2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
14 #include <linux/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 int ext4_resize_begin(struct super_block
*sb
)
21 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
24 if (!capable(CAP_SYS_RESOURCE
))
28 * If we are not using the primary superblock/GDT copy don't resize,
29 * because the user tools have no way of handling this. Probably a
30 * bad time to do it anyways.
32 if (EXT4_B2C(sbi
, sbi
->s_sbh
->b_blocknr
) !=
33 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
34 ext4_warning(sb
, "won't resize using backup superblock at %llu",
35 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
40 * We are not allowed to do online-resizing on a filesystem mounted
41 * with error, because it can destroy the filesystem easily.
43 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
44 ext4_warning(sb
, "There are errors in the filesystem, "
45 "so online resizing is not allowed");
49 if (test_and_set_bit_lock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
))
55 void ext4_resize_end(struct super_block
*sb
)
57 clear_bit_unlock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
);
58 smp_mb__after_atomic();
61 static ext4_group_t
ext4_meta_bg_first_group(struct super_block
*sb
,
63 return (group
>> EXT4_DESC_PER_BLOCK_BITS(sb
)) <<
64 EXT4_DESC_PER_BLOCK_BITS(sb
);
67 static ext4_fsblk_t
ext4_meta_bg_first_block_no(struct super_block
*sb
,
69 group
= ext4_meta_bg_first_group(sb
, group
);
70 return ext4_group_first_block_no(sb
, group
);
73 static ext4_grpblk_t
ext4_group_overhead_blocks(struct super_block
*sb
,
75 ext4_grpblk_t overhead
;
76 overhead
= ext4_bg_num_gdb(sb
, group
);
77 if (ext4_bg_has_super(sb
, group
))
79 le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
83 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
84 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
86 static int verify_group_input(struct super_block
*sb
,
87 struct ext4_new_group_data
*input
)
89 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
90 struct ext4_super_block
*es
= sbi
->s_es
;
91 ext4_fsblk_t start
= ext4_blocks_count(es
);
92 ext4_fsblk_t end
= start
+ input
->blocks_count
;
93 ext4_group_t group
= input
->group
;
94 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
97 struct buffer_head
*bh
= NULL
;
98 ext4_grpblk_t free_blocks_count
, offset
;
101 if (group
!= sbi
->s_groups_count
) {
102 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
103 input
->group
, sbi
->s_groups_count
);
107 overhead
= ext4_group_overhead_blocks(sb
, group
);
108 metaend
= start
+ overhead
;
109 input
->free_blocks_count
= free_blocks_count
=
110 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
112 if (test_opt(sb
, DEBUG
))
113 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
114 "(%d free, %u reserved)\n",
115 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
116 "no-super", input
->group
, input
->blocks_count
,
117 free_blocks_count
, input
->reserved_blocks
);
119 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
121 ext4_warning(sb
, "Last group not full");
122 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
123 ext4_warning(sb
, "Reserved blocks too high (%u)",
124 input
->reserved_blocks
);
125 else if (free_blocks_count
< 0)
126 ext4_warning(sb
, "Bad blocks count %u",
127 input
->blocks_count
);
128 else if (!(bh
= sb_bread(sb
, end
- 1)))
129 ext4_warning(sb
, "Cannot read last block (%llu)",
131 else if (outside(input
->block_bitmap
, start
, end
))
132 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
133 (unsigned long long)input
->block_bitmap
);
134 else if (outside(input
->inode_bitmap
, start
, end
))
135 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
136 (unsigned long long)input
->inode_bitmap
);
137 else if (outside(input
->inode_table
, start
, end
) ||
138 outside(itend
- 1, start
, end
))
139 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
140 (unsigned long long)input
->inode_table
, itend
- 1);
141 else if (input
->inode_bitmap
== input
->block_bitmap
)
142 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
143 (unsigned long long)input
->block_bitmap
);
144 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
145 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
147 (unsigned long long)input
->block_bitmap
,
148 (unsigned long long)input
->inode_table
, itend
- 1);
149 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
150 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
152 (unsigned long long)input
->inode_bitmap
,
153 (unsigned long long)input
->inode_table
, itend
- 1);
154 else if (inside(input
->block_bitmap
, start
, metaend
))
155 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
156 (unsigned long long)input
->block_bitmap
,
158 else if (inside(input
->inode_bitmap
, start
, metaend
))
159 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
160 (unsigned long long)input
->inode_bitmap
,
162 else if (inside(input
->inode_table
, start
, metaend
) ||
163 inside(itend
- 1, start
, metaend
))
164 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
166 (unsigned long long)input
->inode_table
,
167 itend
- 1, start
, metaend
- 1);
176 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
179 struct ext4_new_flex_group_data
{
180 struct ext4_new_group_data
*groups
; /* new_group_data for groups
182 __u16
*bg_flags
; /* block group flags of groups
184 ext4_group_t count
; /* number of groups in @groups
189 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
192 * Returns NULL on failure otherwise address of the allocated structure.
194 static struct ext4_new_flex_group_data
*alloc_flex_gd(unsigned long flexbg_size
)
196 struct ext4_new_flex_group_data
*flex_gd
;
198 flex_gd
= kmalloc(sizeof(*flex_gd
), GFP_NOFS
);
202 if (flexbg_size
>= UINT_MAX
/ sizeof(struct ext4_new_group_data
))
204 flex_gd
->count
= flexbg_size
;
206 flex_gd
->groups
= kmalloc(sizeof(struct ext4_new_group_data
) *
207 flexbg_size
, GFP_NOFS
);
208 if (flex_gd
->groups
== NULL
)
211 flex_gd
->bg_flags
= kmalloc(flexbg_size
* sizeof(__u16
), GFP_NOFS
);
212 if (flex_gd
->bg_flags
== NULL
)
218 kfree(flex_gd
->groups
);
225 static void free_flex_gd(struct ext4_new_flex_group_data
*flex_gd
)
227 kfree(flex_gd
->bg_flags
);
228 kfree(flex_gd
->groups
);
233 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
234 * and inode tables for a flex group.
236 * This function is used by 64bit-resize. Note that this function allocates
237 * group tables from the 1st group of groups contained by @flexgd, which may
238 * be a partial of a flex group.
