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"
25 static void ext4_rcu_ptr_callback(struct rcu_head
*head
)
27 struct ext4_rcu_ptr
*ptr
;
29 ptr
= container_of(head
, struct ext4_rcu_ptr
, rcu
);
34 void ext4_kvfree_array_rcu(void *to_free
)
36 struct ext4_rcu_ptr
*ptr
= kzalloc(sizeof(*ptr
), GFP_KERNEL
);
40 call_rcu(&ptr
->rcu
, ext4_rcu_ptr_callback
);
47 int ext4_resize_begin(struct super_block
*sb
)
49 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
52 if (!capable(CAP_SYS_RESOURCE
))
56 * If we are not using the primary superblock/GDT copy don't resize,
57 * because the user tools have no way of handling this. Probably a
58 * bad time to do it anyways.
60 if (EXT4_B2C(sbi
, sbi
->s_sbh
->b_blocknr
) !=
61 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
62 ext4_warning(sb
, "won't resize using backup superblock at %llu",
63 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
68 * We are not allowed to do online-resizing on a filesystem mounted
69 * with error, because it can destroy the filesystem easily.
71 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
72 ext4_warning(sb
, "There are errors in the filesystem, "
73 "so online resizing is not allowed");
77 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING
,
78 &EXT4_SB(sb
)->s_ext4_flags
))
84 void ext4_resize_end(struct super_block
*sb
)
86 clear_bit_unlock(EXT4_FLAGS_RESIZING
, &EXT4_SB(sb
)->s_ext4_flags
);
87 smp_mb__after_atomic();
90 static ext4_group_t
ext4_meta_bg_first_group(struct super_block
*sb
,
92 return (group
>> EXT4_DESC_PER_BLOCK_BITS(sb
)) <<
93 EXT4_DESC_PER_BLOCK_BITS(sb
);
96 static ext4_fsblk_t
ext4_meta_bg_first_block_no(struct super_block
*sb
,
98 group
= ext4_meta_bg_first_group(sb
, group
);
99 return ext4_group_first_block_no(sb
, group
);
102 static ext4_grpblk_t
ext4_group_overhead_blocks(struct super_block
*sb
,
103 ext4_group_t group
) {
104 ext4_grpblk_t overhead
;
105 overhead
= ext4_bg_num_gdb(sb
, group
);
106 if (ext4_bg_has_super(sb
, group
))
108 le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
112 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
113 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
115 static int verify_group_input(struct super_block
*sb
,
116 struct ext4_new_group_data
*input
)
118 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
119 struct ext4_super_block
*es
= sbi
->s_es
;
120 ext4_fsblk_t start
= ext4_blocks_count(es
);
121 ext4_fsblk_t end
= start
+ input
->blocks_count
;
122 ext4_group_t group
= input
->group
;
123 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
125 ext4_fsblk_t metaend
;
126 struct buffer_head
*bh
= NULL
;
127 ext4_grpblk_t free_blocks_count
, offset
;
130 if (group
!= sbi
->s_groups_count
) {
131 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
132 input
->group
, sbi
->s_groups_count
);
136 overhead
= ext4_group_overhead_blocks(sb
, group
);
137 metaend
= start
+ overhead
;
138 input
->free_clusters_count
= free_blocks_count
=
139 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
141 if (test_opt(sb
, DEBUG
))
142 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
143 "(%d free, %u reserved)\n",
144 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
145 "no-super", input
->group
, input
->blocks_count
,
146 free_blocks_count
, input
->reserved_blocks
);
148 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
150 ext4_warning(sb
, "Last group not full");
151 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
152 ext4_warning(sb
, "Reserved blocks too high (%u)",
153 input
->reserved_blocks
);
154 else if (free_blocks_count
< 0)
155 ext4_warning(sb
, "Bad blocks count %u",
156 input
->blocks_count
);
157 else if (IS_ERR(bh
= ext4_sb_bread(sb
, end
- 1, 0))) {
160 ext4_warning(sb
, "Cannot read last block (%llu)",
162 } else if (outside(input
->block_bitmap
, start
, end
))
163 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
164 (unsigned long long)input
->block_bitmap
);
165 else if (outside(input
->inode_bitmap
, start
, end
))
166 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
167 (unsigned long long)input
->inode_bitmap
);
168 else if (outside(input
->inode_table
, start
, end
) ||
169 outside(itend
- 1, start
, end
))
170 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
171 (unsigned long long)input
->inode_table
, itend
- 1);
172 else if (input
->inode_bitmap
== input
->block_bitmap
)
173 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
174 (unsigned long long)input
->block_bitmap
);
175 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
176 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
178 (unsigned long long)input
->block_bitmap
,
179 (unsigned long long)input
->inode_table
, itend
- 1);
180 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
181 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
183 (unsigned long long)input
->inode_bitmap
,
184 (unsigned long long)input
->inode_table
, itend
- 1);
185 else if (inside(input
->block_bitmap
, start
, metaend
))
186 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
187 (unsigned long long)input
->block_bitmap
,
189 else if (inside(input
->inode_bitmap
, start
, metaend
))
190 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
191 (unsigned long long)input
->inode_bitmap
,
193 else if (inside(input
->inode_table
, start
, metaend
) ||
194 inside(itend
- 1, start
, metaend
))
195 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
197 (unsigned long long)input
->inode_table
,
198 itend
- 1, start
, metaend
- 1);
207 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
210 struct ext4_new_flex_group_data
{
211 struct ext4_new_group_data
*groups
; /* new_group_data for groups
213 __u16
*bg_flags
; /* block group flags of groups
215 ext4_group_t count
; /* number of groups in @groups
220 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
223 * Returns NULL on failure otherwise address of the allocated structure.
225 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
227 struct ext4_new_flex_group_data
*flex_gd
;
229 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
233 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_group_data
))
235 flex_gd
->count
= flexbg_size
;
237 flex_gd
->groups
= kmalloc_array(flexbg_size
,
238 sizeof(struct ext4_new_group_data
),
240 if (flex_gd
->groups
== NULL
)
243 flex_gd
->bg_flags
= kmalloc_array(flexbg_size
, sizeof(__u16
),
245 if (flex_gd
->bg_flags
== NULL
)
251 kfree(flex_gd
->groups
);
258 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
260 kfree(flex_gd
->bg_flags
);
261 kfree(flex_gd
->groups
);
266 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
267 * and inode tables for a flex group.
