks_count
= ext4_blocks_count(es
);
1894 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1895 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1897 if (n_blocks_count
< o_blocks_count
) {
1898 /* On-line shrinking not supported */
1899 ext4_warning(sb
, "can't shrink FS - resize aborted");
1903 if (n_blocks_count
== o_blocks_count
)
1904 /* Nothing need to do */
1907 n_group
= ext4_get_group_number(sb
, n_blocks_count
- 1);
1908 if (n_group
> (0xFFFFFFFFUL
/ EXT4_INODES_PER_GROUP(sb
))) {
1909 ext4_warning(sb
, "resize would cause inodes_count overflow");
1912 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1914 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1915 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1917 meta_bg
= EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_META_BG
);
1919 if (EXT4_HAS_COMPAT_FEATURE(sb
, EXT4_FEATURE_COMPAT_RESIZE_INODE
)) {
1921 ext4_error(sb
, "resize_inode and meta_bg enabled "
1925 if (n_desc_blocks
> o_desc_blocks
+
1926 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1927 n_blocks_count_retry
= n_blocks_count
;
1928 n_desc_blocks
= o_desc_blocks
+
1929 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1930 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1931 n_blocks_count
= n_group
* EXT4_BLOCKS_PER_GROUP(sb
);
1932 n_group
--; /* set to last group number */
1936 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1937 if (IS_ERR(resize_inode
)) {
1938 ext4_warning(sb
, "Error opening resize inode");
1939 return PTR_ERR(resize_inode
);
1943 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
1944 err
= ext4_convert_meta_bg(sb
, resize_inode
);
1949 resize_inode
= NULL
;
1951 if (n_blocks_count_retry
) {
1952 n_blocks_count
= n_blocks_count_retry
;
1953 n_blocks_count_retry
= 0;
1958 /* extend the last group */
1959 if (n_group
== o_group
)
1960 add
= n_blocks_count
- o_blocks_count
;
1962 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1964 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1969 if (ext4_blocks_count(es
) == n_blocks_count
)
1972 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
1976 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
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 if (jiffies
- last_update_time
> HZ
* 10) {
1992 if (last_update_time
)
1993 ext4_msg(sb
, KERN_INFO
,
1994 "resized to %llu blocks",
1995 ext4_blocks_count(es
));
1996 last_update_time
= jiffies
;
1998 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
2000 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
2005 if (!err
&& n_blocks_count_retry
) {
2006 n_blocks_count
= n_blocks_count_retry
;
2007 n_blocks_count_retry
= 0;
2008 free_flex_gd(flex_gd
);
2015 free_flex_gd(flex_gd
);
2016 if (resize_inode
!= NULL
)
2018 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu", n_blocks_count
);