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_blocks_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 (IS_ERR(bh
= ext4_sb_bread(sb
, end
- 1, 0))) {
133 ext4_warning(sb
, "Cannot read last block (%llu)",
135 } else if (outside(input
->block_bitmap
, start
, end
))
136 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
137 (unsigned long long)input
->block_bitmap
);
138 else if (outside(input
->inode_bitmap
, start
, end
))
139 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
140 (unsigned long long)input
->inode_bitmap
);
141 else if (outside(input
->inode_table
, start
, end
) ||
142 outside(itend
- 1, start
, end
))
143 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
144 (unsigned long long)input
->inode_table
, itend
- 1);
145 else if (input
->inode_bitmap
== input
->block_bitmap
)
146 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
147 (unsigned long long)input
->block_bitmap
);
148 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
149 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
151 (unsigned long long)input
->block_bitmap
,
152 (unsigned long long)input
->inode_table
, itend
- 1);
153 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
154 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
156 (unsigned long long)input
->inode_bitmap
,
157 (unsigned long long)input
->inode_table
, itend
- 1);
158 else if (inside(input
->block_bitmap
, start
, metaend
))
159 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
160 (unsigned long long)input
->block_bitmap
,
162 else if (inside(input
->inode_bitmap
, start
, metaend
))
163 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
164 (unsigned long long)input
->inode_bitmap
,
166 else if (inside(input
->inode_table
, start
, metaend
) ||
167 inside(itend
- 1, start
, metaend
))
168 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
170 (unsigned long long)input
->inode_table
,
171 itend
- 1, start
, metaend
- 1);
180 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
183 struct ext4_new_flex_group_data
{
184 struct ext4_new_group_data
*groups
; /* new_group_data for groups
186 __u16
*bg_flags
; /* block group flags of groups
188 ext4_group_t count
; /* number of groups in @groups
193 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
196 * Returns NULL on failure otherwise address of the allocated structure.
198 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
200 struct ext4_new_flex_group_data
*flex_gd
;
202 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
206 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_group_data
))
208 flex_gd
->count
= flexbg_size
;
210 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
211 flexbg_size
, GFP_NOFS
);
212 if (flex_gd
->groups
== NULL
)
215 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
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;
265 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
267 src_group
= group_data
[0].group
;
268 last_group
= src_group
+ flex_gd
->count
- 1;
270 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
271 (last_group
& ~(flexbg_size
- 1))));
273 group
= group_data
[0].group
;
274 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
276 start_blk
= ext4_group_first_block_no(sb
, src_group
);
277 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
279 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
281 start_blk
+= overhead
;
283 /* We collect contiguous blocks as much as possible. */
285 for (; src_group
<= last_group
; src_group
++) {
286 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
288 last_blk
+= group_data
[src_group
- group
].blocks_count
;
293 /* Allocate block bitmaps */
294 for (; bb_index
< flex_gd
->count
; bb_index
++) {
295 if (start_blk
>= last_blk
)
297 group_data
[bb_index
].block_bitmap
= start_blk
++;
298 group
= ext4_get_group_number(sb
, start_blk
- 1);
299 group
-= group_data
[0].group
;
300 group_data
[group
].free_blocks_count
--;
301 flex_gd
->bg_flags
[group
] &= uninit_mask
;
304 /* Allocate inode bitmaps */
305 for (; ib_index
< flex_gd
->count
; ib_index
++) {
306 if (start_blk
>= last_blk
)
308 group_data
[ib_index
].inode_bitmap
= start_blk
++;
309 group
= ext4_get_group_number(sb
, start_blk
- 1);
310 group
-= group_data
[0].group
;
311 group_data
[group
].free_blocks_count
--;
312 flex_gd
->bg_flags
[group
] &= uninit_mask
;
315 /* Allocate inode tables */
316 for (; it_index
< flex_gd
->count
; it_index
++) {
317 unsigned int itb
= EXT4_SB(sb
)->s_itb_per_group
;
318 ext4_fsblk_t next_group_start
;
320 if (start_blk
+ itb
> last_blk
)
322 group_data
[it_index
].inode_table
= start_blk
;
323 group
= ext4_get_group_number(sb
, start_blk
);
324 next_group_start
= ext4_group_first_block_no(sb
, group
+ 1);
325 group
-= group_data
[0].group
;
327 if (start_blk
+ itb
> next_group_start
) {
328 flex_gd
->bg_flags
[group
+ 1] &= uninit_mask
;
329 overhead
= start_blk
+ itb
- next_group_start
;
330 group_data
[group
+ 1].free_blocks_count
-= overhead
;
334 group_data
[group
].free_blocks_count
-= itb
;
335 flex_gd
->bg_flags
[group
] &= uninit_mask
;
336 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
339 if (test_opt(sb
, DEBUG
)) {
341 group
= group_data
[0].group
;
343 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
344 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
347 for (i
= 0; i
< flex_gd
->count
; i
++) {
348 printk(KERN_DEBUG
"adding %s group %u: %u "
349 "blocks (%d free)\n",
350 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
351 "no-super", group
+ i
,
352 group_data
[i
].blocks_count
,
353 group_data
[i
].free_blocks_count
);
359 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
362 struct buffer_head
*bh
;
365 bh
= sb_getblk(sb
, blk
);
367 return ERR_PTR(-ENOMEM
);
368 BUFFER_TRACE(bh
, "get_write_access");
369 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
373 memset(bh
->b_data
, 0, sb
->s_blocksize
);
374 set_buffer_uptodate(bh
);
381 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
382 * If that fails, restart the transaction & regain write access for the
383 * buffer head which is used for block_bitmap modifications.