269 * This function is used by 64bit-resize. Note that this function allocates
270 * group tables from the 1st group of groups contained by @flexgd, which may
271 * be a partial of a flex group.
273 * @sb: super block of fs to which the groups belongs
275 * Returns 0 on a successful allocation of the metadata blocks in the
278 static int ext4_alloc_group_tables(struct super_block
*sb
,
279 struct ext4_new_flex_group_data
*flex_gd
,
282 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
283 ext4_fsblk_t start_blk
;
284 ext4_fsblk_t last_blk
;
285 ext4_group_t src_group
;
286 ext4_group_t bb_index
= 0;
287 ext4_group_t ib_index
= 0;
288 ext4_group_t it_index
= 0;
290 ext4_group_t last_group
;
292 __u16 uninit_mask
= (flexbg_size
> 1) ? ~EXT4_BG_BLOCK_UNINIT
: ~0;
295 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
297 src_group
= group_data
[0].group
;
298 last_group
= src_group
+ flex_gd
->count
- 1;
300 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
301 (last_group
& ~(flexbg_size
- 1))));
303 group
= group_data
[0].group
;
304 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
306 start_blk
= ext4_group_first_block_no(sb
, src_group
);
307 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
309 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
311 start_blk
+= overhead
;
313 /* We collect contiguous blocks as much as possible. */
315 for (; src_group
<= last_group
; src_group
++) {
316 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
318 last_blk
+= group_data
[src_group
- group
].blocks_count
;
323 /* Allocate block bitmaps */
324 for (; bb_index
< flex_gd
->count
; bb_index
++) {
325 if (start_blk
>= last_blk
)
327 group_data
[bb_index
].block_bitmap
= start_blk
++;
328 group
= ext4_get_group_number(sb
, start_blk
- 1);
329 group
-= group_data
[0].group
;
330 group_data
[group
].mdata_blocks
++;
331 flex_gd
->bg_flags
[group
] &= uninit_mask
;
334 /* Allocate inode bitmaps */
335 for (; ib_index
< flex_gd
->count
; ib_index
++) {
336 if (start_blk
>= last_blk
)
338 group_data
[ib_index
].inode_bitmap
= start_blk
++;
339 group
= ext4_get_group_number(sb
, start_blk
- 1);
340 group
-= group_data
[0].group
;
341 group_data
[group
].mdata_blocks
++;
342 flex_gd
->bg_flags
[group
] &= uninit_mask
;
345 /* Allocate inode tables */
346 for (; it_index
< flex_gd
->count
; it_index
++) {
347 unsigned int itb
= EXT4_SB(sb
)->s_itb_per_group
;
348 ext4_fsblk_t next_group_start
;
350 if (start_blk
+ itb
> last_blk
)
352 group_data
[it_index
].inode_table
= start_blk
;
353 group
= ext4_get_group_number(sb
, start_blk
);
354 next_group_start
= ext4_group_first_block_no(sb
, group
+ 1);
355 group
-= group_data
[0].group
;
357 if (start_blk
+ itb
> next_group_start
) {
358 flex_gd
->bg_flags
[group
+ 1] &= uninit_mask
;
359 overhead
= start_blk
+ itb
- next_group_start
;
360 group_data
[group
+ 1].mdata_blocks
+= overhead
;
364 group_data
[group
].mdata_blocks
+= itb
;
365 flex_gd
->bg_flags
[group
] &= uninit_mask
;
366 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
369 /* Update free clusters count to exclude metadata blocks */
370 for (i
= 0; i
< flex_gd
->count
; i
++) {
371 group_data
[i
].free_clusters_count
-=
372 EXT4_NUM_B2C(EXT4_SB(sb
),
373 group_data
[i
].mdata_blocks
);
376 if (test_opt(sb
, DEBUG
)) {
378 group
= group_data
[0].group
;
380 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
381 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
384 for (i
= 0; i
< flex_gd
->count
; i
++) {
386 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
387 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
388 "no-super", group
+ i
,
389 group_data
[i
].blocks_count
,
390 group_data
[i
].free_clusters_count
,
391 group_data
[i
].mdata_blocks
);
397 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
400 struct buffer_head
*bh
;
403 bh
= sb_getblk(sb
, blk
);
405 return ERR_PTR(-ENOMEM
);
406 BUFFER_TRACE(bh
, "get_write_access");
407 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
411 memset(bh
->b_data
, 0, sb
->s_blocksize
);
412 set_buffer_uptodate(bh
);
418 static int ext4_resize_ensure_credits_batch(handle_t
*handle
, int credits
)
420 return ext4_journal_ensure_credits_fn(handle
, credits
,
421 EXT4_MAX_TRANS_DATA
, 0, 0);
425 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
427 * Helper function for ext4_setup_new_group_blocks() which set .
430 * @handle: journal handle
431 * @flex_gd: flex group data
433 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
434 struct ext4_new_flex_group_data
*flex_gd
,
435 ext4_fsblk_t first_cluster
, ext4_fsblk_t last_cluster
)
437 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
438 ext4_group_t count
= last_cluster
- first_cluster
+ 1;
441 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster
,
443 for (count2
= count
; count
> 0;
444 count
-= count2
, first_cluster
+= count2
) {
446 struct buffer_head
*bh
;
450 group
= ext4_get_group_number(sb
, EXT4_C2B(sbi
, first_cluster
));
451 start
= EXT4_B2C(sbi
, ext4_group_first_block_no(sb
, group
));
452 group
-= flex_gd
->groups
[0].group
;
454 count2
= EXT4_CLUSTERS_PER_GROUP(sb
) - (first_cluster
- start
);
458 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
459 BUG_ON(flex_gd
->count
> 1);
463 err
= ext4_resize_ensure_credits_batch(handle
, 1);
467 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
471 BUFFER_TRACE(bh
, "get_write_access");
472 err
= ext4_journal_get_write_access(handle
, bh
);
477 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
478 first_cluster
, first_cluster
- start
, count2
);
479 ext4_set_bits(bh
->b_data
, first_cluster
- start
, count2
);
481 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
491 * Set up the block and inode bitmaps, and the inode table for the new groups.
492 * This doesn't need to be part of the main transaction, since we are only
493 * changing blocks outside the actual filesystem. We still do journaling to
494 * ensure the recovery is correct in case of a failure just after resize.