240 * @sb: super block of fs to which the groups belongs
242 * Returns 0 on a successful allocation of the metadata blocks in the
245 static int ext4_alloc_group_tables(struct super_block
*sb
,
246 struct ext4_new_flex_group_data
*flex_gd
,
249 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
250 ext4_fsblk_t start_blk
;
251 ext4_fsblk_t last_blk
;
252 ext4_group_t src_group
;
253 ext4_group_t bb_index
= 0;
254 ext4_group_t ib_index
= 0;
255 ext4_group_t it_index
= 0;
257 ext4_group_t last_group
;
259 __u16 uninit_mask
= (flexbg_size
> 1) ? ~EXT4_BG_BLOCK_UNINIT
: ~0;
261 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
263 src_group
= group_data
[0].group
;
264 last_group
= src_group
+ flex_gd
->count
- 1;
266 BUG_ON((flexbg_size
> 1) && ((src_group
& ~(flexbg_size
- 1)) !=
267 (last_group
& ~(flexbg_size
- 1))));
269 group
= group_data
[0].group
;
270 if (src_group
>= group_data
[0].group
+ flex_gd
->count
)
272 start_blk
= ext4_group_first_block_no(sb
, src_group
);
273 last_blk
= start_blk
+ group_data
[src_group
- group
].blocks_count
;
275 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
277 start_blk
+= overhead
;
279 /* We collect contiguous blocks as much as possible. */
281 for (; src_group
<= last_group
; src_group
++) {
282 overhead
= ext4_group_overhead_blocks(sb
, src_group
);
284 last_blk
+= group_data
[src_group
- group
].blocks_count
;
289 /* Allocate block bitmaps */
290 for (; bb_index
< flex_gd
->count
; bb_index
++) {
291 if (start_blk
>= last_blk
)
293 group_data
[bb_index
].block_bitmap
= start_blk
++;
294 group
= ext4_get_group_number(sb
, start_blk
- 1);
295 group
-= group_data
[0].group
;
296 group_data
[group
].free_blocks_count
--;
297 flex_gd
->bg_flags
[group
] &= uninit_mask
;
300 /* Allocate inode bitmaps */
301 for (; ib_index
< flex_gd
->count
; ib_index
++) {
302 if (start_blk
>= last_blk
)
304 group_data
[ib_index
].inode_bitmap
= start_blk
++;
305 group
= ext4_get_group_number(sb
, start_blk
- 1);
306 group
-= group_data
[0].group
;
307 group_data
[group
].free_blocks_count
--;
308 flex_gd
->bg_flags
[group
] &= uninit_mask
;
311 /* Allocate inode tables */
312 for (; it_index
< flex_gd
->count
; it_index
++) {
313 unsigned int itb
= EXT4_SB(sb
)->s_itb_per_group
;
314 ext4_fsblk_t next_group_start
;
316 if (start_blk
+ itb
> last_blk
)
318 group_data
[it_index
].inode_table
= start_blk
;
319 group
= ext4_get_group_number(sb
, start_blk
);
320 next_group_start
= ext4_group_first_block_no(sb
, group
+ 1);
321 group
-= group_data
[0].group
;
323 if (start_blk
+ itb
> next_group_start
) {
324 flex_gd
->bg_flags
[group
+ 1] &= uninit_mask
;
325 overhead
= start_blk
+ itb
- next_group_start
;
326 group_data
[group
+ 1].free_blocks_count
-= overhead
;
330 group_data
[group
].free_blocks_count
-= itb
;
331 flex_gd
->bg_flags
[group
] &= uninit_mask
;
332 start_blk
+= EXT4_SB(sb
)->s_itb_per_group
;
335 if (test_opt(sb
, DEBUG
)) {
337 group
= group_data
[0].group
;
339 printk(KERN_DEBUG
"EXT4-fs: adding a flex group with "
340 "%d groups, flexbg size is %d:\n", flex_gd
->count
,
343 for (i
= 0; i
< flex_gd
->count
; i
++) {
344 printk(KERN_DEBUG
"adding %s group %u: %u "
345 "blocks (%d free)\n",
346 ext4_bg_has_super(sb
, group
+ i
) ? "normal" :
347 "no-super", group
+ i
,
348 group_data
[i
].blocks_count
,
349 group_data
[i
].free_blocks_count
);
355 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
358 struct buffer_head
*bh
;
361 bh
= sb_getblk(sb
, blk
);
363 return ERR_PTR(-ENOMEM
);
364 BUFFER_TRACE(bh
, "get_write_access");
365 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
369 memset(bh
->b_data
, 0, sb
->s_blocksize
);
370 set_buffer_uptodate(bh
);
377 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
378 * If that fails, restart the transaction & regain write access for the
379 * buffer head which is used for block_bitmap modifications.
381 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
385 if (ext4_handle_has_enough_credits(handle
, thresh
))
388 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
392 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
401 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
403 * Helper function for ext4_setup_new_group_blocks() which set .
406 * @handle: journal handle
407 * @flex_gd: flex group data
409 static int set_flexbg_block_bitmap(struct super_block
*sb
, handle_t
*handle
,
410 struct ext4_new_flex_group_data
*flex_gd
,
411 ext4_fsblk_t block
, ext4_group_t count
)
415 ext4_debug("mark blocks [%llu/%u] used\n", block
, count
);
416 for (count2
= count
; count
> 0; count
-= count2
, block
+= count2
) {
418 struct buffer_head
*bh
;
422 group
= ext4_get_group_number(sb
, block
);
423 start
= ext4_group_first_block_no(sb
, group
);
424 group
-= flex_gd
->groups
[0].group
;
426 count2
= EXT4_BLOCKS_PER_GROUP(sb
) - (block
- start
);
430 if (flex_gd
->bg_flags
[group
] & EXT4_BG_BLOCK_UNINIT
) {
431 BUG_ON(flex_gd
->count
> 1);
435 err
= extend_or_restart_transaction(handle
, 1);
439 bh
= sb_getblk(sb
, flex_gd
->groups
[group
].block_bitmap
);
443 BUFFER_TRACE(bh
, "get_write_access");
444 err
= ext4_journal_get_write_access(handle
, bh
);
449 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block
,
450 block
- start
, count2
);
451 ext4_set_bits(bh
->b_data
, block
- start
, count2
);
453 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
463 * Set up the block and inode bitmaps, and the inode table for the new groups.
464 * This doesn't need to be part of the main transaction, since we are only
465 * changing blocks outside the actual filesystem. We still do journaling to
466 * ensure the recovery is correct in case of a failure just after resize.
467 * If any part of this fails, we simply abort the resize.
469 * setup_new_flex_group_blocks handles a flex group as follow:
470 * 1. copy super block and GDT, and initialize group tables if necessary.
471 * In this step, we only set bits in blocks bitmaps for blocks taken by
472 * super block and GDT.
473 * 2. allocate group tables in block bitmaps, that is, set bits in block
474 * bitmap for blocks taken by group tables.