385 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
389 if (ext4_handle_has_enough_credits(handle
, thresh
))
392 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
396 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
405 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
407 * Helper function for ext4_setup_new_group_blocks() which set .
410 * @handle: journal handle
411 * @flex_gd: flex group data
413 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
414 struct ext4_new_flex_group_data
*flex_gd
,
415 ext4_fsblk_t block
, ext4_group_t count
)
419 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
420 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
422 struct buffer_head
*bh
;
426 group
= ext4_get_group_number(sb
, block
);
427 start
= ext4_group_first_block_no(sb
, group
);
428 group
-= flex_gd
->groups
[0].group
;
430 count2
= EXT4_BLOCKS_PER_GROUP(sb
) - (block
- start
);
434 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
435 BUG_ON(flex_gd
->count
> 1);
439 err
= extend_or_restart_transaction(handle
, 1);
443 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
447 BUFFER_TRACE(bh
, "get_write_access");
448 err
= ext4_journal_get_write_access(handle
, bh
);
453 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
454 block
- start
, count2
);
455 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
457 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
467 * Set up the block and inode bitmaps, and the inode table for the new groups.
468 * This doesn't need to be part of the main transaction, since we are only
469 * changing blocks outside the actual filesystem. We still do journaling to
470 * ensure the recovery is correct in case of a failure just after resize.
471 * If any part of this fails, we simply abort the resize.
473 * setup_new_flex_group_blocks handles a flex group as follow:
474 * 1. copy super block and GDT, and initialize group tables if necessary.
475 * In this step, we only set bits in blocks bitmaps for blocks taken by
476 * super block and GDT.
477 * 2. allocate group tables in block bitmaps, that is, set bits in block
478 * bitmap for blocks taken by group tables.
480 static int setup_new_flex_group_blocks(struct super_block
*sb
,
481 struct ext4_new_flex_group_data
*flex_gd
)
483 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
486 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
487 struct ext4_super_block
*es
= sbi
->s_es
;
488 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
489 __u16
*bg_flags
= flex_gd
->bg_flags
;
491 ext4_group_t group
, count
;
492 struct buffer_head
*bh
= NULL
;
493 int reserved_gdb
, i
, j
, err
= 0, err2
;
496 BUG_ON(!flex_gd
->count
|| !group_data
||
497 group_data
[0].group
!= sbi
->s_groups_count
);
499 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
500 meta_bg
= ext4_has_feature_meta_bg(sb
);
502 /* This transaction may be extended/restarted along the way */
503 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
505 return PTR_ERR(handle
);
507 group
= group_data
[0].group
;
508 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
509 unsigned long gdblocks
;
510 ext4_grpblk_t overhead
;
512 gdblocks
= ext4_bg_num_gdb(sb
, group
);
513 start
= ext4_group_first_block_no(sb
, group
);
515 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
519 ext4_group_t first_group
;
520 first_group
= ext4_meta_bg_first_group(sb
, group
);
521 if (first_group
!= group
+ 1 &&
522 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
526 block
= start
+ ext4_bg_has_super(sb
, group
);
527 /* Copy all of the GDT blocks into the backup in this group */
528 for (j
= 0; j
< gdblocks
; j
++, block
++) {
529 struct buffer_head
*gdb
;
531 ext4_debug("update backup group %#04llx\n", block
);
532 err
= extend_or_restart_transaction(handle
, 1);
536 gdb
= sb_getblk(sb
, block
);
537 if (unlikely(!gdb
)) {
542 BUFFER_TRACE(gdb
, "get_write_access");
543 err
= ext4_journal_get_write_access(handle
, gdb
);
548 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
550 set_buffer_uptodate(gdb
);
552 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
560 /* Zero out all of the reserved backup group descriptor
563 if (ext4_bg_has_super(sb
, group
)) {
564 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
565 reserved_gdb
, GFP_NOFS
);
571 /* Initialize group tables of the grop @group */
572 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
575 /* Zero out all of the inode table blocks */
576 block
= group_data
[i
].inode_table
;
577 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
578 block
, sbi
->s_itb_per_group
);
579 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
585 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
588 /* Initialize block bitmap of the @group */
589 block
= group_data
[i
].block_bitmap
;
590 err
= extend_or_restart_transaction(handle
, 1);
594 bh
= bclean(handle
, sb
, block
);
599 overhead
= ext4_group_overhead_blocks(sb
, group
);
601 ext4_debug("mark backup superblock %#04llx (+0)\n",
603 ext4_set_bits(bh
->b_data
, 0, overhead
);
605 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
606 sb
->s_blocksize
* 8, bh
->b_data
);
607 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
613 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
616 /* Initialize inode bitmap of the @group */
617 block
= group_data
[i
].inode_bitmap
;
618 err
= extend_or_restart_transaction(handle
, 1);
621 /* Mark unused entries in inode bitmap used */
622 bh
= bclean(handle
, sb
, block
);
628 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
629 sb
->s_blocksize
* 8, bh
->b_data
);
630 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
636 /* Mark group tables in block bitmap */
637 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
638 count
= group_table_count
[j
];
639 start
= (&group_data
[0].block_bitmap
)[j
];
641 for (i
= 1; i
< flex_gd
->count
; i
++) {
642 block
+= group_table_count
[j
];
643 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
644 count
+= group_table_count
[j
];
647 err
= set_flexbg_block_bitmap(sb
, handle
,
648 flex_gd
, start
, count
);
651 count
= group_table_count
[j
];
652 start
= (&group_data
[i
].block_bitmap
)[j
];
657 err
= set_flexbg_block_bitmap(sb
, handle
,
658 flex_gd
, start
, count
);
665 err2
= ext4_journal_stop(handle
);
673 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
674 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
675 * calling this for the first time. In a sparse filesystem it will be the
676 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
677 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
679 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
680 unsigned *five
, unsigned *seven
)
682 unsigned *min
= three
;
686 if (!ext4_has_feature_sparse_super(sb
)) {
708 * Check that all of the backup GDT blocks are held in the primary GDT block.