495 * If any part of this fails, we simply abort the resize.
497 * setup_new_flex_group_blocks handles a flex group as follow:
498 * 1. copy super block and GDT, and initialize group tables if necessary.
499 * In this step, we only set bits in blocks bitmaps for blocks taken by
500 * super block and GDT.
501 * 2. allocate group tables in block bitmaps, that is, set bits in block
502 * bitmap for blocks taken by group tables.
504 static int setup_new_flex_group_blocks(struct super_block
*sb
,
505 struct ext4_new_flex_group_data
*flex_gd
)
507 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
510 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
511 struct ext4_super_block
*es
= sbi
->s_es
;
512 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
513 __u16
*bg_flags
= flex_gd
->bg_flags
;
515 ext4_group_t group
, count
;
516 struct buffer_head
*bh
= NULL
;
517 int reserved_gdb
, i
, j
, err
= 0, err2
;
520 BUG_ON(!flex_gd
->count
|| !group_data
||
521 group_data
[0].group
!= sbi
->s_groups_count
);
523 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
524 meta_bg
= ext4_has_feature_meta_bg(sb
);
526 /* This transaction may be extended/restarted along the way */
527 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
529 return PTR_ERR(handle
);
531 group
= group_data
[0].group
;
532 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
533 unsigned long gdblocks
;
534 ext4_grpblk_t overhead
;
536 gdblocks
= ext4_bg_num_gdb(sb
, group
);
537 start
= ext4_group_first_block_no(sb
, group
);
539 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
543 ext4_group_t first_group
;
544 first_group
= ext4_meta_bg_first_group(sb
, group
);
545 if (first_group
!= group
+ 1 &&
546 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
550 block
= start
+ ext4_bg_has_super(sb
, group
);
551 /* Copy all of the GDT blocks into the backup in this group */
552 for (j
= 0; j
< gdblocks
; j
++, block
++) {
553 struct buffer_head
*gdb
;
555 ext4_debug("update backup group %#04llx\n", block
);
556 err
= ext4_resize_ensure_credits_batch(handle
, 1);
560 gdb
= sb_getblk(sb
, block
);
561 if (unlikely(!gdb
)) {
566 BUFFER_TRACE(gdb
, "get_write_access");
567 err
= ext4_journal_get_write_access(handle
, gdb
);
572 memcpy(gdb
->b_data
, sbi_array_rcu_deref(sbi
,
573 s_group_desc
, j
)->b_data
, gdb
->b_size
);
574 set_buffer_uptodate(gdb
);
576 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
584 /* Zero out all of the reserved backup group descriptor
587 if (ext4_bg_has_super(sb
, group
)) {
588 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
589 reserved_gdb
, GFP_NOFS
);
595 /* Initialize group tables of the grop @group */
596 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
599 /* Zero out all of the inode table blocks */
600 block
= group_data
[i
].inode_table
;
601 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
602 block
, sbi
->s_itb_per_group
);
603 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
609 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
612 /* Initialize block bitmap of the @group */
613 block
= group_data
[i
].block_bitmap
;
614 err
= ext4_resize_ensure_credits_batch(handle
, 1);
618 bh
= bclean(handle
, sb
, block
);
623 overhead
= ext4_group_overhead_blocks(sb
, group
);
625 ext4_debug("mark backup superblock %#04llx (+0)\n",
627 ext4_set_bits(bh
->b_data
, 0,
628 EXT4_NUM_B2C(sbi
, overhead
));
630 ext4_mark_bitmap_end(EXT4_B2C(sbi
, group_data
[i
].blocks_count
),
631 sb
->s_blocksize
* 8, bh
->b_data
);
632 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
638 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
641 /* Initialize inode bitmap of the @group */
642 block
= group_data
[i
].inode_bitmap
;
643 err
= ext4_resize_ensure_credits_batch(handle
, 1);
646 /* Mark unused entries in inode bitmap used */
647 bh
= bclean(handle
, sb
, block
);
653 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
654 sb
->s_blocksize
* 8, bh
->b_data
);
655 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
661 /* Mark group tables in block bitmap */
662 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
663 count
= group_table_count
[j
];
664 start
= (&group_data
[0].block_bitmap
)[j
];
666 for (i
= 1; i
< flex_gd
->count
; i
++) {
667 block
+= group_table_count
[j
];
668 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
669 count
+= group_table_count
[j
];
672 err
= set_flexbg_block_bitmap(sb
, handle
,
674 EXT4_B2C(sbi
, start
),
680 count
= group_table_count
[j
];
681 start
= (&group_data
[i
].block_bitmap
)[j
];
686 err
= set_flexbg_block_bitmap(sb
, handle
,
688 EXT4_B2C(sbi
, start
),
698 err2
= ext4_journal_stop(handle
);
706 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
707 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
708 * calling this for the first time. In a sparse filesystem it will be the
709 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
710 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
712 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
713 unsigned *five
, unsigned *seven
)
715 unsigned *min
= three
;
719 if (!ext4_has_feature_sparse_super(sb
)) {
741 * Check that all of the backup GDT blocks are held in the primary GDT block.
742 * It is assumed that they are stored in group order. Returns the number of
743 * groups in current filesystem that have BACKUPS, or -ve error code.
745 static int verify_reserved_gdb(struct super_block
*sb
,
747 struct buffer_head
*primary
)
749 const ext4_fsblk_t blk
= primary
->b_blocknr
;
754 __le32
*p
= (__le32
*)primary
->b_data
;
757 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
758 if (le32_to_cpu(*p
++) !=
759 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
760 ext4_warning(sb
, "reserved GDT %llu"
761 " missing grp %d (%llu)",
764 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
768 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
776 * Called when we need to bring a reserved group descriptor table block into
777 * use from the resize inode. The primary copy of the new GDT block currently
778 * is an indirect block (under the double indirect block in the resize inode).
779 * The new backup GDT blocks will be stored as leaf blocks in this indirect
780 * block, in group order. Even though we know all the block numbers we need,
781 * we check to ensure that the resize inode has actually reserved these blocks.
783 * Don't need to update the block bitmaps because the blocks are still in use.
785 * We get all of the error cases out of the way, so that we are sure to not
786 * fail once we start modifying the data on disk, because JBD has no rollback.
