1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/resize.c
5 * Support for resizing an ext4 filesystem while it is mounted.
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
9 * This could probably be made into a module, because it is not often in use.
15 #include <linux/errno.h>
16 #include <linux/slab.h>
18 #include "ext4_jbd2.h"
20 int ext4_resize_begin(struct super_block
*sb
)
22 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
25 if (!capable(CAP_SYS_RESOURCE
))
29 * If we are not using the primary superblock/GDT copy don't resize,
30 * because the user tools have no way of handling this. Probably a
31 * bad time to do it anyways.
33 if (EXT4_B2C(sbi
, sbi
->s_sbh
->b_blocknr
) !=
34 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
35 ext4_warning(sb
, "won't resize using backup superblock at %llu",
36 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
41 * We are not allowed to do online-resizing on a filesystem mounted
42 * with error, because it can destroy the filesystem easily.
44 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
45 ext4_warning(sb
, "There are errors in the filesystem, "
46 "so online resizing is not allowed");
50 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING
,
51 &EXT4_SB(sb
)->s_ext4_flags
))
57 void ext4_resize_end(struct super_block
*sb
)
59 clear_bit_unlock(EXT4_FLAGS_RESIZING
, &EXT4_SB(sb
)->s_ext4_flags
);
60 smp_mb__after_atomic();
63 static ext4_group_t
ext4_meta_bg_first_group(struct super_block
*sb
,
65 return (group
>> EXT4_DESC_PER_BLOCK_BITS(sb
)) <<
66 EXT4_DESC_PER_BLOCK_BITS(sb
);
69 static ext4_fsblk_t
ext4_meta_bg_first_block_no(struct super_block
*sb
,
71 group
= ext4_meta_bg_first_group(sb
, group
);
72 return ext4_group_first_block_no(sb
, group
);
75 static ext4_grpblk_t
ext4_group_overhead_blocks(struct super_block
*sb
,
77 ext4_grpblk_t overhead
;
78 overhead
= ext4_bg_num_gdb(sb
, group
);
79 if (ext4_bg_has_super(sb
, group
))
81 le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
85 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
86 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
88 static int verify_group_input(struct super_block
*sb
,
89 struct ext4_new_group_data
*input
)
91 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
92 struct ext4_super_block
*es
= sbi
->s_es
;
93 ext4_fsblk_t start
= ext4_blocks_count(es
);
94 ext4_fsblk_t end
= start
+ input
->blocks_count
;
95 ext4_group_t group
= input
->group
;
96 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
99 struct buffer_head
*bh
= NULL
;
100 ext4_grpblk_t free_blocks_count
, offset
;
103 if (group
!= sbi
->s_groups_count
) {
104 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
105 input
->group
, sbi
->s_groups_count
);
109 overhead
= ext4_group_overhead_blocks(sb
, group
);
110 metaend
= start
+ overhead
;
111 input
->free_clusters_count
= free_blocks_count
=
112 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
114 if (test_opt(sb
, DEBUG
))
115 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
116 "(%d free, %u reserved)\n",
117 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
118 "no-super", input
->group
, input
->blocks_count
,
119 free_blocks_count
, input
->reserved_blocks
);
121 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
123 ext4_warning(sb
, "Last group not full");
124 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
125 ext4_warning(sb
, "Reserved blocks too high (%u)",
126 input
->reserved_blocks
);
127 else if (free_blocks_count
< 0)
128 ext4_warning(sb
, "Bad blocks count %u",
129 input
->blocks_count
);
130 else if (!(bh
= sb_bread(sb
, end
- 1)))
131 ext4_warning(sb
, "Cannot read last block (%llu)",
133 else if (outside(input
->block_bitmap
, start
, end
))
134 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
135 (unsigned long long)input
->block_bitmap
);
136 else if (outside(input
->inode_bitmap
, start
, end
))
137 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
138 (unsigned long long)input
->inode_bitmap
);
139 else if (outside(input
->inode_table
, start
, end
) ||
140 outside(itend
- 1, start
, end
))
141 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
142 (unsigned long long)input
->inode_table
, itend
- 1);
143 else if (input
->inode_bitmap
== input
->block_bitmap
)
144 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
145 (unsigned long long)input
->block_bitmap
);
146 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
147 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
149 (unsigned long long)input
->block_bitmap
,
150 (unsigned long long)input
->inode_table
, itend
- 1);
151 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
152 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
154 (unsigned long long)input
->inode_bitmap
,
155 (unsigned long long)input
->inode_table
, itend
- 1);
156 else if (inside(input
->block_bitmap
, start
, metaend
))
157 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
158 (unsigned long long)input
->block_bitmap
,
160 else if (inside(input
->inode_bitmap
, start
, metaend
))
161 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
162 (unsigned long long)input
->inode_bitmap
,
164 else if (inside(input
->inode_table
, start
, metaend
) ||
165 inside(itend
- 1, start
, metaend
))
166 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
168 (unsigned long long)input
->inode_table
,
169 itend
- 1, start
, metaend
- 1);
178 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
181 struct ext4_new_flex_group_data
{
182 struct ext4_new_group_data
*groups
; /* new_group_data for groups
184 __u16
*bg_flags
; /* block group flags of groups
186 ext4_group_t count
; /* number of groups in @groups
191 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
194 * Returns NULL on failure otherwise address of the allocated structure.
196 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
198 struct ext4_new_flex_group_data
*flex_gd
;
200 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
204 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_group_data
))
206 flex_gd
->count
= flexbg_size
;
208 flex_gd
->groups
= kmalloc_array(flexbg_size
,
209 sizeof(struct ext4_new_group_data
),
211 if (flex_gd
->groups
== NULL
)
214 flex_gd
->bg_flags
= kmalloc_array(flexbg_size
, sizeof(__u16
),
216 if (flex_gd
->bg_flags
== NULL
)
222 kfree(flex_gd
->groups
);
229 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
231 kfree(flex_gd
->bg_flags
);
232 kfree(flex_gd
->groups
);
237 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
238 * and inode tables for a flex group.