709 * It is assumed that they are stored in group order. Returns the number of
710 * groups in current filesystem that have BACKUPS, or -ve error code.
712 static int verify_reserved_gdb(struct super_block
*sb
,
714 struct buffer_head
*primary
)
716 const ext4_fsblk_t blk
= primary
->b_blocknr
;
721 __le32
*p
= (__le32
*)primary
->b_data
;
724 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
725 if (le32_to_cpu(*p
++) !=
726 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
727 ext4_warning(sb
, "reserved GDT %llu"
728 " missing grp %d (%llu)",
731 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
735 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
743 * Called when we need to bring a reserved group descriptor table block into
744 * use from the resize inode. The primary copy of the new GDT block currently
745 * is an indirect block (under the double indirect block in the resize inode).
746 * The new backup GDT blocks will be stored as leaf blocks in this indirect
747 * block, in group order. Even though we know all the block numbers we need,
748 * we check to ensure that the resize inode has actually reserved these blocks.
750 * Don't need to update the block bitmaps because the blocks are still in use.
752 * We get all of the error cases out of the way, so that we are sure to not
753 * fail once we start modifying the data on disk, because JBD has no rollback.
755 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
758 struct super_block
*sb
= inode
->i_sb
;
759 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
760 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
761 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
762 struct buffer_head
**o_group_desc
, **n_group_desc
= NULL
;
763 struct buffer_head
*dind
= NULL
;
764 struct buffer_head
*gdb_bh
= NULL
;
766 struct ext4_iloc iloc
= { .bh
= NULL
};
770 if (test_opt(sb
, DEBUG
))
772 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
775 gdb_bh
= ext4_sb_bread(sb
, gdblock
, 0);
777 return PTR_ERR(gdb_bh
);
779 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
785 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
786 dind
= ext4_sb_bread(sb
, le32_to_cpu(*data
), 0);
793 data
= (__le32
*)dind
->b_data
;
794 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
795 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
801 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
802 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
806 BUFFER_TRACE(gdb_bh
, "get_write_access");
807 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
811 BUFFER_TRACE(dind
, "get_write_access");
812 err
= ext4_journal_get_write_access(handle
, dind
);
814 ext4_std_error(sb
, err
);
816 /* ext4_reserve_inode_write() gets a reference on the iloc */
817 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
821 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
822 sizeof(struct buffer_head
*),
826 ext4_warning(sb
, "not enough memory for %lu groups",
832 * Finally, we have all of the possible failures behind us...
834 * Remove new GDT block from inode double-indirect block and clear out
835 * the new GDT block for use (which also "frees" the backup GDT blocks
836 * from the reserved inode). We don't need to change the bitmaps for
837 * these blocks, because they are marked as in-use from being in the
838 * reserved inode, and will become GDT blocks (primary and backup).
840 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
841 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
843 ext4_std_error(sb
, err
);
846 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
847 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
848 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
849 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
851 ext4_std_error(sb
, err
);
856 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
857 memcpy(n_group_desc
, o_group_desc
,
858 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
859 n_group_desc
[gdb_num
] = gdb_bh
;
860 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
861 EXT4_SB(sb
)->s_gdb_count
++;
862 kvfree(o_group_desc
);
864 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
865 err
= ext4_handle_dirty_super(handle
, sb
);
867 ext4_std_error(sb
, err
);
870 kvfree(n_group_desc
);
875 ext4_debug("leaving with error %d\n", err
);
880 * add_new_gdb_meta_bg is the sister of add_new_gdb.
882 static int add_new_gdb_meta_bg(struct super_block
*sb
,
883 handle_t
*handle
, ext4_group_t group
) {
884 ext4_fsblk_t gdblock
;
885 struct buffer_head
*gdb_bh
;
886 struct buffer_head
**o_group_desc
, **n_group_desc
;
887 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
890 gdblock
= ext4_meta_bg_first_block_no(sb
, group
) +
891 ext4_bg_has_super(sb
, group
);
892 gdb_bh
= ext4_sb_bread(sb
, gdblock
, 0);
894 return PTR_ERR(gdb_bh
);
895 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
896 sizeof(struct buffer_head
*),
901 ext4_warning(sb
, "not enough memory for %lu groups",
906 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
907 memcpy(n_group_desc
, o_group_desc
,
908 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
909 n_group_desc
[gdb_num
] = gdb_bh
;
911 BUFFER_TRACE(gdb_bh
, "get_write_access");
912 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
914 kvfree(n_group_desc
);
919 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
920 EXT4_SB(sb
)->s_gdb_count
++;
921 kvfree(o_group_desc
);
926 * Called when we are adding a new group which has a backup copy of each of
927 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
928 * We need to add these reserved backup GDT blocks to the resize inode, so
929 * that they are kept for future resizing and not allocated to files.