788 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
791 struct super_block
*sb
= inode
->i_sb
;
792 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
793 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
794 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
795 struct buffer_head
**o_group_desc
, **n_group_desc
= NULL
;
796 struct buffer_head
*dind
= NULL
;
797 struct buffer_head
*gdb_bh
= NULL
;
799 struct ext4_iloc iloc
= { .bh
= NULL
};
803 if (test_opt(sb
, DEBUG
))
805 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
808 gdb_bh
= ext4_sb_bread(sb
, gdblock
, 0);
810 return PTR_ERR(gdb_bh
);
812 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
818 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
819 dind
= ext4_sb_bread(sb
, le32_to_cpu(*data
), 0);
826 data
= (__le32
*)dind
->b_data
;
827 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
828 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
834 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
835 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
839 BUFFER_TRACE(gdb_bh
, "get_write_access");
840 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
844 BUFFER_TRACE(dind
, "get_write_access");
845 err
= ext4_journal_get_write_access(handle
, dind
);
847 ext4_std_error(sb
, err
);
851 /* ext4_reserve_inode_write() gets a reference on the iloc */
852 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
856 n_group_desc
= kvmalloc((gdb_num
+ 1) * sizeof(struct buffer_head
*),
860 ext4_warning(sb
, "not enough memory for %lu groups",
866 * Finally, we have all of the possible failures behind us...
868 * Remove new GDT block from inode double-indirect block and clear out
869 * the new GDT block for use (which also "frees" the backup GDT blocks
870 * from the reserved inode). We don't need to change the bitmaps for
871 * these blocks, because they are marked as in-use from being in the
872 * reserved inode, and will become GDT blocks (primary and backup).
874 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
875 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
877 ext4_std_error(sb
, err
);
880 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>>
881 (9 - EXT4_SB(sb
)->s_cluster_bits
);
882 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
883 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
884 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
886 ext4_std_error(sb
, err
);
893 o_group_desc
= rcu_dereference(EXT4_SB(sb
)->s_group_desc
);
894 memcpy(n_group_desc
, o_group_desc
,
895 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
897 n_group_desc
[gdb_num
] = gdb_bh
;
898 rcu_assign_pointer(EXT4_SB(sb
)->s_group_desc
, n_group_desc
);
899 EXT4_SB(sb
)->s_gdb_count
++;
900 ext4_kvfree_array_rcu(o_group_desc
);
902 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
903 err
= ext4_handle_dirty_super(handle
, sb
);
905 ext4_std_error(sb
, err
);
908 kvfree(n_group_desc
);
913 ext4_debug("leaving with error %d\n", err
);
918 * add_new_gdb_meta_bg is the sister of add_new_gdb.
920 static int add_new_gdb_meta_bg(struct super_block
*sb
,
921 handle_t
*handle
, ext4_group_t group
) {
922 ext4_fsblk_t gdblock
;
923 struct buffer_head
*gdb_bh
;
924 struct buffer_head
**o_group_desc
, **n_group_desc
;
925 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
928 gdblock
= ext4_meta_bg_first_block_no(sb
, group
) +
929 ext4_bg_has_super(sb
, group
);
930 gdb_bh
= ext4_sb_bread(sb
, gdblock
, 0);
932 return PTR_ERR(gdb_bh
);
933 n_group_desc
= kvmalloc((gdb_num
+ 1) * sizeof(struct buffer_head
*),
938 ext4_warning(sb
, "not enough memory for %lu groups",
944 o_group_desc
= rcu_dereference(EXT4_SB(sb
)->s_group_desc
);
945 memcpy(n_group_desc
, o_group_desc
,
946 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
948 n_group_desc
[gdb_num
] = gdb_bh
;
950 BUFFER_TRACE(gdb_bh
, "get_write_access");
951 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
953 kvfree(n_group_desc
);
958 rcu_assign_pointer(EXT4_SB(sb
)->s_group_desc
, n_group_desc
);
959 EXT4_SB(sb
)->s_gdb_count
++;
960 ext4_kvfree_array_rcu(o_group_desc
);
965 * Called when we are adding a new group which has a backup copy of each of
966 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
967 * We need to add these reserved backup GDT blocks to the resize inode, so
968 * that they are kept for future resizing and not allocated to files.
970 * Each reserved backup GDT block will go into a different indirect block.
971 * The indirect blocks are actually the primary reserved GDT blocks,
972 * so we know in advance what their block numbers are. We only get the
973 * double-indirect block to verify it is pointing to the primary reserved
974 * GDT blocks so we don't overwrite a data block by accident. The reserved
975 * backup GDT blocks are stored in their reserved primary GDT block.
977 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
980 struct super_block
*sb
= inode
->i_sb
;
981 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
982 int cluster_bits
= EXT4_SB(sb
)->s_cluster_bits
;
983 struct buffer_head
**primary
;
984 struct buffer_head
*dind
;
985 struct ext4_iloc iloc
;
992 primary
= kmalloc_array(reserved_gdb
, sizeof(*primary
), GFP_NOFS
);
996 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
997 dind
= ext4_sb_bread(sb
, le32_to_cpu(*data
), 0);
1004 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
1005 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
1006 EXT4_ADDR_PER_BLOCK(sb
));
1007 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
1009 /* Get each reserved primary GDT block and verify it holds backups */
1010 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
1011 if (le32_to_cpu(*data
) != blk
) {
1012 ext4_warning(sb
, "reserved block %llu"
1013 " not at offset %ld",
1015 (long)(data
- (__le32
*)dind
->b_data
));
1019 primary
[res
] = ext4_sb_bread(sb
, blk
, 0);
1020 if (IS_ERR(primary
[res
])) {
1021 err
= PTR_ERR(primary
[res
]);
1022 primary
[res
] = NULL
;
1025 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
1026 if (gdbackups
< 0) {
1027 brelse(primary
[res
]);
1032 data
= (__le32
*)dind
->b_data
;
1035 for (i
= 0; i
< reserved_gdb
; i
++) {
1036 BUFFER_TRACE(primary
[i
], "get_write_access");
1037 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
])))
1041 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
1045 * Finally we can add each of the reserved backup GDT blocks from
1046 * the new group to its reserved primary GDT block.