476 static int setup_new_flex_group_blocks(struct super_block
*sb
,
477 struct ext4_new_flex_group_data
*flex_gd
)
479 int group_table_count
[] = {1, 1, EXT4_SB(sb
)->s_itb_per_group
};
482 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
483 struct ext4_super_block
*es
= sbi
->s_es
;
484 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
485 __u16
*bg_flags
= flex_gd
->bg_flags
;
487 ext4_group_t group
, count
;
488 struct buffer_head
*bh
= NULL
;
489 int reserved_gdb
, i
, j
, err
= 0, err2
;
492 BUG_ON(!flex_gd
->count
|| !group_data
||
493 group_data
[0].group
!= sbi
->s_groups_count
);
495 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
496 meta_bg
= ext4_has_feature_meta_bg(sb
);
498 /* This transaction may be extended/restarted along the way */
499 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
501 return PTR_ERR(handle
);
503 group
= group_data
[0].group
;
504 for (i
= 0; i
< flex_gd
->count
; i
++, group
++) {
505 unsigned long gdblocks
;
506 ext4_grpblk_t overhead
;
508 gdblocks
= ext4_bg_num_gdb(sb
, group
);
509 start
= ext4_group_first_block_no(sb
, group
);
511 if (meta_bg
== 0 && !ext4_bg_has_super(sb
, group
))
515 ext4_group_t first_group
;
516 first_group
= ext4_meta_bg_first_group(sb
, group
);
517 if (first_group
!= group
+ 1 &&
518 first_group
!= group
+ EXT4_DESC_PER_BLOCK(sb
) - 1)
522 block
= start
+ ext4_bg_has_super(sb
, group
);
523 /* Copy all of the GDT blocks into the backup in this group */
524 for (j
= 0; j
< gdblocks
; j
++, block
++) {
525 struct buffer_head
*gdb
;
527 ext4_debug("update backup group %#04llx\n", block
);
528 err
= extend_or_restart_transaction(handle
, 1);
532 gdb
= sb_getblk(sb
, block
);
533 if (unlikely(!gdb
)) {
538 BUFFER_TRACE(gdb
, "get_write_access");
539 err
= ext4_journal_get_write_access(handle
, gdb
);
544 memcpy(gdb
->b_data
, sbi
->s_group_desc
[j
]->b_data
,
546 set_buffer_uptodate(gdb
);
548 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
556 /* Zero out all of the reserved backup group descriptor
559 if (ext4_bg_has_super(sb
, group
)) {
560 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1,
561 reserved_gdb
, GFP_NOFS
);
567 /* Initialize group tables of the grop @group */
568 if (!(bg_flags
[i
] & EXT4_BG_INODE_ZEROED
))
571 /* Zero out all of the inode table blocks */
572 block
= group_data
[i
].inode_table
;
573 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
574 block
, sbi
->s_itb_per_group
);
575 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
,
581 if (bg_flags
[i
] & EXT4_BG_BLOCK_UNINIT
)
584 /* Initialize block bitmap of the @group */
585 block
= group_data
[i
].block_bitmap
;
586 err
= extend_or_restart_transaction(handle
, 1);
590 bh
= bclean(handle
, sb
, block
);
595 overhead
= ext4_group_overhead_blocks(sb
, group
);
597 ext4_debug("mark backup superblock %#04llx (+0)\n",
599 ext4_set_bits(bh
->b_data
, 0, overhead
);
601 ext4_mark_bitmap_end(group_data
[i
].blocks_count
,
602 sb
->s_blocksize
* 8, bh
->b_data
);
603 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
609 if (bg_flags
[i
] & EXT4_BG_INODE_UNINIT
)
612 /* Initialize inode bitmap of the @group */
613 block
= group_data
[i
].inode_bitmap
;
614 err
= extend_or_restart_transaction(handle
, 1);
617 /* Mark unused entries in inode bitmap used */
618 bh
= bclean(handle
, sb
, block
);
624 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
),
625 sb
->s_blocksize
* 8, bh
->b_data
);
626 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
632 /* Mark group tables in block bitmap */
633 for (j
= 0; j
< GROUP_TABLE_COUNT
; j
++) {
634 count
= group_table_count
[j
];
635 start
= (&group_data
[0].block_bitmap
)[j
];
637 for (i
= 1; i
< flex_gd
->count
; i
++) {
638 block
+= group_table_count
[j
];
639 if (block
== (&group_data
[i
].block_bitmap
)[j
]) {
640 count
+= group_table_count
[j
];
643 err
= set_flexbg_block_bitmap(sb
, handle
,
644 flex_gd
, start
, count
);
647 count
= group_table_count
[j
];
648 start
= (&group_data
[i
].block_bitmap
)[j
];
653 err
= set_flexbg_block_bitmap(sb
, handle
,
654 flex_gd
, start
, count
);
661 err2
= ext4_journal_stop(handle
);
669 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
670 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
671 * calling this for the first time. In a sparse filesystem it will be the
672 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
673 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
675 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
676 unsigned *five
, unsigned *seven
)
678 unsigned *min
= three
;
682 if (!ext4_has_feature_sparse_super(sb
)) {
704 * Check that all of the backup GDT blocks are held in the primary GDT block.
705 * It is assumed that they are stored in group order. Returns the number of
706 * groups in current filesystem that have BACKUPS, or -ve error code.
708 static int verify_reserved_gdb(struct super_block
*sb
,
710 struct buffer_head
*primary
)
712 const ext4_fsblk_t blk
= primary
->b_blocknr
;
717 __le32
*p
= (__le32
*)primary
->b_data
;
720 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
721 if (le32_to_cpu(*p
++) !=
722 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
723 ext4_warning(sb
, "reserved GDT %llu"
724 " missing grp %d (%llu)",
727 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
731 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
739 * Called when we need to bring a reserved group descriptor table block into
740 * use from the resize inode. The primary copy of the new GDT block currently
741 * is an indirect block (under the double indirect block in the resize inode).
742 * The new backup GDT blocks will be stored as leaf blocks in this indirect
743 * block, in group order. Even though we know all the block numbers we need,
744 * we check to ensure that the resize inode has actually reserved these blocks.
746 * Don't need to update the block bitmaps because the blocks are still in use.
748 * We get all of the error cases out of the way, so that we are sure to not
749 * fail once we start modifying the data on disk, because JBD has no rollback.
751 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
754 struct super_block
*sb
= inode
->i_sb
;
755 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
756 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
757 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
758 struct buffer_head
**o_group_desc
, **n_group_desc
;
759 struct buffer_head
*dind
;
760 struct buffer_head
*gdb_bh
;
762 struct ext4_iloc iloc
;
766 if (test_opt(sb
, DEBUG
))
768 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
771 gdb_bh
= sb_bread(sb
, gdblock
);
775 gdbackups
= verify_reserved_gdb(sb
, group
, gdb_bh
);
781 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
782 dind
= sb_bread(sb
, le32_to_cpu(*data
));
788 data
= (__le32
*)dind
->b_data
;
789 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
790 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
796 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
797 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
801 BUFFER_TRACE(gdb_bh
, "get_write_access");
802 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
806 BUFFER_TRACE(dind
, "get_write_access");
807 err
= ext4_journal_get_write_access(handle
, dind
);
809 ext4_std_error(sb
, err
);
811 /* ext4_reserve_inode_write() gets a reference on the iloc */
812 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
816 n_group_desc
= ext4_kvmalloc((gdb_num
+ 1) *
817 sizeof(struct buffer_head
*),
821 ext4_warning(sb
, "not enough memory for %lu groups",
827 * Finally, we have all of the possible failures behind us...