931 * Each reserved backup GDT block will go into a different indirect block.
932 * The indirect blocks are actually the primary reserved GDT blocks,
933 * so we know in advance what their block numbers are. We only get the
934 * double-indirect block to verify it is pointing to the primary reserved
935 * GDT blocks so we don't overwrite a data block by accident. The reserved
936 * backup GDT blocks are stored in their reserved primary GDT block.
938 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
941 struct super_block
*sb
= inode
->i_sb
;
942 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
943 struct buffer_head
**primary
;
944 struct buffer_head
*dind
;
945 struct ext4_iloc iloc
;
952 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
956 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
957 dind
= ext4_sb_bread(sb
, le32_to_cpu(*data
), 0);
964 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
965 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
966 EXT4_ADDR_PER_BLOCK(sb
));
967 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
969 /* Get each reserved primary GDT block and verify it holds backups */
970 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
971 if (le32_to_cpu(*data
) != blk
) {
972 ext4_warning(sb
, "reserved block %llu"
973 " not at offset %ld",
975 (long)(data
- (__le32
*)dind
->b_data
));
979 primary
[res
] = ext4_sb_bread(sb
, blk
, 0);
980 if (IS_ERR(primary
[res
])) {
981 err
= PTR_ERR(primary
[res
]);
985 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
987 brelse(primary
[res
]);
992 data
= (__le32
*)dind
->b_data
;
995 for (i
= 0; i
< reserved_gdb
; i
++) {
996 BUFFER_TRACE(primary
[i
], "get_write_access");
997 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
])))
1001 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
1005 * Finally we can add each of the reserved backup GDT blocks from
1006 * the new group to its reserved primary GDT block.
1008 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1009 for (i
= 0; i
< reserved_gdb
; i
++) {
1011 data
= (__le32
*)primary
[i
]->b_data
;
1012 /* printk("reserving backup %lu[%u] = %lu\n",
1013 primary[i]->b_blocknr, gdbackups,
1014 blk + primary[i]->b_blocknr); */
1015 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1016 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1020 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
1021 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1025 brelse(primary
[res
]);
1035 * Update the backup copies of the ext4 metadata. These don't need to be part
1036 * of the main resize transaction, because e2fsck will re-write them if there
1037 * is a problem (basically only OOM will cause a problem). However, we
1038 * _should_ update the backups if possible, in case the primary gets trashed
1039 * for some reason and we need to run e2fsck from a backup superblock. The
1040 * important part is that the new block and inode counts are in the backup
1041 * superblocks, and the location of the new group metadata in the GDT backups.
1043 * We do not need take the s_resize_lock for this, because these
1044 * blocks are not otherwise touched by the filesystem code when it is
1045 * mounted. We don't need to worry about last changing from
1046 * sbi->s_groups_count, because the worst that can happen is that we
1047 * do not copy the full number of backups at this time. The resize
1048 * which changed s_groups_count will backup again.
1050 static void update_backups(struct super_block
*sb
, sector_t blk_off
, char *data
,
1051 int size
, int meta_bg
)
1053 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1055 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1059 ext4_group_t group
= 0;
1060 int rest
= sb
->s_blocksize
- size
;
1064 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
1065 if (IS_ERR(handle
)) {
1067 err
= PTR_ERR(handle
);
1072 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1073 last
= sbi
->s_groups_count
;
1075 group
= ext4_get_group_number(sb
, blk_off
) + 1;
1076 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1079 while (group
< sbi
->s_groups_count
) {
1080 struct buffer_head
*bh
;
1081 ext4_fsblk_t backup_block
;
1083 /* Out of journal space, and can't get more - abort - so sad */
1084 if (ext4_handle_valid(handle
) &&
1085 handle
->h_buffer_credits
== 0 &&
1086 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1087 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1091 backup_block
= ((ext4_fsblk_t
)group
) * bpg
+ blk_off
;
1093 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1094 ext4_bg_has_super(sb
, group
));
1096 bh
= sb_getblk(sb
, backup_block
);
1097 if (unlikely(!bh
)) {
1101 ext4_debug("update metadata backup %llu(+%llu)\n",
1102 backup_block
, backup_block
-
1103 ext4_group_first_block_no(sb
, group
));
1104 BUFFER_TRACE(bh
, "get_write_access");
1105 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
1110 memcpy(bh
->b_data
, data
, size
);
1112 memset(bh
->b_data
+ size
, 0, rest
);
1113 set_buffer_uptodate(bh
);
1115 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1117 ext4_std_error(sb
, err
);
1121 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1122 else if (group
== last
)
1127 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1131 * Ugh! Need to have e2fsck write the backup copies. It is too
1132 * late to revert the resize, we shouldn't fail just because of
1133 * the backup copies (they are only needed in case of corruption).