240 * This function is used by 64bit-resize. Note that this function allocates
241 * group tables from the 1st group of groups contained by @flexgd, which may
242 * be a partial of a flex group.
244 * @sb: super block of fs to which the groups belongs
246 * Returns 0 on a successful allocation of the metadata blocks in the
249 static int ext4_alloc_group_tables(struct super_block
*sb
,
250 struct ext4_new_flex_group_data
*flex_gd
,
253 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
254 ext4_fsblk_t start_blk
;
255 ext4_fsblk_t last_blk
;
256 ext4_group_t src_group
;
257 ext4_group_t bb_index
= 0;
258 ext4_group_t ib_index
= 0;
259 ext4_group_t it_index
= 0;
261 ext4_group_t last_group
;
263 __u16 uninit_mask
= (flexbg_size
> 1) ? ~EXT4_BG_BLOCK_UNINIT
: ~0;
266 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
268 src_group
= group_data
[0].group
;
269 last_group
= src_group
+ flex_gd
->count
- 1;
271 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
272 (last_group
& ~(flexbg_size
- 1))));
274 group
= group_data
[0].group
;
275 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
277 start_blk
= ext4_group_first_block_no(sb
, src_group
);
278 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
280 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
282 start_blk
+= overhead
;
284 /* We collect contiguous blocks as much as possible. */
286 for (; src_group
<= last_group
; src_group
++) {
287 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
289 last_blk
+= group_data
[src_group
- group
].blocks_count
;
294 /* Allocate block bitmaps */
295 for (; bb_index
< flex_gd
->count
; bb_index
++) {
296 if (start_blk
>= last_blk
)
298 group_data
[bb_index
].block_bitmap
= start_blk
++;
299 group
= ext4_get_group_number(sb
, start_blk
- 1);
300 group
-= group_data
[0].group
;
301 group_data
[group
].mdata_blocks
++;
302 flex_gd
->bg_flags
[group
] &= uninit_mask
;
305 /* Allocate inode bitmaps */
306 for (; ib_index
< flex_gd
->count
; ib_index
++) {
307 if (start_blk
>= last_blk
)
309 group_data
[ib_index
].inode_bitmap
= start_blk
++;
310 group
= ext4_get_group_number(sb
, start_blk
- 1);
311 group
-= group_data
[0].group
;
312 group_data
[group
].mdata_blocks
++;
313 flex_gd
->bg_flags
[group
] &= uninit_mask
;
316 /* Allocate inode tables */
317 for (; it_index
< flex_gd
->count
; it_index
++) {
318 unsigned int itb
= EXT4_SB(sb
)->s_itb_per_group
;
319 ext4_fsblk_t next_group_start
;
321 if (start_blk
+ itb
> last_blk
)
323 group_data
[it_index
].inode_table
= start_blk
;
324 group
= ext4_get_group_number(sb
, start_blk
);
325 next_group_start
= ext4_group_first_block_no(sb
, group
+ 1);
326 group
-= group_data
[0].group
;
328 if (start_blk
+ itb
> next_group_start
) {
329 flex_gd
->bg_flags
[group
+ 1] &= uninit_mask
;
330 overhead
= start_blk
+ itb
- next_group_start
;
331 group_data
[group
+ 1].mdata_blocks
+= overhead
;
335 group_data
[group
].mdata_blocks
+= itb
;
336 flex_gd
->bg_flags
[group
] &= uninit_mask
;
337 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
340 /* Update free clusters count to exclude metadata blocks */
341 for (i
= 0; i
< flex_gd
->count
; i
++) {
342 group_data
[i
].free_clusters_count
-=
343 EXT4_NUM_B2C(EXT4_SB(sb
),
344 group_data
[i
].mdata_blocks
);
347 if (test_opt(sb
, DEBUG
)) {
349 group
= group_data
[0].group
;
351 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
352 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
355 for (i
= 0; i
< flex_gd
->count
; i
++) {
357 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
358 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
359 "no-super", group
+ i
,
360 group_data
[i
].blocks_count
,
361 group_data
[i
].free_clusters_count
,
362 group_data
[i
].mdata_blocks
);
368 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
371 struct buffer_head
*bh
;
374 bh
= sb_getblk(sb
, blk
);
376 return ERR_PTR(-ENOMEM
);
377 BUFFER_TRACE(bh
, "get_write_access");
378 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
382 memset(bh
->b_data
, 0, sb
->s_blocksize
);
383 set_buffer_uptodate(bh
);
390 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
391 * If that fails, restart the transaction & regain write access for the
392 * buffer head which is used for block_bitmap modifications.
394 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
398 if (ext4_handle_has_enough_credits(handle
, thresh
))
401 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
405 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
414 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
416 * Helper function for ext4_setup_new_group_blocks() which set .
419 * @handle: journal handle
420 * @flex_gd: flex group data
422 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
423 struct ext4_new_flex_group_data
*flex_gd
,
424 ext4_fsblk_t first_cluster
, ext4_fsblk_t last_cluster
)
426 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
427 ext4_group_t count
= last_cluster
- first_cluster
+ 1;
430 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster
,
432 for (count2
= count
; count
> 0;
433 count
-= count2
, first_cluster
+= count2
) {
435 struct buffer_head
*bh
;
439 group
= ext4_get_group_number(sb
, EXT4_C2B(sbi
, first_cluster
));
440 start
= EXT4_B2C(sbi
, ext4_group_first_block_no(sb
, group
));
441 group
-= flex_gd
->groups
[0].group
;
443 count2
= EXT4_CLUSTERS_PER_GROUP(sb
) - (first_cluster
- start
);
447 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
448 BUG_ON(flex_gd
->count
> 1);
452 err
= extend_or_restart_transaction(handle
, 1);
456 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
460 BUFFER_TRACE(bh
, "get_write_access");
461 err
= ext4_journal_get_write_access(handle
, bh
);
466 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
467 first_cluster
, first_cluster
- start
, count2
);
468 ext4_set_bits(bh
->b_data
, first_cluster
- start
, count2
);
470 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
480 * Set up the block and inode bitmaps, and the inode table for the new groups.