1048 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1049 for (i
= 0; i
< reserved_gdb
; i
++) {
1051 data
= (__le32
*)primary
[i
]->b_data
;
1052 /* printk("reserving backup %lu[%u] = %lu\n",
1053 primary[i]->b_blocknr, gdbackups,
1054 blk + primary[i]->b_blocknr); */
1055 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1056 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1061 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> (9 - cluster_bits
);
1062 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1066 brelse(primary
[res
]);
1076 * Update the backup copies of the ext4 metadata. These don't need to be part
1077 * of the main resize transaction, because e2fsck will re-write them if there
1078 * is a problem (basically only OOM will cause a problem). However, we
1079 * _should_ update the backups if possible, in case the primary gets trashed
1080 * for some reason and we need to run e2fsck from a backup superblock. The
1081 * important part is that the new block and inode counts are in the backup
1082 * superblocks, and the location of the new group metadata in the GDT backups.
1084 * We do not need take the s_resize_lock for this, because these
1085 * blocks are not otherwise touched by the filesystem code when it is
1086 * mounted. We don't need to worry about last changing from
1087 * sbi->s_groups_count, because the worst that can happen is that we
1088 * do not copy the full number of backups at this time. The resize
1089 * which changed s_groups_count will backup again.
1091 static void update_backups(struct super_block
*sb
, sector_t blk_off
, char *data
,
1092 int size
, int meta_bg
)
1094 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1096 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1100 ext4_group_t group
= 0;
1101 int rest
= sb
->s_blocksize
- size
;
1105 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
1106 if (IS_ERR(handle
)) {
1108 err
= PTR_ERR(handle
);
1113 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1114 last
= sbi
->s_groups_count
;
1116 group
= ext4_get_group_number(sb
, blk_off
) + 1;
1117 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1120 while (group
< sbi
->s_groups_count
) {
1121 struct buffer_head
*bh
;
1122 ext4_fsblk_t backup_block
;
1124 /* Out of journal space, and can't get more - abort - so sad */
1125 err
= ext4_resize_ensure_credits_batch(handle
, 1);
1130 backup_block
= ((ext4_fsblk_t
)group
) * bpg
+ blk_off
;
1132 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1133 ext4_bg_has_super(sb
, group
));
1135 bh
= sb_getblk(sb
, backup_block
);
1136 if (unlikely(!bh
)) {
1140 ext4_debug("update metadata backup %llu(+%llu)\n",
1141 backup_block
, backup_block
-
1142 ext4_group_first_block_no(sb
, group
));
1143 BUFFER_TRACE(bh
, "get_write_access");
1144 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
1149 memcpy(bh
->b_data
, data
, size
);
1151 memset(bh
->b_data
+ size
, 0, rest
);
1152 set_buffer_uptodate(bh
);
1154 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1156 ext4_std_error(sb
, err
);
1160 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1161 else if (group
== last
)
1166 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1170 * Ugh! Need to have e2fsck write the backup copies. It is too
1171 * late to revert the resize, we shouldn't fail just because of
1172 * the backup copies (they are only needed in case of corruption).
1174 * However, if we got here we have a journal problem too, so we
1175 * can't really start a transaction to mark the superblock.
1176 * Chicken out and just set the flag on the hope it will be written
1177 * to disk, and if not - we will simply wait until next fsck.
1181 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1182 "forcing fsck on next reboot", group
, err
);
1183 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1184 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1185 mark_buffer_dirty(sbi
->s_sbh
);
1190 * ext4_add_new_descs() adds @count group descriptor of groups
1191 * starting at @group
1193 * @handle: journal handle
1195 * @group: the group no. of the first group desc to be added
1196 * @resize_inode: the resize inode
1197 * @count: number of group descriptors to be added
1199 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1200 ext4_group_t group
, struct inode
*resize_inode
,
1203 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1204 struct ext4_super_block
*es
= sbi
->s_es
;
1205 struct buffer_head
*gdb_bh
;
1206 int i
, gdb_off
, gdb_num
, err
= 0;
1209 meta_bg
= ext4_has_feature_meta_bg(sb
);
1210 for (i
= 0; i
< count
; i
++, group
++) {
1211 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1212 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1214 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1215 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1218 * We will only either add reserved group blocks to a backup group
1219 * or remove reserved blocks for the first group in a new group block.
1220 * Doing both would be mean more complex code, and sane people don't
1221 * use non-sparse filesystems anymore. This is already checked above.
1224 gdb_bh
= sbi_array_rcu_deref(sbi
, s_group_desc
,
1226 BUFFER_TRACE(gdb_bh
, "get_write_access");
1227 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1229 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1230 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1231 } else if (meta_bg
!= 0) {
1232 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1234 err
= add_new_gdb(handle
, resize_inode
, group
);
1242 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1244 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1247 if (!bh_uptodate_or_lock(bh
)) {
1248 if (ext4_read_bh(bh
, 0, NULL
) < 0) {
1257 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1259 struct ext4_group_desc
*gdp
,
1260 struct ext4_new_group_data
*group_data
)
1262 struct buffer_head
*bh
;
1264 if (!ext4_has_metadata_csum(sb
))
1267 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1270 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1271 EXT4_INODES_PER_GROUP(sb
) / 8);
1274 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1277 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1284 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1286 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1287 struct ext4_new_flex_group_data
*flex_gd
)
1289 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1290 struct ext4_group_desc
*gdp
;
1291 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1292 struct buffer_head
*gdb_bh
;
1294 __u16
*bg_flags
= flex_gd
->bg_flags
;
1295 int i
, gdb_off
, gdb_num
, err
= 0;
1298 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1299 group
= group_data
->group
;
1301 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1302 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1305 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1307 gdb_bh
= sbi_array_rcu_deref(sbi
, s_group_desc
, gdb_num
);
1308 /* Update group descriptor block for new group */
1309 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1310 gdb_off
* EXT4_DESC_SIZE(sb
));
1312 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1313 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1314 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1315 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1317 ext4_std_error(sb
, err
);
1321 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1322 ext4_free_group_clusters_set(sb
, gdp
,
1323 group_data
->free_clusters_count
);
1324 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1325 if (ext4_has_group_desc_csum(sb
))
1326 ext4_itable_unused_set(sb
, gdp
,
1327 EXT4_INODES_PER_GROUP(sb
));
1328 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1329 ext4_group_desc_csum_set(sb
, group
, gdp
);
1331 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1332 if (unlikely(err
)) {
1333 ext4_std_error(sb
, err
);
1338 * We can allocate memory for mb_alloc based on the new group
1341 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1349 * ext4_update_super() updates the super block so that the newly added
1350 * groups can be seen by the filesystem.