1135 * However, if we got here we have a journal problem too, so we
1136 * can't really start a transaction to mark the superblock.
1137 * Chicken out and just set the flag on the hope it will be written
1138 * to disk, and if not - we will simply wait until next fsck.
1142 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1143 "forcing fsck on next reboot", group
, err
);
1144 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1145 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1146 mark_buffer_dirty(sbi
->s_sbh
);
1151 * ext4_add_new_descs() adds @count group descriptor of groups
1152 * starting at @group
1154 * @handle: journal handle
1156 * @group: the group no. of the first group desc to be added
1157 * @resize_inode: the resize inode
1158 * @count: number of group descriptors to be added
1160 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1161 ext4_group_t group
, struct inode
*resize_inode
,
1164 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1165 struct ext4_super_block
*es
= sbi
->s_es
;
1166 struct buffer_head
*gdb_bh
;
1167 int i
, gdb_off
, gdb_num
, err
= 0;
1170 meta_bg
= ext4_has_feature_meta_bg(sb
);
1171 for (i
= 0; i
< count
; i
++, group
++) {
1172 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1173 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1175 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1176 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1179 * We will only either add reserved group blocks to a backup group
1180 * or remove reserved blocks for the first group in a new group block.
1181 * Doing both would be mean more complex code, and sane people don't
1182 * use non-sparse filesystems anymore. This is already checked above.
1185 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1186 BUFFER_TRACE(gdb_bh
, "get_write_access");
1187 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1189 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1190 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1191 } else if (meta_bg
!= 0) {
1192 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1194 err
= add_new_gdb(handle
, resize_inode
, group
);
1202 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1204 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1207 if (!bh_uptodate_or_lock(bh
)) {
1208 if (bh_submit_read(bh
) < 0) {
1217 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1219 struct ext4_group_desc
*gdp
,
1220 struct ext4_new_group_data
*group_data
)
1222 struct buffer_head
*bh
;
1224 if (!ext4_has_metadata_csum(sb
))
1227 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1230 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1231 EXT4_INODES_PER_GROUP(sb
) / 8);
1234 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1237 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1244 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1246 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1247 struct ext4_new_flex_group_data
*flex_gd
)
1249 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1250 struct ext4_group_desc
*gdp
;
1251 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1252 struct buffer_head
*gdb_bh
;
1254 __u16
*bg_flags
= flex_gd
->bg_flags
;
1255 int i
, gdb_off
, gdb_num
, err
= 0;
1258 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1259 group
= group_data
->group
;
1261 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1262 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1265 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1267 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1268 /* Update group descriptor block for new group */
1269 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1270 gdb_off
* EXT4_DESC_SIZE(sb
));
1272 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1273 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1274 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1275 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1277 ext4_std_error(sb
, err
);
1281 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1282 ext4_free_group_clusters_set(sb
, gdp
,
1283 EXT4_NUM_B2C(sbi
, group_data
->free_blocks_count
));
1284 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1285 if (ext4_has_group_desc_csum(sb
))
1286 ext4_itable_unused_set(sb
, gdp
,
1287 EXT4_INODES_PER_GROUP(sb
));
1288 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1289 ext4_group_desc_csum_set(sb
, group
, gdp
);
1291 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1292 if (unlikely(err
)) {
1293 ext4_std_error(sb
, err
);
1298 * We can allocate memory for mb_alloc based on the new group
1301 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1309 * ext4_update_super() updates the super block so that the newly added
1310 * groups can be seen by the filesystem.
1313 * @flex_gd: new added groups
1315 static void ext4_update_super(struct super_block
*sb
,
1316 struct ext4_new_flex_group_data
*flex_gd
)
1318 ext4_fsblk_t blocks_count
= 0;
1319 ext4_fsblk_t free_blocks
= 0;
1320 ext4_fsblk_t reserved_blocks
= 0;
1321 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1322 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1323 struct ext4_super_block
*es
= sbi
->s_es
;
1326 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1328 * Make the new blocks and inodes valid next. We do this before
1329 * increasing the group count so that once the group is enabled,
1330 * all of its blocks and inodes are already valid.
1332 * We always allocate group-by-group, then block-by-block or
1333 * inode-by-inode within a group, so enabling these
1334 * blocks/inodes before the group is live won't actually let us
1335 * allocate the new space yet.
1337 for (i
= 0; i
< flex_gd
->count
; i
++) {
1338 blocks_count
+= group_data
[i
].blocks_count
;
1339 free_blocks
+= group_data
[i
].free_blocks_count
;
1342 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1343 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1344 reserved_blocks
*= blocks_count
;
1345 do_div(reserved_blocks
, 100);
1347 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1348 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1349 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1351 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1354 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1356 * We need to protect s_groups_count against other CPUs seeing
1357 * inconsistent state in the superblock.
1359 * The precise rules we use are:
1361 * * Writers must perform a smp_wmb() after updating all
1362 * dependent data and before modifying the groups count
1364 * * Readers must perform an smp_rmb() after reading the groups
1365 * count and before reading any dependent data.