829 * Remove new GDT block from inode double-indirect block and clear out
830 * the new GDT block for use (which also "frees" the backup GDT blocks
831 * from the reserved inode). We don't need to change the bitmaps for
832 * these blocks, because they are marked as in-use from being in the
833 * reserved inode, and will become GDT blocks (primary and backup).
835 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
836 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
838 ext4_std_error(sb
, err
);
841 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
842 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
843 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
844 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
846 ext4_std_error(sb
, err
);
852 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
853 memcpy(n_group_desc
, o_group_desc
,
854 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
855 n_group_desc
[gdb_num
] = gdb_bh
;
856 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
857 EXT4_SB(sb
)->s_gdb_count
++;
858 kvfree(o_group_desc
);
860 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
861 err
= ext4_handle_dirty_super(handle
, sb
);
863 ext4_std_error(sb
, err
);
868 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
= sb_bread(sb
, gdblock
);
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
= sb_bread(sb
, le32_to_cpu(*data
));
963 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
964 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
965 EXT4_ADDR_PER_BLOCK(sb
));
966 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
968 /* Get each reserved primary GDT block and verify it holds backups */
969 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
970 if (le32_to_cpu(*data
) != blk
) {
971 ext4_warning(sb
, "reserved block %llu"
972 " not at offset %ld",
974 (long)(data
- (__le32
*)dind
->b_data
));
978 primary
[res
] = sb_bread(sb
, blk
);
983 gdbackups
= verify_reserved_gdb(sb
, group
, primary
[res
]);
985 brelse(primary
[res
]);
990 data
= (__le32
*)dind
->b_data
;
993 for (i
= 0; i
< reserved_gdb
; i
++) {
994 BUFFER_TRACE(primary
[i
], "get_write_access");
995 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
])))
999 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
1003 * Finally we can add each of the reserved backup GDT blocks from
1004 * the new group to its reserved primary GDT block.
1006 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
1007 for (i
= 0; i
< reserved_gdb
; i
++) {
1009 data
= (__le32
*)primary
[i
]->b_data
;
1010 /* printk("reserving backup %lu[%u] = %lu\n",
1011 primary[i]->b_blocknr, gdbackups,
1012 blk + primary[i]->b_blocknr); */
1013 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
1014 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
1018 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
1019 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
1023 brelse(primary
[res
]);
1033 * Update the backup copies of the ext4 metadata. These don't need to be part
1034 * of the main resize transaction, because e2fsck will re-write them if there
1035 * is a problem (basically only OOM will cause a problem). However, we
1036 * _should_ update the backups if possible, in case the primary gets trashed
1037 * for some reason and we need to run e2fsck from a backup superblock. The
1038 * important part is that the new block and inode counts are in the backup
1039 * superblocks, and the location of the new group metadata in the GDT backups.
1041 * We do not need take the s_resize_lock for this, because these
1042 * blocks are not otherwise touched by the filesystem code when it is
1043 * mounted. We don't need to worry about last changing from
1044 * sbi->s_groups_count, because the worst that can happen is that we
1045 * do not copy the full number of backups at this time. The resize
1046 * which changed s_groups_count will backup again.
1048 static void update_backups(struct super_block
*sb
, sector_t blk_off
, char *data
,
1049 int size
, int meta_bg
)
1051 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1053 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
1057 ext4_group_t group
= 0;
1058 int rest
= sb
->s_blocksize
- size
;
1062 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, EXT4_MAX_TRANS_DATA
);
1063 if (IS_ERR(handle
)) {
1065 err
= PTR_ERR(handle
);
1070 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1071 last
= sbi
->s_groups_count
;
1073 group
= ext4_get_group_number(sb
, blk_off
) + 1;
1074 last
= (ext4_group_t
)(group
+ EXT4_DESC_PER_BLOCK(sb
) - 2);
1077 while (group
< sbi
->s_groups_count
) {
1078 struct buffer_head
*bh
;
1079 ext4_fsblk_t backup_block
;
1081 /* Out of journal space, and can't get more - abort - so sad */
1082 if (ext4_handle_valid(handle
) &&
1083 handle
->h_buffer_credits
== 0 &&
1084 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
1085 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
1089 backup_block
= ((ext4_fsblk_t
)group
) * bpg
+ blk_off
;
1091 backup_block
= (ext4_group_first_block_no(sb
, group
) +
1092 ext4_bg_has_super(sb
, group
));
1094 bh
= sb_getblk(sb
, backup_block
);
1095 if (unlikely(!bh
)) {
1099 ext4_debug("update metadata backup %llu(+%llu)\n",
1100 backup_block
, backup_block
-
1101 ext4_group_first_block_no(sb
, group
));
1102 BUFFER_TRACE(bh
, "get_write_access");
1103 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
1108 memcpy(bh
->b_data
, data
, size
);
1110 memset(bh
->b_data
+ size
, 0, rest
);
1111 set_buffer_uptodate(bh
);
1113 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1115 ext4_std_error(sb
, err
);
1119 group
= ext4_list_backups(sb
, &three
, &five
, &seven
);
1120 else if (group
== last
)
1125 if ((err2
= ext4_journal_stop(handle
)) && !err
)
1129 * Ugh! Need to have e2fsck write the backup copies. It is too
1130 * late to revert the resize, we shouldn't fail just because of
1131 * the backup copies (they are only needed in case of corruption).
1133 * However, if we got here we have a journal problem too, so we
1134 * can't really start a transaction to mark the superblock.
1135 * Chicken out and just set the flag on the hope it will be written
1136 * to disk, and if not - we will simply wait until next fsck.
1140 ext4_warning(sb
, "can't update backup for group %u (err %d), "
1141 "forcing fsck on next reboot", group
, err
);
1142 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
1143 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
1144 mark_buffer_dirty(sbi
->s_sbh
);
1149 * ext4_add_new_descs() adds @count group descriptor of groups
1150 * starting at @group
1152 * @handle: journal handle
1154 * @group: the group no. of the first group desc to be added
1155 * @resize_inode: the resize inode
1156 * @count: number of group descriptors to be added
1158 static int ext4_add_new_descs(handle_t
*handle
, struct super_block
*sb
,
1159 ext4_group_t group
, struct inode
*resize_inode
,
1162 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1163 struct ext4_super_block
*es
= sbi
->s_es
;
1164 struct buffer_head
*gdb_bh
;
1165 int i
, gdb_off
, gdb_num
, err
= 0;
1168 meta_bg
= ext4_has_feature_meta_bg(sb
);
1169 for (i
= 0; i
< count
; i
++, group
++) {
1170 int reserved_gdb
= ext4_bg_has_super(sb
, group
) ?
1171 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1173 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1174 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1177 * We will only either add reserved group blocks to a backup group
1178 * or remove reserved blocks for the first group in a new group block.