1353 * @flex_gd: new added groups
1355 static void ext4_update_super(struct super_block
*sb
,
1356 struct ext4_new_flex_group_data
*flex_gd
)
1358 ext4_fsblk_t blocks_count
= 0;
1359 ext4_fsblk_t free_blocks
= 0;
1360 ext4_fsblk_t reserved_blocks
= 0;
1361 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1362 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1363 struct ext4_super_block
*es
= sbi
->s_es
;
1366 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1368 * Make the new blocks and inodes valid next. We do this before
1369 * increasing the group count so that once the group is enabled,
1370 * all of its blocks and inodes are already valid.
1372 * We always allocate group-by-group, then block-by-block or
1373 * inode-by-inode within a group, so enabling these
1374 * blocks/inodes before the group is live won't actually let us
1375 * allocate the new space yet.
1377 for (i
= 0; i
< flex_gd
->count
; i
++) {
1378 blocks_count
+= group_data
[i
].blocks_count
;
1379 free_blocks
+= EXT4_C2B(sbi
, group_data
[i
].free_clusters_count
);
1382 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1383 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1384 reserved_blocks
*= blocks_count
;
1385 do_div(reserved_blocks
, 100);
1387 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1388 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1389 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1391 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1394 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1396 * We need to protect s_groups_count against other CPUs seeing
1397 * inconsistent state in the superblock.
1399 * The precise rules we use are:
1401 * * Writers must perform a smp_wmb() after updating all
1402 * dependent data and before modifying the groups count
1404 * * Readers must perform an smp_rmb() after reading the groups
1405 * count and before reading any dependent data.
1407 * NB. These rules can be relaxed when checking the group count
1408 * while freeing data, as we can only allocate from a block
1409 * group after serialising against the group count, and we can
1410 * only then free after serialising in turn against that
1415 /* Update the global fs size fields */
1416 sbi
->s_groups_count
+= flex_gd
->count
;
1417 sbi
->s_blockfile_groups
= min_t(ext4_group_t
, sbi
->s_groups_count
,
1418 (EXT4_MAX_BLOCK_FILE_PHYS
/ EXT4_BLOCKS_PER_GROUP(sb
)));
1420 /* Update the reserved block counts only once the new group is
1422 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1425 /* Update the free space counts */
1426 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1427 EXT4_NUM_B2C(sbi
, free_blocks
));
1428 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1429 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1431 ext4_debug("free blocks count %llu",
1432 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1433 if (ext4_has_feature_flex_bg(sb
) && sbi
->s_log_groups_per_flex
) {
1434 ext4_group_t flex_group
;
1435 struct flex_groups
*fg
;
1437 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1438 fg
= sbi_array_rcu_deref(sbi
, s_flex_groups
, flex_group
);
1439 atomic64_add(EXT4_NUM_B2C(sbi
, free_blocks
),
1440 &fg
->free_clusters
);
1441 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1446 * Update the fs overhead information
1448 ext4_calculate_overhead(sb
);
1450 if (test_opt(sb
, DEBUG
))
1451 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1452 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1453 blocks_count
, free_blocks
, reserved_blocks
);
1456 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1457 * _before_ we start modifying the filesystem, because we cannot abort the
1458 * transaction and not have it write the data to disk.
1460 static int ext4_flex_group_add(struct super_block
*sb
,
1461 struct inode
*resize_inode
,
1462 struct ext4_new_flex_group_data
*flex_gd
)
1464 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1465 struct ext4_super_block
*es
= sbi
->s_es
;
1466 ext4_fsblk_t o_blocks_count
;
1470 unsigned reserved_gdb
;
1471 int err
= 0, err2
= 0, credit
;
1473 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1475 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1476 o_blocks_count
= ext4_blocks_count(es
);
1477 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1480 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1484 * We will always be modifying at least the superblock and GDT
1485 * blocks. If we are adding a group past the last current GDT block,
1486 * we will also modify the inode and the dindirect block. If we
1487 * are adding a group with superblock/GDT backups we will also
1488 * modify each of the reserved GDT dindirect blocks.
1490 credit
= 3; /* sb, resize inode, resize inode dindirect */
1492 credit
+= 1 + DIV_ROUND_UP(flex_gd
->count
, EXT4_DESC_PER_BLOCK(sb
));
1493 credit
+= reserved_gdb
; /* Reserved GDT dindirect blocks */
1494 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credit
);
1495 if (IS_ERR(handle
)) {
1496 err
= PTR_ERR(handle
);
1500 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1501 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1505 group
= flex_gd
->groups
[0].group
;
1506 BUG_ON(group
!= sbi
->s_groups_count
);
1507 err
= ext4_add_new_descs(handle
, sb
, group
,
1508 resize_inode
, flex_gd
->count
);
1512 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1516 ext4_update_super(sb
, flex_gd
);
1518 err
= ext4_handle_dirty_super(handle
, sb
);
1521 err2
= ext4_journal_stop(handle
);
1526 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1527 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1528 EXT4_DESC_PER_BLOCK(sb
));
1529 int meta_bg
= ext4_has_feature_meta_bg(sb
);
1530 sector_t old_gdb
= 0;
1532 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1533 sizeof(struct ext4_super_block
), 0);
1534 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1535 struct buffer_head
*gdb_bh
;
1537 gdb_bh
= sbi_array_rcu_deref(sbi
, s_group_desc
,
1539 if (old_gdb
== gdb_bh
->b_blocknr
)
1541 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1542 gdb_bh
->b_size
, meta_bg
);
1543 old_gdb
= gdb_bh
->b_blocknr
;
1550 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1551 struct ext4_new_flex_group_data
*flex_gd
,
1552 ext4_fsblk_t n_blocks_count
,
1553 unsigned long flexbg_size
)
1555 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1556 struct ext4_super_block
*es
= sbi
->s_es
;
1557 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1558 ext4_fsblk_t o_blocks_count
;
1559 ext4_group_t n_group
;
1561 ext4_group_t last_group
;
1563 ext4_grpblk_t clusters_per_group
;
1566 clusters_per_group
= EXT4_CLUSTERS_PER_GROUP(sb
);
1568 o_blocks_count
= ext4_blocks_count(es
);
1570 if (o_blocks_count
== n_blocks_count
)
1573 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1575 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1577 last_group
= group
| (flexbg_size
- 1);
1578 if (last_group
> n_group
)
1579 last_group
= n_group
;
1581 flex_gd
->count
= last_group
- group
+ 1;
1583 for (i
= 0; i
< flex_gd
->count
; i
++) {
1586 group_data
[i
].group
= group
+ i
;
1587 group_data
[i
].blocks_count
= EXT4_BLOCKS_PER_GROUP(sb
);
1588 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1589 group_data
[i
].mdata_blocks
= overhead
;
1590 group_data
[i
].free_clusters_count
= EXT4_CLUSTERS_PER_GROUP(sb
);
1591 if (ext4_has_group_desc_csum(sb
)) {
1592 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1593 EXT4_BG_INODE_UNINIT
;
1594 if (!test_opt(sb
, INIT_INODE_TABLE
))
1595 flex_gd
->bg_flags
[i
] |= EXT4_BG_INODE_ZEROED
;
1597 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1600 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1601 /* We need to initialize block bitmap of last group. */
1602 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1604 if ((last_group
== n_group
) && (last
!= clusters_per_group
- 1)) {
1605 group_data
[i
- 1].blocks_count
= EXT4_C2B(sbi
, last
+ 1);
1606 group_data
[i
- 1].free_clusters_count
-= clusters_per_group
-
1613 /* Add group descriptor data to an existing or new group descriptor block.