1367 * NB. These rules can be relaxed when checking the group count
1368 * while freeing data, as we can only allocate from a block
1369 * group after serialising against the group count, and we can
1370 * only then free after serialising in turn against that
1375 /* Update the global fs size fields */
1376 sbi
->s_groups_count
+= flex_gd
->count
;
1377 sbi
->s_blockfile_groups
= min_t(ext4_group_t
, sbi
->s_groups_count
,
1378 (EXT4_MAX_BLOCK_FILE_PHYS
/ EXT4_BLOCKS_PER_GROUP(sb
)));
1380 /* Update the reserved block counts only once the new group is
1382 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1385 /* Update the free space counts */
1386 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1387 EXT4_NUM_B2C(sbi
, free_blocks
));
1388 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1389 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1391 ext4_debug("free blocks count %llu",
1392 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1393 if (ext4_has_feature_flex_bg(sb
) && sbi
->s_log_groups_per_flex
) {
1394 ext4_group_t flex_group
;
1395 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1396 atomic64_add(EXT4_NUM_B2C(sbi
, free_blocks
),
1397 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1398 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1399 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1403 * Update the fs overhead information
1405 ext4_calculate_overhead(sb
);
1407 if (test_opt(sb
, DEBUG
))
1408 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1409 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1410 blocks_count
, free_blocks
, reserved_blocks
);
1413 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1414 * _before_ we start modifying the filesystem, because we cannot abort the
1415 * transaction and not have it write the data to disk.
1417 static int ext4_flex_group_add(struct super_block
*sb
,
1418 struct inode
*resize_inode
,
1419 struct ext4_new_flex_group_data
*flex_gd
)
1421 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1422 struct ext4_super_block
*es
= sbi
->s_es
;
1423 ext4_fsblk_t o_blocks_count
;
1427 unsigned reserved_gdb
;
1428 int err
= 0, err2
= 0, credit
;
1430 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1432 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1433 o_blocks_count
= ext4_blocks_count(es
);
1434 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1437 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1441 * We will always be modifying at least the superblock and GDT
1442 * blocks. If we are adding a group past the last current GDT block,
1443 * we will also modify the inode and the dindirect block. If we
1444 * are adding a group with superblock/GDT backups we will also
1445 * modify each of the reserved GDT dindirect blocks.
1447 credit
= 3; /* sb, resize inode, resize inode dindirect */
1449 credit
+= 1 + DIV_ROUND_UP(flex_gd
->count
, EXT4_DESC_PER_BLOCK(sb
));
1450 credit
+= reserved_gdb
; /* Reserved GDT dindirect blocks */
1451 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credit
);
1452 if (IS_ERR(handle
)) {
1453 err
= PTR_ERR(handle
);
1457 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1458 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1462 group
= flex_gd
->groups
[0].group
;
1463 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1464 err
= ext4_add_new_descs(handle
, sb
, group
,
1465 resize_inode
, flex_gd
->count
);
1469 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1473 ext4_update_super(sb
, flex_gd
);
1475 err
= ext4_handle_dirty_super(handle
, sb
);
1478 err2
= ext4_journal_stop(handle
);
1483 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1484 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1485 EXT4_DESC_PER_BLOCK(sb
));
1486 int meta_bg
= ext4_has_feature_meta_bg(sb
);
1487 sector_t old_gdb
= 0;
1489 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1490 sizeof(struct ext4_super_block
), 0);
1491 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1492 struct buffer_head
*gdb_bh
;
1494 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1495 if (old_gdb
== gdb_bh
->b_blocknr
)
1497 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1498 gdb_bh
->b_size
, meta_bg
);
1499 old_gdb
= gdb_bh
->b_blocknr
;
1506 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1507 struct ext4_new_flex_group_data
*flex_gd
,
1508 ext4_fsblk_t n_blocks_count
,
1509 unsigned long flexbg_size
)
1511 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1512 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1513 ext4_fsblk_t o_blocks_count
;
1514 ext4_group_t n_group
;
1516 ext4_group_t last_group
;
1518 ext4_grpblk_t blocks_per_group
;
1521 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1523 o_blocks_count
= ext4_blocks_count(es
);
1525 if (o_blocks_count
== n_blocks_count
)
1528 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1530 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1532 last_group
= group
| (flexbg_size
- 1);
1533 if (last_group
> n_group
)
1534 last_group
= n_group
;
1536 flex_gd
->count
= last_group
- group
+ 1;
1538 for (i
= 0; i
< flex_gd
->count
; i
++) {
1541 group_data
[i
].group
= group
+ i
;
1542 group_data
[i
].blocks_count
= blocks_per_group
;
1543 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1544 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1545 if (ext4_has_group_desc_csum(sb
)) {
1546 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1547 EXT4_BG_INODE_UNINIT
;
1548 if (!test_opt(sb
, INIT_INODE_TABLE
))
1549 flex_gd
->bg_flags
[i
] |= EXT4_BG_INODE_ZEROED
;
1551 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1554 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1555 /* We need to initialize block bitmap of last group. */
1556 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1558 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1559 group_data
[i
- 1].blocks_count
= last
+ 1;
1560 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1567 /* Add group descriptor data to an existing or new group descriptor block.
1568 * Ensure we handle all possible error conditions _before_ we start modifying
1569 * the filesystem, because we cannot abort the transaction and not have it
1570 * write the data to disk.
1572 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1573 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1575 * We only need to hold the superblock lock while we are actually adding
1576 * in the new group's counts to the superblock. Prior to that we have
1577 * not really "added" the group at all. We re-check that we are still
1578 * adding in the last group in case things have changed since verifying.