1179 * Doing both would be mean more complex code, and sane people don't
1180 * use non-sparse filesystems anymore. This is already checked above.
1183 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1184 BUFFER_TRACE(gdb_bh
, "get_write_access");
1185 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
1187 if (!err
&& reserved_gdb
&& ext4_bg_num_gdb(sb
, group
))
1188 err
= reserve_backup_gdb(handle
, resize_inode
, group
);
1189 } else if (meta_bg
!= 0) {
1190 err
= add_new_gdb_meta_bg(sb
, handle
, group
);
1192 err
= add_new_gdb(handle
, resize_inode
, group
);
1200 static struct buffer_head
*ext4_get_bitmap(struct super_block
*sb
, __u64 block
)
1202 struct buffer_head
*bh
= sb_getblk(sb
, block
);
1205 if (!bh_uptodate_or_lock(bh
)) {
1206 if (bh_submit_read(bh
) < 0) {
1215 static int ext4_set_bitmap_checksums(struct super_block
*sb
,
1217 struct ext4_group_desc
*gdp
,
1218 struct ext4_new_group_data
*group_data
)
1220 struct buffer_head
*bh
;
1222 if (!ext4_has_metadata_csum(sb
))
1225 bh
= ext4_get_bitmap(sb
, group_data
->inode_bitmap
);
1228 ext4_inode_bitmap_csum_set(sb
, group
, gdp
, bh
,
1229 EXT4_INODES_PER_GROUP(sb
) / 8);
1232 bh
= ext4_get_bitmap(sb
, group_data
->block_bitmap
);
1235 ext4_block_bitmap_csum_set(sb
, group
, gdp
, bh
);
1242 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1244 static int ext4_setup_new_descs(handle_t
*handle
, struct super_block
*sb
,
1245 struct ext4_new_flex_group_data
*flex_gd
)
1247 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1248 struct ext4_group_desc
*gdp
;
1249 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1250 struct buffer_head
*gdb_bh
;
1252 __u16
*bg_flags
= flex_gd
->bg_flags
;
1253 int i
, gdb_off
, gdb_num
, err
= 0;
1256 for (i
= 0; i
< flex_gd
->count
; i
++, group_data
++, bg_flags
++) {
1257 group
= group_data
->group
;
1259 gdb_off
= group
% EXT4_DESC_PER_BLOCK(sb
);
1260 gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1263 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1265 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1266 /* Update group descriptor block for new group */
1267 gdp
= (struct ext4_group_desc
*)(gdb_bh
->b_data
+
1268 gdb_off
* EXT4_DESC_SIZE(sb
));
1270 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
1271 ext4_block_bitmap_set(sb
, gdp
, group_data
->block_bitmap
);
1272 ext4_inode_bitmap_set(sb
, gdp
, group_data
->inode_bitmap
);
1273 err
= ext4_set_bitmap_checksums(sb
, group
, gdp
, group_data
);
1275 ext4_std_error(sb
, err
);
1279 ext4_inode_table_set(sb
, gdp
, group_data
->inode_table
);
1280 ext4_free_group_clusters_set(sb
, gdp
,
1281 EXT4_NUM_B2C(sbi
, group_data
->free_blocks_count
));
1282 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
1283 if (ext4_has_group_desc_csum(sb
))
1284 ext4_itable_unused_set(sb
, gdp
,
1285 EXT4_INODES_PER_GROUP(sb
));
1286 gdp
->bg_flags
= cpu_to_le16(*bg_flags
);
1287 ext4_group_desc_csum_set(sb
, group
, gdp
);
1289 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
1290 if (unlikely(err
)) {
1291 ext4_std_error(sb
, err
);
1296 * We can allocate memory for mb_alloc based on the new group
1299 err
= ext4_mb_add_groupinfo(sb
, group
, gdp
);
1307 * ext4_update_super() updates the super block so that the newly added
1308 * groups can be seen by the filesystem.
1311 * @flex_gd: new added groups
1313 static void ext4_update_super(struct super_block
*sb
,
1314 struct ext4_new_flex_group_data
*flex_gd
)
1316 ext4_fsblk_t blocks_count
= 0;
1317 ext4_fsblk_t free_blocks
= 0;
1318 ext4_fsblk_t reserved_blocks
= 0;
1319 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1320 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1321 struct ext4_super_block
*es
= sbi
->s_es
;
1324 BUG_ON(flex_gd
->count
== 0 || group_data
== NULL
);
1326 * Make the new blocks and inodes valid next. We do this before
1327 * increasing the group count so that once the group is enabled,
1328 * all of its blocks and inodes are already valid.
1330 * We always allocate group-by-group, then block-by-block or
1331 * inode-by-inode within a group, so enabling these
1332 * blocks/inodes before the group is live won't actually let us
1333 * allocate the new space yet.
1335 for (i
= 0; i
< flex_gd
->count
; i
++) {
1336 blocks_count
+= group_data
[i
].blocks_count
;
1337 free_blocks
+= group_data
[i
].free_blocks_count
;
1340 reserved_blocks
= ext4_r_blocks_count(es
) * 100;
1341 reserved_blocks
= div64_u64(reserved_blocks
, ext4_blocks_count(es
));
1342 reserved_blocks
*= blocks_count
;
1343 do_div(reserved_blocks
, 100);
1345 ext4_blocks_count_set(es
, ext4_blocks_count(es
) + blocks_count
);
1346 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + free_blocks
);
1347 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1349 le32_add_cpu(&es
->s_free_inodes_count
, EXT4_INODES_PER_GROUP(sb
) *
1352 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es
));
1354 * We need to protect s_groups_count against other CPUs seeing
1355 * inconsistent state in the superblock.
1357 * The precise rules we use are:
1359 * * Writers must perform a smp_wmb() after updating all
1360 * dependent data and before modifying the groups count
1362 * * Readers must perform an smp_rmb() after reading the groups
1363 * count and before reading any dependent data.