1614 * Ensure we handle all possible error conditions _before_ we start modifying
1615 * the filesystem, because we cannot abort the transaction and not have it
1616 * write the data to disk.
1618 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1619 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1621 * We only need to hold the superblock lock while we are actually adding
1622 * in the new group's counts to the superblock. Prior to that we have
1623 * not really "added" the group at all. We re-check that we are still
1624 * adding in the last group in case things have changed since verifying.
1626 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1628 struct ext4_new_flex_group_data flex_gd
;
1629 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1630 struct ext4_super_block
*es
= sbi
->s_es
;
1631 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1632 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1633 struct inode
*inode
= NULL
;
1638 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1640 if (gdb_off
== 0 && !ext4_has_feature_sparse_super(sb
)) {
1641 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1645 if (ext4_blocks_count(es
) + input
->blocks_count
<
1646 ext4_blocks_count(es
)) {
1647 ext4_warning(sb
, "blocks_count overflow");
1651 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1652 le32_to_cpu(es
->s_inodes_count
)) {
1653 ext4_warning(sb
, "inodes_count overflow");
1657 if (reserved_gdb
|| gdb_off
== 0) {
1658 if (!ext4_has_feature_resize_inode(sb
) ||
1659 !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1661 "No reserved GDT blocks, can't resize");
1664 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
, EXT4_IGET_SPECIAL
);
1665 if (IS_ERR(inode
)) {
1666 ext4_warning(sb
, "Error opening resize inode");
1667 return PTR_ERR(inode
);
1672 err
= verify_group_input(sb
, input
);
1676 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1680 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1685 flex_gd
.groups
= input
;
1686 flex_gd
.bg_flags
= &bg_flags
;
1687 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1691 } /* ext4_group_add */
1694 * extend a group without checking assuming that checking has been done.
1696 static int ext4_group_extend_no_check(struct super_block
*sb
,
1697 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1699 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1703 /* We will update the superblock, one block bitmap, and
1704 * one group descriptor via ext4_group_add_blocks().
1706 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, 3);
1707 if (IS_ERR(handle
)) {
1708 err
= PTR_ERR(handle
);
1709 ext4_warning(sb
, "error %d on journal start", err
);
1713 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1714 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1716 ext4_warning(sb
, "error %d on journal write access", err
);
1720 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1721 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1722 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1723 o_blocks_count
+ add
);
1724 /* We add the blocks to the bitmap and set the group need init bit */
1725 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1728 ext4_handle_dirty_super(handle
, sb
);
1729 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1730 o_blocks_count
+ add
);
1732 err2
= ext4_journal_stop(handle
);
1737 if (test_opt(sb
, DEBUG
))
1738 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1739 "blocks\n", ext4_blocks_count(es
));
1740 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
,
1741 (char *)es
, sizeof(struct ext4_super_block
), 0);
1747 * Extend the filesystem to the new number of blocks specified. This entry
1748 * point is only used to extend the current filesystem to the end of the last
1749 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1750 * for emergencies (because it has no dependencies on reserved blocks).
1752 * If we _really_ wanted, we could use default values to call ext4_group_add()
1753 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1754 * GDT blocks are reserved to grow to the desired size.