1580 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1582 struct ext4_new_flex_group_data flex_gd
;
1583 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1584 struct ext4_super_block
*es
= sbi
->s_es
;
1585 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1586 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1587 struct inode
*inode
= NULL
;
1592 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1594 if (gdb_off
== 0 && !ext4_has_feature_sparse_super(sb
)) {
1595 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1599 if (ext4_blocks_count(es
) + input
->blocks_count
<
1600 ext4_blocks_count(es
)) {
1601 ext4_warning(sb
, "blocks_count overflow");
1605 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1606 le32_to_cpu(es
->s_inodes_count
)) {
1607 ext4_warning(sb
, "inodes_count overflow");
1611 if (reserved_gdb
|| gdb_off
== 0) {
1612 if (!ext4_has_feature_resize_inode(sb
) ||
1613 !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1615 "No reserved GDT blocks, can't resize");
1618 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1619 if (IS_ERR(inode
)) {
1620 ext4_warning(sb
, "Error opening resize inode");
1621 return PTR_ERR(inode
);
1626 err
= verify_group_input(sb
, input
);
1630 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1634 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1639 flex_gd
.groups
= input
;
1640 flex_gd
.bg_flags
= &bg_flags
;
1641 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1645 } /* ext4_group_add */
1648 * extend a group without checking assuming that checking has been done.
1650 static int ext4_group_extend_no_check(struct super_block
*sb
,
1651 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1653 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1657 /* We will update the superblock, one block bitmap, and
1658 * one group descriptor via ext4_group_add_blocks().
1660 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, 3);
1661 if (IS_ERR(handle
)) {
1662 err
= PTR_ERR(handle
);
1663 ext4_warning(sb
, "error %d on journal start", err
);
1667 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1668 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1670 ext4_warning(sb
, "error %d on journal write access", err
);
1674 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1675 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1676 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1677 o_blocks_count
+ add
);
1678 /* We add the blocks to the bitmap and set the group need init bit */
1679 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1682 ext4_handle_dirty_super(handle
, sb
);
1683 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1684 o_blocks_count
+ add
);
1686 err2
= ext4_journal_stop(handle
);
1691 if (test_opt(sb
, DEBUG
))
1692 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1693 "blocks\n", ext4_blocks_count(es
));
1694 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
,
1695 (char *)es
, sizeof(struct ext4_super_block
), 0);
1701 * Extend the filesystem to the new number of blocks specified. This entry
1702 * point is only used to extend the current filesystem to the end of the last
1703 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1704 * for emergencies (because it has no dependencies on reserved blocks).
1706 * If we _really_ wanted, we could use default values to call ext4_group_add()
1707 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1708 * GDT blocks are reserved to grow to the desired size.
1710 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1711 ext4_fsblk_t n_blocks_count
)
1713 ext4_fsblk_t o_blocks_count
;
1716 struct buffer_head
*bh
;
1720 o_blocks_count
= ext4_blocks_count(es
);
1722 if (test_opt(sb
, DEBUG
))
1723 ext4_msg(sb
, KERN_DEBUG
,
1724 "extending last group from %llu to %llu blocks",
1725 o_blocks_count
, n_blocks_count
);
1727 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1730 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1731 ext4_msg(sb
, KERN_ERR
,
1732 "filesystem too large to resize to %llu blocks safely",
1734 if (sizeof(sector_t
) < 8)
1735 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1739 if (n_blocks_count
< o_blocks_count
) {
1740 ext4_warning(sb
, "can't shrink FS - resize aborted");
1744 /* Handle the remaining blocks in the last group only. */
1745 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1748 ext4_warning(sb
, "need to use ext2online to resize further");
1752 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1754 if (o_blocks_count
+ add
< o_blocks_count
) {
1755 ext4_warning(sb
, "blocks_count overflow");
1759 if (o_blocks_count
+ add
> n_blocks_count
)
1760 add
= n_blocks_count
- o_blocks_count
;
1762 if (o_blocks_count
+ add
< n_blocks_count
)
1763 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1764 o_blocks_count
+ add
, add
);
1766 /* See if the device is actually as big as what was requested */
1767 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1769 ext4_warning(sb
, "can't read last block, resize aborted");
1774 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1776 } /* ext4_group_extend */
1779 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1781 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1785 * Release the resize inode and drop the resize_inode feature if there
1786 * are no more reserved gdt blocks, and then convert the file system
1789 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1792 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1793 struct ext4_super_block
*es
= sbi
->s_es
;
1794 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1796 int i
, ret
, err
= 0;
1799 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1801 if (es
->s_reserved_gdt_blocks
) {
1802 ext4_error(sb
, "Unexpected non-zero "
1803 "s_reserved_gdt_blocks");
1807 /* Do a quick sanity check of the resize inode */
1808 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
- 9))