1365 * NB. These rules can be relaxed when checking the group count
1366 * while freeing data, as we can only allocate from a block
1367 * group after serialising against the group count, and we can
1368 * only then free after serialising in turn against that
1373 /* Update the global fs size fields */
1374 sbi
->s_groups_count
+= flex_gd
->count
;
1375 sbi
->s_blockfile_groups
= min_t(ext4_group_t
, sbi
->s_groups_count
,
1376 (EXT4_MAX_BLOCK_FILE_PHYS
/ EXT4_BLOCKS_PER_GROUP(sb
)));
1378 /* Update the reserved block counts only once the new group is
1380 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
1383 /* Update the free space counts */
1384 percpu_counter_add(&sbi
->s_freeclusters_counter
,
1385 EXT4_NUM_B2C(sbi
, free_blocks
));
1386 percpu_counter_add(&sbi
->s_freeinodes_counter
,
1387 EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
);
1389 ext4_debug("free blocks count %llu",
1390 percpu_counter_read(&sbi
->s_freeclusters_counter
));
1391 if (ext4_has_feature_flex_bg(sb
) && sbi
->s_log_groups_per_flex
) {
1392 ext4_group_t flex_group
;
1393 flex_group
= ext4_flex_group(sbi
, group_data
[0].group
);
1394 atomic64_add(EXT4_NUM_B2C(sbi
, free_blocks
),
1395 &sbi
->s_flex_groups
[flex_group
].free_clusters
);
1396 atomic_add(EXT4_INODES_PER_GROUP(sb
) * flex_gd
->count
,
1397 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
1401 * Update the fs overhead information
1403 ext4_calculate_overhead(sb
);
1405 if (test_opt(sb
, DEBUG
))
1406 printk(KERN_DEBUG
"EXT4-fs: added group %u:"
1407 "%llu blocks(%llu free %llu reserved)\n", flex_gd
->count
,
1408 blocks_count
, free_blocks
, reserved_blocks
);
1411 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1412 * _before_ we start modifying the filesystem, because we cannot abort the
1413 * transaction and not have it write the data to disk.
1415 static int ext4_flex_group_add(struct super_block
*sb
,
1416 struct inode
*resize_inode
,
1417 struct ext4_new_flex_group_data
*flex_gd
)
1419 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1420 struct ext4_super_block
*es
= sbi
->s_es
;
1421 ext4_fsblk_t o_blocks_count
;
1425 unsigned reserved_gdb
;
1426 int err
= 0, err2
= 0, credit
;
1428 BUG_ON(!flex_gd
->count
|| !flex_gd
->groups
|| !flex_gd
->bg_flags
);
1430 reserved_gdb
= le16_to_cpu(es
->s_reserved_gdt_blocks
);
1431 o_blocks_count
= ext4_blocks_count(es
);
1432 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1435 err
= setup_new_flex_group_blocks(sb
, flex_gd
);
1439 * We will always be modifying at least the superblock and GDT
1440 * blocks. If we are adding a group past the last current GDT block,
1441 * we will also modify the inode and the dindirect block. If we
1442 * are adding a group with superblock/GDT backups we will also
1443 * modify each of the reserved GDT dindirect blocks.
1445 credit
= 3; /* sb, resize inode, resize inode dindirect */
1447 credit
+= 1 + DIV_ROUND_UP(flex_gd
->count
, EXT4_DESC_PER_BLOCK(sb
));
1448 credit
+= reserved_gdb
; /* Reserved GDT dindirect blocks */
1449 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credit
);
1450 if (IS_ERR(handle
)) {
1451 err
= PTR_ERR(handle
);
1455 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1456 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1460 group
= flex_gd
->groups
[0].group
;
1461 BUG_ON(group
!= EXT4_SB(sb
)->s_groups_count
);
1462 err
= ext4_add_new_descs(handle
, sb
, group
,
1463 resize_inode
, flex_gd
->count
);
1467 err
= ext4_setup_new_descs(handle
, sb
, flex_gd
);
1471 ext4_update_super(sb
, flex_gd
);
1473 err
= ext4_handle_dirty_super(handle
, sb
);
1476 err2
= ext4_journal_stop(handle
);
1481 int gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
1482 int gdb_num_end
= ((group
+ flex_gd
->count
- 1) /
1483 EXT4_DESC_PER_BLOCK(sb
));
1484 int meta_bg
= ext4_has_feature_meta_bg(sb
);
1485 sector_t old_gdb
= 0;
1487 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
1488 sizeof(struct ext4_super_block
), 0);
1489 for (; gdb_num
<= gdb_num_end
; gdb_num
++) {
1490 struct buffer_head
*gdb_bh
;
1492 gdb_bh
= sbi
->s_group_desc
[gdb_num
];
1493 if (old_gdb
== gdb_bh
->b_blocknr
)
1495 update_backups(sb
, gdb_bh
->b_blocknr
, gdb_bh
->b_data
,
1496 gdb_bh
->b_size
, meta_bg
);
1497 old_gdb
= gdb_bh
->b_blocknr
;
1504 static int ext4_setup_next_flex_gd(struct super_block
*sb
,
1505 struct ext4_new_flex_group_data
*flex_gd
,
1506 ext4_fsblk_t n_blocks_count
,
1507 unsigned long flexbg_size
)
1509 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1510 struct ext4_new_group_data
*group_data
= flex_gd
->groups
;
1511 ext4_fsblk_t o_blocks_count
;
1512 ext4_group_t n_group
;
1514 ext4_group_t last_group
;
1516 ext4_grpblk_t blocks_per_group
;
1519 blocks_per_group
= EXT4_BLOCKS_PER_GROUP(sb
);
1521 o_blocks_count
= ext4_blocks_count(es
);
1523 if (o_blocks_count
== n_blocks_count
)
1526 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1528 ext4_get_group_no_and_offset(sb
, n_blocks_count
- 1, &n_group
, &last
);
1530 last_group
= group
| (flexbg_size
- 1);
1531 if (last_group
> n_group
)
1532 last_group
= n_group
;
1534 flex_gd
->count
= last_group
- group
+ 1;
1536 for (i
= 0; i
< flex_gd
->count
; i
++) {
1539 group_data
[i
].group
= group
+ i
;
1540 group_data
[i
].blocks_count
= blocks_per_group
;
1541 overhead
= ext4_group_overhead_blocks(sb
, group
+ i
);
1542 group_data
[i
].free_blocks_count
= blocks_per_group
- overhead
;
1543 if (ext4_has_group_desc_csum(sb
)) {
1544 flex_gd
->bg_flags
[i
] = EXT4_BG_BLOCK_UNINIT
|
1545 EXT4_BG_INODE_UNINIT
;
1546 if (!test_opt(sb
, INIT_INODE_TABLE
))
1547 flex_gd
->bg_flags
[i
] |= EXT4_BG_INODE_ZEROED
;
1549 flex_gd
->bg_flags
[i
] = EXT4_BG_INODE_ZEROED
;
1552 if (last_group
== n_group
&& ext4_has_group_desc_csum(sb
))
1553 /* We need to initialize block bitmap of last group. */
1554 flex_gd
->bg_flags
[i
- 1] &= ~EXT4_BG_BLOCK_UNINIT
;
1556 if ((last_group
== n_group
) && (last
!= blocks_per_group
- 1)) {
1557 group_data
[i
- 1].blocks_count
= last
+ 1;
1558 group_data
[i
- 1].free_blocks_count
-= blocks_per_group
-
1565 /* Add group descriptor data to an existing or new group descriptor block.
1566 * Ensure we handle all possible error conditions _before_ we start modifying
1567 * the filesystem, because we cannot abort the transaction and not have it
1568 * write the data to disk.