481 * This doesn't need to be part of the main transaction, since we are only
482 * changing blocks outside the actual filesystem. We still do journaling to
483 * ensure the recovery is correct in case of a failure just after resize.
484 * If any part of this fails, we simply abort the resize.
486 * setup_new_flex_group_blocks handles a flex group as follow:
487 * 1. copy super block and GDT, and initialize group tables if necessary.
488 * In this step, we only set bits in blocks bitmaps for blocks taken by
489 * super block and GDT.
490 * 2. allocate group tables in block bitmaps, that is, set bits in block
491 * bitmap for blocks taken by group tables.
493 static int setup_new_flex_group_blocks(struct super_block
*sb
,
494 struct ext4_new_flex_group_data
*flex_gd
)
496 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
499 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
500 struct ext4_super_block
*es
= sbi
->s_es
;
501 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
502 __u16
*bg_flags
= flex_gd
->bg_flags
;
504 ext4_group_t group
, count
;
505 struct buffer_head
*bh
= NULL
;
506 int reserved_gdb
, i
, j
, err
= 0, err2
;
509 BUG_ON(!flex_gd
->count
|| !group_data
||
510 group_data
[0].group
!= sbi
->s_groups_count
);
512 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
513 meta_bg
= ext4_has_feature_meta_bg(sb
);
515 /* This transaction may be extended/restarted along the way */
516 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
518 return PTR_ERR(handle
);
520 group
= group_data
[0].group
;
521 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
522 unsigned long gdblocks
;
523 ext4_grpblk_t overhead
;
525 gdblocks
= ext4_bg_num_gdb(sb
, group
);
526 start
= ext4_group_first_block_no(sb
, group
);
528 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
532 ext4_group_t first_group
;
533 first_group
= ext4_meta_bg_first_group(sb
, group
);
534 if (first_group
!= group
+ 1 &&
535 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
539 block
= start
+ ext4_bg_has_super(sb
, group
);
540 /* Copy all of the GDT blocks into the backup in this group */
541 for (j
= 0; j
< gdblocks
; j
++, block
++) {
542 struct buffer_head
*gdb
;
544 ext4_debug("update backup group %#04llx\n", block
);
545 err
= extend_or_restart_transaction(handle
, 1);
549 gdb
= sb_getblk(sb
, block
);
550 if (unlikely(!gdb
)) {
555 BUFFER_TRACE(gdb
, "get_write_access");
556 err
= ext4_journal_get_write_access(handle
, gdb
);
561 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
563 set_buffer_uptodate(gdb
);
565 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
573 /* Zero out all of the reserved backup group descriptor
576 if (ext4_bg_has_super(sb
, group
)) {
577 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
578 reserved_gdb
, GFP_NOFS
);
584 /* Initialize group tables of the grop @group */
585 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
588 /* Zero out all of the inode table blocks */
589 block
= group_data
[i
].inode_table
;
590 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
591 block
, sbi
->s_itb_per_group
);
592 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
598 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
601 /* Initialize block bitmap of the @group */
602 block
= group_data
[i
].block_bitmap
;
603 err
= extend_or_restart_transaction(handle
, 1);
607 bh
= bclean(handle
, sb
, block
);
612 overhead
= ext4_group_overhead_blocks(sb
, group
);
614 ext4_debug("mark backup superblock %#04llx (+0)\n",
616 ext4_set_bits(bh
->b_data
, 0,
617 EXT4_NUM_B2C(sbi
, overhead
));
619 ext4_mark_bitmap_end(EXT4_B2C(sbi
, group_data
[i
].blocks_count
),
620 sb
->s_blocksize
* 8, bh
->b_data
);
621 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
627 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
630 /* Initialize inode bitmap of the @group */
631 block
= group_data
[i
].inode_bitmap
;
632 err
= extend_or_restart_transaction(handle
, 1);
635 /* Mark unused entries in inode bitmap used */
636 bh
= bclean(handle
, sb
, block
);
642 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
643 sb
->s_blocksize
* 8, bh
->b_data
);
644 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
650 /* Mark group tables in block bitmap */
651 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
652 count
= group_table_count
[j
];
653 start
= (&group_data
[0].block_bitmap
)[j
];
655 for (i
= 1; i
< flex_gd
->count
; i
++) {
656 block
+= group_table_count
[j
];
657 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
658 count
+= group_table_count
[j
];
661 err
= set_flexbg_block_bitmap(sb
, handle
,
663 EXT4_B2C(sbi
, start
),
669 count
= group_table_count
[j
];
670 start
= (&group_data
[i
].block_bitmap
)[j
];
675 err
= set_flexbg_block_bitmap(sb
, handle
,
677 EXT4_B2C(sbi
, start
),
687 err2
= ext4_journal_stop(handle
);
695 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
696 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
697 * calling this for the first time. In a sparse filesystem it will be the
698 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
699 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
701 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
702 unsigned *five
, unsigned *seven
)
704 unsigned *min
= three
;
708 if (!ext4_has_feature_sparse_super(sb
)) {
730 * Check that all of the backup GDT blocks are held in the primary GDT block.
731 * It is assumed that they are stored in group order. Returns the number of
732 * groups in current filesystem that have BACKUPS, or -ve error code.
734 static int verify_reserved_gdb(struct super_block
*sb
,
736 struct buffer_head
*primary
)
738 const ext4_fsblk_t blk
= primary
->b_blocknr
;
743 __le32
*p
= (__le32
*)primary
->b_data
;
746 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
747 if (le32_to_cpu(*p
++) !=
748 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
749 ext4_warning(sb
, "reserved GDT %llu"
750 " missing grp %d (%llu)",
753 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
757 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
765 * Called when we need to bring a reserved group descriptor table block into
766 * use from the resize inode. The primary copy of the new GDT block currently
767 * is an indirect block (under the double indirect block in the resize inode).
768 * The new backup GDT blocks will be stored as leaf blocks in this indirect
769 * block, in group order. Even though we know all the block numbers we need,
770 * we check to ensure that the resize inode has actually reserved these blocks.
772 * Don't need to update the block bitmaps because the blocks are still in use.