1756 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1757 ext4_fsblk_t n_blocks_count
)
1759 ext4_fsblk_t o_blocks_count
;
1762 struct buffer_head
*bh
;
1766 o_blocks_count
= ext4_blocks_count(es
);
1768 if (test_opt(sb
, DEBUG
))
1769 ext4_msg(sb
, KERN_DEBUG
,
1770 "extending last group from %llu to %llu blocks",
1771 o_blocks_count
, n_blocks_count
);
1773 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1776 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1777 ext4_msg(sb
, KERN_ERR
,
1778 "filesystem too large to resize to %llu blocks safely",
1783 if (n_blocks_count
< o_blocks_count
) {
1784 ext4_warning(sb
, "can't shrink FS - resize aborted");
1788 /* Handle the remaining blocks in the last group only. */
1789 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1792 ext4_warning(sb
, "need to use ext2online to resize further");
1796 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1798 if (o_blocks_count
+ add
< o_blocks_count
) {
1799 ext4_warning(sb
, "blocks_count overflow");
1803 if (o_blocks_count
+ add
> n_blocks_count
)
1804 add
= n_blocks_count
- o_blocks_count
;
1806 if (o_blocks_count
+ add
< n_blocks_count
)
1807 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1808 o_blocks_count
+ add
, add
);
1810 /* See if the device is actually as big as what was requested */
1811 bh
= ext4_sb_bread(sb
, o_blocks_count
+ add
- 1, 0);
1813 ext4_warning(sb
, "can't read last block, resize aborted");
1818 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1820 } /* ext4_group_extend */
1823 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1825 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1829 * Release the resize inode and drop the resize_inode feature if there
1830 * are no more reserved gdt blocks, and then convert the file system
1833 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1836 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1837 struct ext4_super_block
*es
= sbi
->s_es
;
1838 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1840 int i
, ret
, err
= 0;
1843 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1845 if (es
->s_reserved_gdt_blocks
) {
1846 ext4_error(sb
, "Unexpected non-zero "
1847 "s_reserved_gdt_blocks");
1851 /* Do a quick sanity check of the resize inode */
1852 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
-
1853 (9 - sbi
->s_cluster_bits
)))
1854 goto invalid_resize_inode
;
1855 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1856 if (i
== EXT4_DIND_BLOCK
) {
1860 goto invalid_resize_inode
;
1863 goto invalid_resize_inode
;
1865 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1868 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credits
);
1870 return PTR_ERR(handle
);
1872 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1873 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1877 ext4_clear_feature_resize_inode(sb
);
1878 ext4_set_feature_meta_bg(sb
);
1879 sbi
->s_es
->s_first_meta_bg
=
1880 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1882 err
= ext4_handle_dirty_super(handle
, sb
);
1884 ext4_std_error(sb
, err
);
1889 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1890 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1891 EXT4_FREE_BLOCKS_METADATA
|
1892 EXT4_FREE_BLOCKS_FORGET
);
1893 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1894 inode
->i_blocks
= 0;
1896 err
= ext4_mark_inode_dirty(handle
, inode
);
1898 ext4_std_error(sb
, err
);
1902 ret
= ext4_journal_stop(handle
);
1907 invalid_resize_inode
:
1908 ext4_error(sb
, "corrupted/inconsistent resize inode");
1913 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1915 * @sb: super block of the fs to be resized
1916 * @n_blocks_count: the number of blocks resides in the resized fs
1918 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1920 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1921 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1922 struct ext4_super_block
*es
= sbi
->s_es
;
1923 struct buffer_head
*bh
;
1924 struct inode
*resize_inode
= NULL
;
1925 ext4_grpblk_t add
, offset
;
1926 unsigned long n_desc_blocks
;
1927 unsigned long o_desc_blocks
;
1928 ext4_group_t o_group
;
1929 ext4_group_t n_group
;
1930 ext4_fsblk_t o_blocks_count
;
1931 ext4_fsblk_t n_blocks_count_retry
= 0;
1932 unsigned long last_update_time
= 0;
1933 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1936 /* See if the device is actually as big as what was requested */
1937 bh
= ext4_sb_bread(sb
, n_blocks_count
- 1, 0);
1939 ext4_warning(sb
, "can't read last block, resize aborted");
1945 o_blocks_count
= ext4_blocks_count(es
);
1947 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1948 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1950 if (n_blocks_count
< o_blocks_count
) {
1951 /* On-line shrinking not supported */
1952 ext4_warning(sb
, "can't shrink FS - resize aborted");
1956 if (n_blocks_count
== o_blocks_count
)
1957 /* Nothing need to do */
1960 n_group
= ext4_get_group_number(sb
, n_blocks_count
- 1);
1961 if (n_group
>= (0xFFFFFFFFUL
/ EXT4_INODES_PER_GROUP(sb
))) {
1962 ext4_warning(sb
, "resize would cause inodes_count overflow");
1965 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1967 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1968 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1970 meta_bg
= ext4_has_feature_meta_bg(sb
);
1972 if (ext4_has_feature_resize_inode(sb
)) {
1974 ext4_error(sb
, "resize_inode and meta_bg enabled "
1978 if (n_desc_blocks
> o_desc_blocks
+
1979 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1980 n_blocks_count_retry
= n_blocks_count
;
1981 n_desc_blocks
= o_desc_blocks
+
1982 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1983 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1984 n_blocks_count
= (ext4_fsblk_t
)n_group
*
1985 EXT4_BLOCKS_PER_GROUP(sb
) +
1986 le32_to_cpu(es
->s_first_data_block
);
1987 n_group
--; /* set to last group number */
1991 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
,
1993 if (IS_ERR(resize_inode
)) {
1994 ext4_warning(sb
, "Error opening resize inode");
1995 return PTR_ERR(resize_inode
);
1999 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
2000 err
= ext4_convert_meta_bg(sb
, resize_inode
);
2005 resize_inode
= NULL
;
2007 if (n_blocks_count_retry
) {
2008 n_blocks_count
= n_blocks_count_retry
;
2009 n_blocks_count_retry
= 0;
2015 * Make sure the last group has enough space so that it's
2016 * guaranteed to have enough space for all metadata blocks
2017 * that it might need to hold. (We might not need to store
2018 * the inode table blocks in the last block group, but there
2019 * will be cases where this might be needed.)
2021 if ((ext4_group_first_block_no(sb
, n_group
) +
2022 ext4_group_overhead_blocks(sb
, n_group
) + 2 +
2023 sbi
->s_itb_per_group
+ sbi
->s_cluster_ratio
) >= n_blocks_count
) {
2024 n_blocks_count
= ext4_group_first_block_no(sb
, n_group
);
2026 n_blocks_count_retry
= 0;
2029 resize_inode
= NULL
;
2034 /* extend the last group */
2035 if (n_group
== o_group
)
2036 add
= n_blocks_count
- o_blocks_count
;
2038 add
= EXT4_C2B(sbi
, EXT4_CLUSTERS_PER_GROUP(sb
) - (offset
+ 1));
2040 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
2045 if (ext4_blocks_count(es
) == n_blocks_count
)
2048 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
2052 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
2056 flex_gd
= alloc_flex_gd(flexbg_size
);
2057 if (flex_gd
== NULL
) {
2062 /* Add flex groups. Note that a regular group is a
2063 * flex group with 1 group.
2065 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
2067 if (jiffies
- last_update_time
> HZ
* 10) {
2068 if (last_update_time
)
2069 ext4_msg(sb
, KERN_INFO
,
2070 "resized to %llu blocks",
2071 ext4_blocks_count(es
));
2072 last_update_time
= jiffies
;
2074 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
2076 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
2081 if (!err
&& n_blocks_count_retry
) {
2082 n_blocks_count
= n_blocks_count_retry
;
2083 n_blocks_count_retry
= 0;
2084 free_flex_gd(flex_gd
);
2088 resize_inode
= NULL
;
2095 free_flex_gd(flex_gd
);
2096 if (resize_inode
!= NULL
)
2099 ext4_warning(sb
, "error (%d) occurred during "
2100 "file system resize", err
);
2101 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu",
2102 ext4_blocks_count(es
));