1570 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1571 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1573 * We only need to hold the superblock lock while we are actually adding
1574 * in the new group's counts to the superblock. Prior to that we have
1575 * not really "added" the group at all. We re-check that we are still
1576 * adding in the last group in case things have changed since verifying.
1578 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
1580 struct ext4_new_flex_group_data flex_gd
;
1581 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1582 struct ext4_super_block
*es
= sbi
->s_es
;
1583 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
1584 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
1585 struct inode
*inode
= NULL
;
1590 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
1592 if (gdb_off
== 0 && !ext4_has_feature_sparse_super(sb
)) {
1593 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
1597 if (ext4_blocks_count(es
) + input
->blocks_count
<
1598 ext4_blocks_count(es
)) {
1599 ext4_warning(sb
, "blocks_count overflow");
1603 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
1604 le32_to_cpu(es
->s_inodes_count
)) {
1605 ext4_warning(sb
, "inodes_count overflow");
1609 if (reserved_gdb
|| gdb_off
== 0) {
1610 if (!ext4_has_feature_resize_inode(sb
) ||
1611 !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1613 "No reserved GDT blocks, can't resize");
1616 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1617 if (IS_ERR(inode
)) {
1618 ext4_warning(sb
, "Error opening resize inode");
1619 return PTR_ERR(inode
);
1624 err
= verify_group_input(sb
, input
);
1628 err
= ext4_alloc_flex_bg_array(sb
, input
->group
+ 1);
1632 err
= ext4_mb_alloc_groupinfo(sb
, input
->group
+ 1);
1637 flex_gd
.groups
= input
;
1638 flex_gd
.bg_flags
= &bg_flags
;
1639 err
= ext4_flex_group_add(sb
, inode
, &flex_gd
);
1643 } /* ext4_group_add */
1646 * extend a group without checking assuming that checking has been done.
1648 static int ext4_group_extend_no_check(struct super_block
*sb
,
1649 ext4_fsblk_t o_blocks_count
, ext4_grpblk_t add
)
1651 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
1655 /* We will update the superblock, one block bitmap, and
1656 * one group descriptor via ext4_group_add_blocks().
1658 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, 3);
1659 if (IS_ERR(handle
)) {
1660 err
= PTR_ERR(handle
);
1661 ext4_warning(sb
, "error %d on journal start", err
);
1665 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get_write_access");
1666 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
1668 ext4_warning(sb
, "error %d on journal write access", err
);
1672 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1673 ext4_free_blocks_count_set(es
, ext4_free_blocks_count(es
) + add
);
1674 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1675 o_blocks_count
+ add
);
1676 /* We add the blocks to the bitmap and set the group need init bit */
1677 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1680 ext4_handle_dirty_super(handle
, sb
);
1681 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1682 o_blocks_count
+ add
);
1684 err2
= ext4_journal_stop(handle
);
1689 if (test_opt(sb
, DEBUG
))
1690 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu "
1691 "blocks\n", ext4_blocks_count(es
));
1692 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
,
1693 (char *)es
, sizeof(struct ext4_super_block
), 0);
1699 * Extend the filesystem to the new number of blocks specified. This entry
1700 * point is only used to extend the current filesystem to the end of the last
1701 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1702 * for emergencies (because it has no dependencies on reserved blocks).
1704 * If we _really_ wanted, we could use default values to call ext4_group_add()
1705 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1706 * GDT blocks are reserved to grow to the desired size.
1708 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
1709 ext4_fsblk_t n_blocks_count
)
1711 ext4_fsblk_t o_blocks_count
;
1714 struct buffer_head
*bh
;
1718 o_blocks_count
= ext4_blocks_count(es
);
1720 if (test_opt(sb
, DEBUG
))
1721 ext4_msg(sb
, KERN_DEBUG
,
1722 "extending last group from %llu to %llu blocks",
1723 o_blocks_count
, n_blocks_count
);
1725 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1728 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1729 ext4_msg(sb
, KERN_ERR
,
1730 "filesystem too large to resize to %llu blocks safely",
1732 if (sizeof(sector_t
) < 8)
1733 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1737 if (n_blocks_count
< o_blocks_count
) {
1738 ext4_warning(sb
, "can't shrink FS - resize aborted");
1742 /* Handle the remaining blocks in the last group only. */
1743 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1746 ext4_warning(sb
, "need to use ext2online to resize further");
1750 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1752 if (o_blocks_count
+ add
< o_blocks_count
) {
1753 ext4_warning(sb
, "blocks_count overflow");
1757 if (o_blocks_count
+ add
> n_blocks_count
)
1758 add
= n_blocks_count
- o_blocks_count
;
1760 if (o_blocks_count
+ add
< n_blocks_count
)
1761 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1762 o_blocks_count
+ add
, add
);
1764 /* See if the device is actually as big as what was requested */
1765 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1767 ext4_warning(sb
, "can't read last block, resize aborted");
1772 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1774 } /* ext4_group_extend */
1777 static int num_desc_blocks(struct super_block
*sb
, ext4_group_t groups
)
1779 return (groups
+ EXT4_DESC_PER_BLOCK(sb
) - 1) / EXT4_DESC_PER_BLOCK(sb
);
1783 * Release the resize inode and drop the resize_inode feature if there
1784 * are no more reserved gdt blocks, and then convert the file system
1787 static int ext4_convert_meta_bg(struct super_block
*sb
, struct inode
*inode
)
1790 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1791 struct ext4_super_block
*es
= sbi
->s_es
;
1792 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1794 int i
, ret
, err
= 0;
1797 ext4_msg(sb
, KERN_INFO