774 * We get all of the error cases out of the way, so that we are sure to not
775 * fail once we start modifying the data on disk, because JBD has no rollback.
777 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
780 struct super_block
*sb
= inode
->i_sb
;
781 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
782 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
783 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
784 struct buffer_head
**o_group_desc
, **n_group_desc
;
785 struct buffer_head
*dind
;
786 struct buffer_head
*gdb_bh
;
788 struct ext4_iloc iloc
;
792 if (test_opt(sb
, DEBUG
))
794 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
797 gdb_bh
= sb_bread(sb
, gdblock
);
801 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
807 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
808 dind
= sb_bread(sb
, le32_to_cpu(*data
));
814 data
= (__le32
*)dind
->b_data
;
815 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
816 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
822 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
823 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
827 BUFFER_TRACE(gdb_bh
, "get_write_access");
828 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
832 BUFFER_TRACE(dind
, "get_write_access");
833 err
= ext4_journal_get_write_access(handle
, dind
);
835 ext4_std_error(sb
, err
);
837 /* ext4_reserve_inode_write() gets a reference on the iloc */
838 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
842 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
843 sizeof(struct buffer_head
*),
847 ext4_warning(sb
, "not enough memory for %lu groups",
853 * Finally, we have all of the possible failures behind us...
855 * Remove new GDT block from inode double-indirect block and clear out
856 * the new GDT block for use (which also "frees" the backup GDT blocks
857 * from the reserved inode). We don't need to change the bitmaps for
858 * these blocks, because they are marked as in-use from being in the
859 * reserved inode, and will become GDT blocks (primary and backup).
861 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
862 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
864 ext4_std_error(sb
, err
);
867 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>>
868 (9 - EXT4_SB(sb
)->s_cluster_bits
);
869 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
870 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
871 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
873 ext4_std_error(sb
, err
);
879 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
880 memcpy(n_group_desc
, o_group_desc
,
881 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
882 n_group_desc
[gdb_num
] = gdb_bh
;
883 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
884 EXT4_SB(sb
)->s_gdb_count
++;
885 kvfree(o_group_desc
);
887 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
888 err
= ext4_handle_dirty_super(handle
, sb
);
890 ext4_std_error(sb
, err
);
895 kvfree(n_group_desc
);
902 ext4_debug("leaving with error %d\n", err
);
907 * add_new_gdb_meta_bg is the sister of add_new_gdb.
909 static int add_new_gdb_meta_bg(struct super_block
*sb
,
910 handle_t
*handle
, ext4_group_t group
) {
911 ext4_fsblk_t gdblock
;
912 struct buffer_head
*gdb_bh
;
913 struct buffer_head
**o_group_desc
, **n_group_desc
;
914 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
917 gdblock
= ext4_meta_bg_first_block_no(sb
, group
) +
918 ext4_bg_has_super(sb
, group
);
919 gdb_bh
= sb_bread(sb
, gdblock
);
922 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
923 sizeof(struct buffer_head
*),
928 ext4_warning(sb
, "not enough memory for %lu groups",
933 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
934 memcpy(n_group_desc
, o_group_desc
,
935 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
936 n_group_desc
[gdb_num
] = gdb_bh
;
937 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
938 EXT4_SB(sb
)->s_gdb_count
++;
939 kvfree(o_group_desc
);
940 BUFFER_TRACE(gdb_bh
, "get_write_access");
941 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
946 * Called when we are adding a new group which has a backup copy of each of
947 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
948 * We need to add these reserved backup GDT blocks to the resize inode, so
949 * that they are kept for future resizing and not allocated to files.
951 * Each reserved backup GDT block will go into a different indirect block.
952 * The indirect blocks are actually the primary reserved GDT blocks,
953 * so we know in advance what their block numbers are. We only get the
954 * double-indirect block to verify it is pointing to the primary reserved
955 * GDT blocks so we don't overwrite a data block by accident. The reserved
956 * backup GDT blocks are stored in their reserved primary GDT block.
958 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
961 struct super_block
*sb
= inode
->i_sb
;
962 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
963 int cluster_bits
= EXT4_SB(sb
)->s_cluster_bits
;
964 struct buffer_head
**primary
;
965 struct buffer_head
*dind
;
966 struct ext4_iloc iloc
;
973 primary
= kmalloc_array(reserved_gdb
, sizeof(*primary
), GFP_NOFS
);
977 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
978 dind
= sb_bread(sb
, le32_to_cpu(*data
));
984 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
985 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
986 EXT4_ADDR_PER_BLOCK(sb
));
987 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
989 /* Get each reserved primary GDT block and verify it holds backups */
990 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
991 if (le32_to_cpu(*data
) != blk
) {
992 ext4_warning(sb
, "reserved block %llu"
993 " not at offset %ld",
995 (long)(data
- (__le32
*)dind
->b_data
));
999 primary
[res
] = sb_bread(sb
, blk
);
1000 if (!primary
[res
]) {
1004 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
1005 if (gdbackups
< 0) {
1006 brelse(primary
[res
]);
1011 data
= (__le32
*)dind
->b_data
;
1014 for (i
= 0; i
< reserved_gdb
; i
++) {
1015 BUFFER_TRACE(primary
[i
], "get_write_access");
1016 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
])))
1020 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
1024 * Finally we can add each of the reserved backup GDT blocks from
1025 * the new group to its reserved primary GDT block.
1027 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1028 for (i
= 0; i
< reserved_gdb
; i
++) {
1030 data
= (__le32
*)primary
[i
]->b_data
;
1031 /* printk("reserving backup %lu[%u] = %lu\n",
1032 primary[i]->b_blocknr, gdbackups,
1033 blk + primary[i]->b_blocknr); */
1034 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1035 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1040 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> (9 - cluster_bits
);
1041 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1045 brelse(primary
[res
]);
1055 * Update the backup copies of the ext4 metadata. These don't need to be part
1056 * of the main resize transaction, because e2fsck will re-write them if there
1057 * is a problem (basically only OOM will cause a problem). However, we
1058 * _should_ update the backups if possible, in case the primary gets trashed
1059 * for some reason and we need to run e2fsck from a backup superblock. The
1060 * important part is that the new block and inode counts are in the backup
1061 * superblocks, and the location of the new group metadata in the GDT backups.