1809 goto invalid_resize_inode
;
1810 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1811 if (i
== EXT4_DIND_BLOCK
) {
1815 goto invalid_resize_inode
;
1818 goto invalid_resize_inode
;
1820 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1823 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credits
);
1825 return PTR_ERR(handle
);
1827 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1828 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1832 ext4_clear_feature_resize_inode(sb
);
1833 ext4_set_feature_meta_bg(sb
);
1834 sbi
->s_es
->s_first_meta_bg
=
1835 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1837 err
= ext4_handle_dirty_super(handle
, sb
);
1839 ext4_std_error(sb
, err
);
1844 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1845 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1846 EXT4_FREE_BLOCKS_METADATA
|
1847 EXT4_FREE_BLOCKS_FORGET
);
1848 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1849 inode
->i_blocks
= 0;
1851 err
= ext4_mark_inode_dirty(handle
, inode
);
1853 ext4_std_error(sb
, err
);
1857 ret
= ext4_journal_stop(handle
);
1862 invalid_resize_inode
:
1863 ext4_error(sb
, "corrupted/inconsistent resize inode");
1868 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1870 * @sb: super block of the fs to be resized
1871 * @n_blocks_count: the number of blocks resides in the resized fs
1873 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1875 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1876 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1877 struct ext4_super_block
*es
= sbi
->s_es
;
1878 struct buffer_head
*bh
;
1879 struct inode
*resize_inode
= NULL
;
1880 ext4_grpblk_t add
, offset
;
1881 unsigned long n_desc_blocks
;
1882 unsigned long o_desc_blocks
;
1883 ext4_group_t o_group
;
1884 ext4_group_t n_group
;
1885 ext4_fsblk_t o_blocks_count
;
1886 ext4_fsblk_t n_blocks_count_retry
= 0;
1887 unsigned long last_update_time
= 0;
1888 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1891 /* See if the device is actually as big as what was requested */
1892 bh
= sb_bread(sb
, n_blocks_count
- 1);
1894 ext4_warning(sb
, "can't read last block, resize aborted");
1900 o_blocks_count
= ext4_blocks_count(es
);
1902 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1903 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1905 if (n_blocks_count
< o_blocks_count
) {
1906 /* On-line shrinking not supported */
1907 ext4_warning(sb
, "can't shrink FS - resize aborted");
1911 if (n_blocks_count
== o_blocks_count
)
1912 /* Nothing need to do */
1915 n_group
= ext4_get_group_number(sb
, n_blocks_count
- 1);
1916 if (n_group
>= (0xFFFFFFFFUL
/ EXT4_INODES_PER_GROUP(sb
))) {
1917 ext4_warning(sb
, "resize would cause inodes_count overflow");
1920 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1922 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1923 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1925 meta_bg
= ext4_has_feature_meta_bg(sb
);
1927 if (ext4_has_feature_resize_inode(sb
)) {
1929 ext4_error(sb
, "resize_inode and meta_bg enabled "
1933 if (n_desc_blocks
> o_desc_blocks
+
1934 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1935 n_blocks_count_retry
= n_blocks_count
;
1936 n_desc_blocks
= o_desc_blocks
+
1937 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1938 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1939 n_blocks_count
= (ext4_fsblk_t
)n_group
*
1940 EXT4_BLOCKS_PER_GROUP(sb
) +
1941 le32_to_cpu(es
->s_first_data_block
);
1942 n_group
--; /* set to last group number */
1946 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1947 if (IS_ERR(resize_inode
)) {
1948 ext4_warning(sb
, "Error opening resize inode");
1949 return PTR_ERR(resize_inode
);
1953 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
1954 err
= ext4_convert_meta_bg(sb
, resize_inode
);
1959 resize_inode
= NULL
;
1961 if (n_blocks_count_retry
) {
1962 n_blocks_count
= n_blocks_count_retry
;
1963 n_blocks_count_retry
= 0;
1969 * Make sure the last group has enough space so that it's
1970 * guaranteed to have enough space for all metadata blocks
1971 * that it might need to hold. (We might not need to store
1972 * the inode table blocks in the last block group, but there
1973 * will be cases where this might be needed.)
1975 if ((ext4_group_first_block_no(sb
, n_group
) +
1976 ext4_group_overhead_blocks(sb
, n_group
) + 2 +
1977 sbi
->s_itb_per_group
+ sbi
->s_cluster_ratio
) >= n_blocks_count
) {
1978 n_blocks_count
= ext4_group_first_block_no(sb
, n_group
);
1980 n_blocks_count_retry
= 0;
1983 resize_inode
= NULL
;
1988 /* extend the last group */
1989 if (n_group
== o_group
)
1990 add
= n_blocks_count
- o_blocks_count
;
1992 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1994 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1999 if (ext4_blocks_count(es
) == n_blocks_count
)
2002 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
2006 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
2010 flex_gd
= alloc_flex_gd(flexbg_size
);
2011 if (flex_gd
== NULL
) {
2016 /* Add flex groups. Note that a regular group is a
2017 * flex group with 1 group.
2019 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
2021 if (jiffies
- last_update_time
> HZ
* 10) {
2022 if (last_update_time
)
2023 ext4_msg(sb
, KERN_INFO
,
2024 "resized to %llu blocks",
2025 ext4_blocks_count(es
));
2026 last_update_time
= jiffies
;
2028 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
2030 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
2035 if (!err
&& n_blocks_count_retry
) {
2036 n_blocks_count
= n_blocks_count_retry
;
2037 n_blocks_count_retry
= 0;
2038 free_flex_gd(flex_gd
);
2042 resize_inode
= NULL
;
2049 free_flex_gd(flex_gd
);
2050 if (resize_inode
!= NULL
)
2053 ext4_warning(sb
, "error (%d) occurred during "
2054 "file system resize", err
);
2055 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu",
2056 ext4_blocks_count(es
));