, "Converting file system to meta_bg");
1799 if (es
->s_reserved_gdt_blocks
) {
1800 ext4_error(sb
, "Unexpected non-zero "
1801 "s_reserved_gdt_blocks");
1805 /* Do a quick sanity check of the resize inode */
1806 if (inode
->i_blocks
!= 1 << (inode
->i_blkbits
- 9))
1807 goto invalid_resize_inode
;
1808 for (i
= 0; i
< EXT4_N_BLOCKS
; i
++) {
1809 if (i
== EXT4_DIND_BLOCK
) {
1813 goto invalid_resize_inode
;
1816 goto invalid_resize_inode
;
1818 credits
+= 3; /* block bitmap, bg descriptor, resize inode */
1821 handle
= ext4_journal_start_sb(sb
, EXT4_HT_RESIZE
, credits
);
1823 return PTR_ERR(handle
);
1825 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
1826 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
1830 ext4_clear_feature_resize_inode(sb
);
1831 ext4_set_feature_meta_bg(sb
);
1832 sbi
->s_es
->s_first_meta_bg
=
1833 cpu_to_le32(num_desc_blocks(sb
, sbi
->s_groups_count
));
1835 err
= ext4_handle_dirty_super(handle
, sb
);
1837 ext4_std_error(sb
, err
);
1842 nr
= le32_to_cpu(ei
->i_data
[EXT4_DIND_BLOCK
]);
1843 ext4_free_blocks(handle
, inode
, NULL
, nr
, 1,
1844 EXT4_FREE_BLOCKS_METADATA
|
1845 EXT4_FREE_BLOCKS_FORGET
);
1846 ei
->i_data
[EXT4_DIND_BLOCK
] = 0;
1847 inode
->i_blocks
= 0;
1849 err
= ext4_mark_inode_dirty(handle
, inode
);
1851 ext4_std_error(sb
, err
);
1855 ret
= ext4_journal_stop(handle
);
1860 invalid_resize_inode
:
1861 ext4_error(sb
, "corrupted/inconsistent resize inode");
1866 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1868 * @sb: super block of the fs to be resized
1869 * @n_blocks_count: the number of blocks resides in the resized fs
1871 int ext4_resize_fs(struct super_block
*sb
, ext4_fsblk_t n_blocks_count
)
1873 struct ext4_new_flex_group_data
*flex_gd
= NULL
;
1874 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
1875 struct ext4_super_block
*es
= sbi
->s_es
;
1876 struct buffer_head
*bh
;
1877 struct inode
*resize_inode
= NULL
;
1878 ext4_grpblk_t add
, offset
;
1879 unsigned long n_desc_blocks
;
1880 unsigned long o_desc_blocks
;
1881 ext4_group_t o_group
;
1882 ext4_group_t n_group
;
1883 ext4_fsblk_t o_blocks_count
;
1884 ext4_fsblk_t n_blocks_count_retry
= 0;
1885 unsigned long last_update_time
= 0;
1886 int err
= 0, flexbg_size
= 1 << sbi
->s_log_groups_per_flex
;
1889 /* See if the device is actually as big as what was requested */
1890 bh
= sb_bread(sb
, n_blocks_count
- 1);
1892 ext4_warning(sb
, "can't read last block, resize aborted");
1898 o_blocks_count
= ext4_blocks_count(es
);
1900 ext4_msg(sb
, KERN_INFO
, "resizing filesystem from %llu "
1901 "to %llu blocks", o_blocks_count
, n_blocks_count
);
1903 if (n_blocks_count
< o_blocks_count
) {
1904 /* On-line shrinking not supported */
1905 ext4_warning(sb
, "can't shrink FS - resize aborted");
1909 if (n_blocks_count
== o_blocks_count
)
1910 /* Nothing need to do */
1913 n_group
= ext4_get_group_number(sb
, n_blocks_count
- 1);
1914 if (n_group
>= (0xFFFFFFFFUL
/ EXT4_INODES_PER_GROUP(sb
))) {
1915 ext4_warning(sb
, "resize would cause inodes_count overflow");
1918 ext4_get_group_no_and_offset(sb
, o_blocks_count
- 1, &o_group
, &offset
);
1920 n_desc_blocks
= num_desc_blocks(sb
, n_group
+ 1);
1921 o_desc_blocks
= num_desc_blocks(sb
, sbi
->s_groups_count
);
1923 meta_bg
= ext4_has_feature_meta_bg(sb
);
1925 if (ext4_has_feature_resize_inode(sb
)) {
1927 ext4_error(sb
, "resize_inode and meta_bg enabled "
1931 if (n_desc_blocks
> o_desc_blocks
+
1932 le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
1933 n_blocks_count_retry
= n_blocks_count
;
1934 n_desc_blocks
= o_desc_blocks
+
1935 le16_to_cpu(es
->s_reserved_gdt_blocks
);
1936 n_group
= n_desc_blocks
* EXT4_DESC_PER_BLOCK(sb
);
1937 n_blocks_count
= (ext4_fsblk_t
)n_group
*
1938 EXT4_BLOCKS_PER_GROUP(sb
) +
1939 le32_to_cpu(es
->s_first_data_block
);
1940 n_group
--; /* set to last group number */
1944 resize_inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
1945 if (IS_ERR(resize_inode
)) {
1946 ext4_warning(sb
, "Error opening resize inode");
1947 return PTR_ERR(resize_inode
);
1951 if ((!resize_inode
&& !meta_bg
) || n_blocks_count
== o_blocks_count
) {
1952 err
= ext4_convert_meta_bg(sb
, resize_inode
);
1957 resize_inode
= NULL
;
1959 if (n_blocks_count_retry
) {
1960 n_blocks_count
= n_blocks_count_retry
;
1961 n_blocks_count_retry
= 0;
1967 * Make sure the last group has enough space so that it's
1968 * guaranteed to have enough space for all metadata blocks
1969 * that it might need to hold. (We might not need to store
1970 * the inode table blocks in the last block group, but there
1971 * will be cases where this might be needed.)
1973 if ((ext4_group_first_block_no(sb
, n_group
) +
1974 ext4_group_overhead_blocks(sb
, n_group
) + 2 +
1975 sbi
->s_itb_per_group
+ sbi
->s_cluster_ratio
) >= n_blocks_count
) {
1976 n_blocks_count
= ext4_group_first_block_no(sb
, n_group
);
1978 n_blocks_count_retry
= 0;
1981 resize_inode
= NULL
;
1986 /* extend the last group */
1987 if (n_group
== o_group
)
1988 add
= n_blocks_count
- o_blocks_count
;
1990 add
= EXT4_BLOCKS_PER_GROUP(sb
) - (offset
+ 1);
1992 err
= ext4_group_extend_no_check(sb
, o_blocks_count
, add
);
1997 if (ext4_blocks_count(es
) == n_blocks_count
)
2000 err
= ext4_alloc_flex_bg_array(sb
, n_group
+ 1);
2004 err
= ext4_mb_alloc_groupinfo(sb
, n_group
+ 1);
2008 flex_gd
= alloc_flex_gd(flexbg_size
);
2009 if (flex_gd
== NULL
) {
2014 /* Add flex groups. Note that a regular group is a
2015 * flex group with 1 group.
2017 while (ext4_setup_next_flex_gd(sb
, flex_gd
, n_blocks_count
,
2019 if (jiffies
- last_update_time
> HZ
* 10) {
2020 if (last_update_time
)
2021 ext4_msg(sb
, KERN_INFO
,
2022 "resized to %llu blocks",
2023 ext4_blocks_count(es
));
2024 last_update_time
= jiffies
;
2026 if (ext4_alloc_group_tables(sb
, flex_gd
, flexbg_size
) != 0)
2028 err
= ext4_flex_group_add(sb
, resize_inode
, flex_gd
);
2033 if (!err
&& n_blocks_count_retry
) {
2034 n_blocks_count
= n_blocks_count_retry
;
2035 n_blocks_count_retry
= 0;
2036 free_flex_gd(flex_gd
);
2040 resize_inode
= NULL
;
2047 free_flex_gd(flex_gd
);
2048 if (resize_inode
!= NULL
)
2050 ext4_msg(sb
, KERN_INFO
, "resized filesystem to %llu", n_blocks_count
);