1063 * We do not need take the s_resize_lock for this, because these
1064 * blocks are not otherwise touched by the filesystem code when it is
1065 * mounted. We don't need to worry about last changing from
1066 * sbi->s_groups_count, because the worst that can happen is that we
1067 * do not copy the full number of backups at this time. The resize
1068 * which changed s_groups_count will backup again.
1070 static void update_backups(struct super_block
*sb
, sector_t blk_off
, char *data
,
1071 int size
, int meta_bg
)
1073 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1075 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1079 ext4_group_t group
= 0;
1080 int rest
= sb
->s_blocksize
- size
;
1084 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
1085 if (IS_ERR(handle
)) {
1087 err
= PTR_ERR(handle
);
1092 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1093 last
= sbi
->s_groups_count
;
1095 group
= ext4_get_group_number(sb
, blk_off
) + 1;
1096 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1099 while (group
< sbi
->s_groups_count
) {
1100 struct buffer_head
*bh
;
1101 ext4_fsblk_t backup_block
;
1103 /* Out of journal space, and can't get more - abort - so sad */
1104 if (ext4_handle_valid(handle
) &&
1105 handle
->h_buffer_credits
== 0 &&
1106 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1107 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1111 backup_block
= ((ext4_fsblk_t
)group
) * bpg
+ blk_off
;
1113 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1114 ext4_bg_has_super(sb
, group
));
1116 bh
= sb_getblk(sb
, backup_block
);
1117 if (unlikely(!bh
)) {
1121 ext4_debug("update metadata backup %llu(+%llu)\n",
1122 backup_block
, backup_block
-
1123 ext4_group_first_block_no(sb
, group
));
1124 BUFFER_TRACE(bh
, "get_write_access");
1125 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
1130 memcpy(bh
->b_data
, data
, size
);
1132 memset(bh
->b_data
+ size
, 0, rest
);
1133 set_buffer_uptodate(bh
);
1135 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1137 ext4_std_error(sb
, err
);
1141 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1142 else if (group
== last
)
1147 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1151 * Ugh! Need to have e2fsck write the backup copies. It is too
1152 * late to revert the resize, we shouldn't fail just because of
1153 * the backup copies (they are only needed in case of corruption).
1155 * However, if we got here we have a journal problem too, so we
1156 * can't really start a transaction to mark the superblock.
1157 * Chicken out and just set the flag on the hope it will be written
1158 * to disk, and if not - we will simply wait until next fsck.
1162 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1163 "forcing fsck on next reboot", group
, err
);
1164 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1165 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1166 mark_buffer_dirty(sbi
->s_sbh
);
1171 * ext4_add_new_descs() adds @count group descriptor of groups
1172 * starting at @group
1174 * @handle: journal handle
1176 * @group: the group no. of the first group desc to be added
1177 * @resize_inode: the resize inode
1178 * @count: number of group descriptors to be added
1180 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1181 ext4_group_t group
, struct inode
*resize_inode
,
1184 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1185 struct ext4_super_block
*es
= sbi
->s_es
;
1186 struct buffer_head
*gdb_bh
;
1187 int i
, gdb_off
, gdb_num
, err
= 0;
1190 meta_bg
= ext4_has_feature_meta_bg(sb
);
1191 for (i
= 0; i
< count
; i
++, group
++) {
1192 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1193 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1195 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1196 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1199 * We will only either add reserved group blocks to a backup group
1200 * or remove reserved blocks for the first group in a new group block.
1201 * Doing both would be mean more complex code, and sane people don't
1202 * use non-sparse filesystems anymore. This is already checked above.
1205 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1206 BUFFER_TRACE(gdb_bh
, "get_write_access");
1207 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1209 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1210 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1211 } else if (meta_bg
!= 0) {
1212 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1214 err
= add_new_gdb(handle
, resize_inode
, group
);
1222 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1224 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1227 if (!bh_uptodate_or_lock(bh
)) {
1228 if (bh_submit_read(bh
) < 0) {
1237 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1239 struct ext4_group_desc
*gdp
,
1240 struct ext4_new_group_data
*group_data
)
1242 struct buffer_head
*bh
;
1244 if (!ext4_has_metadata_csum(sb
))
1247 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1250 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1251 EXT4_INODES_PER_GROUP(sb
) / 8);
1254 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1257 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1264 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1266 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1267 struct ext4_new_flex_group_data
*flex_gd
)
1269 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1270 struct ext4_group_desc
*gdp
;
1271 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1272 struct buffer_head
*gdb_bh
;
1274 __u16
*bg_flags
= flex_gd
->bg_flags
;
1275 int i
, gdb_off
, gdb_num
, err
= 0;
1278 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1279 group
= group_data
->group
;
1281 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1282 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1285 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1287 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1288 /* Update group descriptor block for new group */
1289 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1290 gdb_off
* EXT4_DESC_SIZE(sb
));
1292 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1293 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1294 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1295 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1297 ext4_std_error(sb
, err
);
1301 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1302 ext4_free_group_clusters_set(sb
, gdp
,
1303 group_data
->free_clusters_count
);
1304 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1305 if (ext4_has_group_desc_csum(sb
))
1306 ext4_itable_unused_set(sb
, gdp
,
1307 EXT4_INODES_PER_GROUP(sb
));
1308 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1309 ext4_group_desc_csum_set(sb
, group
, gdp
);
1311 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1312 if (unlikely(err
)) {
1313 ext4_std_error(sb
, err
);
1318 * We can allocate memory for mb_alloc based on the new group
1321 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1329 * ext4_update_super() updates the super block so that the newly added
1330 * groups can be seen by the filesystem.
1333 * @flex_gd: new added groups
1335 static void ext4_update_super(struct super_block
*sb
,
1336 struct ext4_new_flex_group_data
*flex_gd
)
1338 ext4_fsblk_t blocks_count
= 0;
1339 ext4_fsblk_t free_blocks
= 0;
1340 ext4_fsblk_t reserved_blocks
= 0;
1341 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1342 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1343 struct ext4_super_block
*es
= sbi
->s_es
;
1346 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1348 * Make the new blocks and inodes valid next. We do this before
1349 * increasing the group count so that once the group is enabled,
1350 * all of its blocks and inodes are already valid.
1352 * We always allocate group-by-group, then block-by-block or
1353 * inode-by-inode within a group, so enabling these
1354 * blocks/inodes before the group is live won't actually let us
1355 * allocate the new space yet.
1357 for (i
= 0; i
< flex_gd
->count
; i
++) {
1358 blocks_count
+= group_data
[i
].blocks_count
;
1359 free_blocks
+= EXT4_C2B(sbi
, group_data
[i
].free_clusters_count
);
1362 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1363 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1364 reserved_blocks
*= blocks_count
;
1365 do_div(reserved_blocks
, 100);
1367 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1368 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1369 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1371 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1374 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1376 * We need to protect s_groups_count against other CPUs seeing
1377 * inconsistent state in the superblock.
1379 * The precise rules we use are:
1381 * * Writers must perform a smp_wmb() after updating all
1382 * dependent data and before modifying the groups count
1384 * * Readers must perform an smp_rmb() after reading the groups
1385 * count and before reading any dependent data.
1387 * NB. These rules can be relaxed when checking the group count
1388 * while freeing data, as we can only allocate from a block
1389 * group after serialising against the group count, and we can
1390 * only then free after serialising in turn against that
1395 /* Update the global fs size fields */
1396 sbi
->s_groups_count
+= flex_gd
->count
;
1397 sbi
->s_blockfile_groups
= min_t(ext4_group_t
, sbi
->s_groups_count
,
1398 (EXT4_MAX_BLOCK_FILE_PHYS
/ EXT4_BLOCKS_PER_GROUP(sb
)));
1400 /* Update the reserved block counts only once the new group is
1402 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1405 /* Update the free space counts */
1406 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1407 EXT4_NUM_B2C(sbi
, free_blocks
));
1408 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1409 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1411 ext4_debug("free blocks count %llu",
1412 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1413 if (ext4_has_feature_flex_bg(sb
) && sbi
->s_log_groups_per_flex
) {
1414 ext4_group_t flex_group
;
1415 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1416 atomic64_add(EXT4_NUM_B2C(sbi
, free_blocks
),
1417 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1418 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1419 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1423 * Update the fs overhead information
1425 ext4_calculate_overhead(sb
);
1427 if (test_opt(sb
, DEBUG
))
1428 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1429 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1430 blocks_count
, free_blocks
, reserved_blocks
);
1433 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1434 * _before_ we start modifying the filesystem, because we cannot abort the
1435 * transaction and not have it write the data to disk.
1437 static int ext4_flex_group_add(struct super_block
*sb
,
1438 struct inode
*resize_inode
,
1439 struct ext4_new_flex_group_data
*flex_gd
)
1441 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1442 struct ext4_super_block
*es
= sbi
->s_es
;
1443 ext4_fsblk_t o_blocks_count
;
1447 unsigned reserved_gdb
;
1448 int err
= 0, err2
= 0, credit
;
1450 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1452 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1453 o_blocks_count
= ext4_blocks_count(es
);
1454 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1457 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1461 * We will always be modifying at least the superblock and GDT
1462 * blocks. If we are adding a group past the last current GDT block,
1463 * we will also modify the inode and the dindirect block. If we
1464 * are adding a group with superblock/GDT backups we will also
1465 * modify each of the reserved GDT dindirect blocks.
1467 credit
= 3; /* sb, resize inode, resize inode dindirect */
1469 credit
+= 1 + DIV_ROUND_UP(flex_gd
->count
, EXT4_DESC_PER_BLOCK(sb
));
1470 credit
+= reserved_gdb
; /* Reserved GDT dindirect blocks */
1471 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credit
);
1472 if (IS_ERR(handle
)) {
1473 err
= PTR_ERR(handle
);
1477 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1478 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1482 group
= flex_gd
->groups
[0].group
;
1483 BUG_ON(group
!= sbi
->s_groups_count
);
1484 err
= ext4_add_new_descs(handle
, sb
, group
,
1485 resize_inode
, flex_gd
->count
);
1489 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1493 ext4_update_super(sb
, flex_gd
);
1495 err
= ext4_handle_dirty_super(handle
, sb
);
1498 err2
= ext4_journal_stop(handle
);
1503 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1504 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1505 EXT4_DESC_PER_BLOCK(sb
));
1506 int meta_bg
= ext4_has_feature_meta_bg(sb
);
1507 sector_t old_gdb
= 0;
1509 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1510 sizeof(struct ext4_super_block
), 0);
1511 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1512 struct buffer_head
*gdb_bh
;
1514 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1515 if (old_gdb
== gdb_bh
->b_blocknr
)
1517 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1518 gdb_bh
->b_size
, meta_bg
);
1519 old_gdb
= gdb_bh
->b_blocknr
;
1526 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1527 struct ext4_new_flex_group_data
*flex_gd
,
1528 ext4_fsblk_t n_blocks_count
,
1529 unsigned long flexbg_size
)
1531 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1532 struct ext4_super_block
*es
= sbi
->s_es
;
1533 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1534 ext4_fsblk_t o_blocks_count
;
1535 ext4_group_t n_group
;
1537 ext4_group_t last_group
;
1539 ext4_grpblk_t clusters_per_group
;
1542 clusters_per_group
= EXT4_CLUSTERS_PER_GROUP(sb
);
1544 o_blocks_count
= ext4_blocks_count(es
);
1546 if (o_blocks_count
== n_blocks_count
)
1549 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1551 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1553 last_group
= group
| (flexbg_size
- 1);
1554 if (last_group
> n_group
)
1555 last_group
= n_group
;
1557 flex_gd
->count
= last_group
- group
+ 1;
1559 for (i
= 0; i
< flex_gd
->count
; i
++) {
1562 group_data
[i
].group
= group
+ i
;
1563 group_data
[i
].blocks_count
= EXT4_BLOCKS_PER_GROUP(sb
);
1564 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1565 group_data
[i
].mdata_blocks
= overhead
;
1566 group_data
[i
].free_clusters_count
= EXT4_CLUSTERS_PER_GROUP(sb
);
1567 if (ext4_has_group_desc_csum(sb
)) {
1568 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1569 EXT4_BG_INODE_UNINIT
;
1570 if (!test_opt(sb
, INIT_INODE_TABLE
))
1571 flex_gd
->bg_flags
[i
] |= EXT4_BG_INODE_ZEROED
;
1573 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1576 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1577 /* We need to initialize block bitmap of last group. */
1578 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1580 if ((last_group
== n_group
) && (last
!= clusters_per_group
- 1)) {
1581 group_data
[i
- 1].blocks_count
= EXT4_C2B(sbi
, last
+ 1);
1582 group_data
[i
- 1].free_clusters_count
-= clusters_per_group
-
1589 /* Add group descriptor data to an existing or new group descriptor block.
1590 * Ensure we handle all possible error conditions _before_ we start modifying
1591 * the filesystem, because we cannot abort the transaction and not have it
1592 * write the data to disk.
1594 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1595 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1597 * We only need to hold the superblock lock while we are actually adding
1598 * in the new group's counts to the superblock. Prior to that we have
1599 * not really "added" the group at all. We re-check that we are still
1600 * adding in the last group in case things have changed since verifying.
1602 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1604 struct ext4_new_flex_group_data flex_gd
;
1605 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1606 struct ext4_super_block
*es
= sbi
->s_es
;
1607 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1608 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1609 struct inode
*inode
= NULL
;
1614 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1616 if (gdb_off
== 0 && !ext4_has_feature_sparse_super(sb
)) {
1617 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1621 if (ext4_blocks_count(es
) + input
->blocks_count
<
1622 ext4_blocks_count(es
)) {
1623 ext4_warning(sb
, "blocks_count overflow");
1627 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1628 le32_to_cpu(es
->s_inodes_count
)) {
1629 ext4_warning(sb
, "inodes_count overflow");
1633 if (reserved_gdb
|| gdb_off
== 0) {
1634 if (ext4_has_feature_resize_inode(sb
) ||
1635 !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1637 "No reserved GDT blocks, can't resize");
1640 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1641 if (IS_ERR(inode
)) {
1642 ext4_warning(sb
, "Error opening resize inode");
1643 return PTR_ERR(inode
);
1648 err
= verify_group_input(sb
, input
);
1652 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1656 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1661 flex_gd
.groups
= input
;
1662 flex_gd
.bg_flags
= &bg_flags
;
1663 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1667 } /* ext4_group_add */
1670 * extend a group without checking assuming that checking has been done.
1672 static int ext4_group_extend_no_check(struct super_block
*sb
,
1673 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1675 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1679 /* We will update the superblock, one block bitmap, and
1680 * one group descriptor via ext4_group_add_blocks().
1682 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, 3);
1683 if (IS_ERR(handle
)) {
1684 err
= PTR_ERR(handle
);
1685 ext4_warning(sb
, "error %d on journal start", err
);
1689 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1690 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1692 ext4_warning(sb
, "error %d on journal write access", err
);
1696 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1697 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1698 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1699 o_blocks_count
+ add
);
1700 /* We add the blocks to the bitmap and set the group need init bit */
1701 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1704 ext4_handle_dirty_super(handle
, sb
);
1705 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1706 o_blocks_count
+ add
);
1708 err2
= ext4_journal_stop(handle
);
1713 if (test_opt(sb
, DEBUG
))
1714 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1715 "blocks\n", ext4_blocks_count(es
));
1716 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
,
1717 (char *)es
, sizeof(struct ext4_super_block
), 0);
1723 * Extend the filesystem to the new number of blocks specified. This entry
1724 * point is only used to extend the current filesystem to the end of the last
1725 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1726 * for emergencies (because it has no dependencies on reserved blocks).
1728 * If we _really_ wanted, we could use default values to call ext4_group_add()
1729 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1730 * GDT blocks are reserved to grow to the desired size.
1732 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1733 ext4_fsblk_t n_blocks_count
)
1735 ext4_fsblk_t o_blocks_count
;
1738 struct buffer_head
*bh
;
1742 o_blocks_count
= ext4_blocks_count(es
);
1744 if (test_opt(sb
, DEBUG
))
1745 ext4_msg(sb
, KERN_DEBUG
,
1746 "extending last group from %llu to %llu blocks",
1747 o_blocks_count
, n_blocks_count
);
1749 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1752 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1753 ext4_msg(sb
, KERN_ERR
,
1754 "filesystem too large to resize to %llu blocks safely",
1756 if (sizeof(sector_t
) < 8)
1757 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1761 if (n_blocks_count
< o_blocks_count
) {
1762 ext4_warning(sb
, "can't shrink FS - resize aborted");
1766 /* Handle the remaining blocks in the last group only. */
1767 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1770 ext4_warning(sb
, "need to use ext2online to resize further");
1774 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1776 if (o_blocks_count
+ add
< o_blocks_count
) {
1777 ext4_warning(sb
, "blocks_count overflow");
1781 if (o_blocks_count
+ add
> n_blocks_count
)
1782 add
= n_blocks_count
- o_blocks_count
;
1784 if (o_blocks_count
+ add
< n_blocks_count
)
1785 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1786 o_blocks_count
+ add
, add
);
1788 /* See if the device is actually as big as what was requested */
1789 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1791 ext4_warning(sb
, "can't read last block, resize aborted");
1796 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1798 } /* ext4_group_extend */
1801 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1803 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1807 * Release the resize inode and drop the resize_inode feature if there
1808 * are no more reserved gdt blocks, and then convert the file system
1811 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1814 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1815 struct ext4_super_block
*es
= sbi
->s_es
;
1816 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1818 int i
, ret
, err
= 0;
1821 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1823 if (es
->s_reserved_gdt_blocks
) {
1824 ext4_error(sb
, "Unexpected non-zero "
1825 "s_reserved_gdt_blocks");
1829 /* Do a quick sanity check of the resize inode */
1830 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
-
1831 (9 - sbi
->s_cluster_bits
)))
1832 goto invalid_resize_inode
;
1833 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1834 if (i
== EXT4_DIND_BLOCK
) {
1838 goto invalid_resize_inode
;
1841 goto invalid_resize_inode
;
1843 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1846 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credits
);
1848 return PTR_ERR(handle
);
1850 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1851 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1855 ext4_clear_feature_resize_inode(sb
);
1856 ext4_set_feature_meta_bg(sb
);
1857 sbi
->s_es
->s_first_meta_bg
=
1858 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1860 err
= ext4_handle_dirty_super(handle
, sb
);
1862 ext4_std_error(sb
, err
);
1867 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1868 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1869 EXT4_FREE_BLOCKS_METADATA
|
1870 EXT4_FREE_BLOCKS_FORGET
);
1871 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1872 inode
->i_blocks
= 0;
1874 err
= ext4_mark_inode_dirty(handle
, inode
);
1876 ext4_std_error(sb
, err
);
1880 ret
= ext4_journal_stop(handle
);
1885 invalid_resize_inode
:
1886 ext4_error(sb
, "corrupted/inconsistent resize inode");
1891 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1893 * @sb: super block of the fs to be resized
1894 * @n_blocks_count: the number of blocks resides in the resized fs
1896 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1898 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1899 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1900 struct ext4_super_block
*es
= sbi
->s_es
;
1901 struct buffer_head
*bh
;
1902 struct inode
*resize_inode
= NULL
;
1903 ext4_grpblk_t add
, offset
;
1904 unsigned long n_desc_blocks
;
1905 unsigned long o_desc_blocks
;
1906 ext4_group_t o_group
;
1907 ext4_group_t n_group
;
1908 ext4_fsblk_t o_blocks_count
;
1909 ext4_fsblk_t n_blocks_count_retry
= 0;
1910 unsigned long last_update_time
= 0;
1911 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1914 /* See if the device is actually as big as what was requested */
1915 bh
= sb_bread(sb
, n_blocks_count
- 1);
1917 ext4_warning(sb
, "can't read last block, resize aborted");
1923 o_blocks_count
= ext4_blocks_count(es
);
1925 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1926 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1928 if (n_blocks_count
< o_blocks_count
) {
1929 /* On-line shrinking not supported */
1930 ext4_warning(sb
, "can't shrink FS - resize aborted");
1934 if (n_blocks_count
== o_blocks_count
)
1935 /* Nothing need to do */
1938 n_group
= ext4_get_group_number(sb
, n_blocks_count
- 1);
1939 if (n_group
>= (0xFFFFFFFFUL
/ EXT4_INODES_PER_GROUP(sb
))) {
1940 ext4_warning(sb
, "resize would cause inodes_count overflow");
1943 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1945 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1946 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1948 meta_bg
= ext4_has_feature_meta_bg(sb
);
1950 if (ext4_has_feature_resize_inode(sb
)) {
1952 ext4_error(sb
, "resize_inode and meta_bg enabled "
1956 if (n_desc_blocks
> o_desc_blocks
+
1957 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1958 n_blocks_count_retry
= n_blocks_count
;
1959 n_desc_blocks
= o_desc_blocks
+
1960 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1961 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1962 n_blocks_count
= (ext4_fsblk_t
)n_group
*
1963 EXT4_BLOCKS_PER_GROUP(sb
);
1964 n_group
--; /* set to last group number */
1968 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1969 if (IS_ERR(resize_inode
)) {
1970 ext4_warning(sb
, "Error opening resize inode");
1971 return PTR_ERR(resize_inode
);
1975 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
1976 err
= ext4_convert_meta_bg(sb
, resize_inode
);
1981 resize_inode
= NULL
;
1983 if (n_blocks_count_retry
) {
1984 n_blocks_count
= n_blocks_count_retry
;
1985 n_blocks_count_retry
= 0;
1991 * Make sure the last group has enough space so that it's
1992 * guaranteed to have enough space for all metadata blocks
1993 * that it might need to hold. (We might not need to store
1994 * the inode table blocks in the last block group, but there
1995 * will be cases where this might be needed.)
1997 if ((ext4_group_first_block_no(sb
, n_group
) +
1998 ext4_group_overhead_blocks(sb
, n_group
) + 2 +
1999 sbi
->s_itb_per_group
+ sbi
->s_cluster_ratio
) >= n_blocks_count
) {
2000 n_blocks_count
= ext4_group_first_block_no(sb
, n_group
);
2002 n_blocks_count_retry
= 0;
2005 resize_inode
= NULL
;
2010 /* extend the last group */
2011 if (n_group
== o_group
)
2012 add
= n_blocks_count
- o_blocks_count
;
2014 add
= EXT4_C2B(sbi
, EXT4_CLUSTERS_PER_GROUP(sb
) - (offset
+ 1));
2016 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
2021 if (ext4_blocks_count(es
) == n_blocks_count
)
2024 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
2028 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
2032 flex_gd
= alloc_flex_gd(flexbg_size
);
2033 if (flex_gd
== NULL
) {
2038 /* Add flex groups. Note that a regular group is a
2039 * flex group with 1 group.
2041 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
2043 if (jiffies
- last_update_time
> HZ
* 10) {
2044 if (last_update_time
)
2045 ext4_msg(sb
, KERN_INFO
,
2046 "resized to %llu blocks",
2047 ext4_blocks_count(es
));
2048 last_update_time
= jiffies
;
2050 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
2052 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
2057 if (!err
&& n_blocks_count_retry
) {
2058 n_blocks_count
= n_blocks_count_retry
;
2059 n_blocks_count_retry
= 0;
2060 free_flex_gd(flex_gd
);
2064 resize_inode
= NULL
;
2071 free_flex_gd(flex_gd
);
2072 if (resize_inode
!= NULL
)
2074 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu", n_blocks_count
);