2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
14 #include <linux/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 int ext4_resize_begin(struct super_block
*sb
)
23 if (!capable(CAP_SYS_RESOURCE
))
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
30 if (EXT4_SB(sb
)->s_mount_state
& EXT4_ERROR_FS
) {
31 ext4_warning(sb
, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
36 if (test_and_set_bit_lock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
))
42 void ext4_resize_end(struct super_block
*sb
)
44 clear_bit_unlock(EXT4_RESIZING
, &EXT4_SB(sb
)->s_resize_flags
);
45 smp_mb__after_clear_bit();
48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
49 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
51 static int verify_group_input(struct super_block
*sb
,
52 struct ext4_new_group_data
*input
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
55 struct ext4_super_block
*es
= sbi
->s_es
;
56 ext4_fsblk_t start
= ext4_blocks_count(es
);
57 ext4_fsblk_t end
= start
+ input
->blocks_count
;
58 ext4_group_t group
= input
->group
;
59 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
60 unsigned overhead
= ext4_bg_has_super(sb
, group
) ?
61 (1 + ext4_bg_num_gdb(sb
, group
) +
62 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
63 ext4_fsblk_t metaend
= start
+ overhead
;
64 struct buffer_head
*bh
= NULL
;
65 ext4_grpblk_t free_blocks_count
, offset
;
68 input
->free_blocks_count
= free_blocks_count
=
69 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
71 if (test_opt(sb
, DEBUG
))
72 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
73 "(%d free, %u reserved)\n",
74 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
75 "no-super", input
->group
, input
->blocks_count
,
76 free_blocks_count
, input
->reserved_blocks
);
78 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
79 if (group
!= sbi
->s_groups_count
)
80 ext4_warning(sb
, "Cannot add at group %u (only %u groups)",
81 input
->group
, sbi
->s_groups_count
);
83 ext4_warning(sb
, "Last group not full");
84 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
85 ext4_warning(sb
, "Reserved blocks too high (%u)",
86 input
->reserved_blocks
);
87 else if (free_blocks_count
< 0)
88 ext4_warning(sb
, "Bad blocks count %u",
90 else if (!(bh
= sb_bread(sb
, end
- 1)))
91 ext4_warning(sb
, "Cannot read last block (%llu)",
93 else if (outside(input
->block_bitmap
, start
, end
))
94 ext4_warning(sb
, "Block bitmap not in group (block %llu)",
95 (unsigned long long)input
->block_bitmap
);
96 else if (outside(input
->inode_bitmap
, start
, end
))
97 ext4_warning(sb
, "Inode bitmap not in group (block %llu)",
98 (unsigned long long)input
->inode_bitmap
);
99 else if (outside(input
->inode_table
, start
, end
) ||
100 outside(itend
- 1, start
, end
))
101 ext4_warning(sb
, "Inode table not in group (blocks %llu-%llu)",
102 (unsigned long long)input
->inode_table
, itend
- 1);
103 else if (input
->inode_bitmap
== input
->block_bitmap
)
104 ext4_warning(sb
, "Block bitmap same as inode bitmap (%llu)",
105 (unsigned long long)input
->block_bitmap
);
106 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
107 ext4_warning(sb
, "Block bitmap (%llu) in inode table "
109 (unsigned long long)input
->block_bitmap
,
110 (unsigned long long)input
->inode_table
, itend
- 1);
111 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
112 ext4_warning(sb
, "Inode bitmap (%llu) in inode table "
114 (unsigned long long)input
->inode_bitmap
,
115 (unsigned long long)input
->inode_table
, itend
- 1);
116 else if (inside(input
->block_bitmap
, start
, metaend
))
117 ext4_warning(sb
, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 (unsigned long long)input
->block_bitmap
,
120 else if (inside(input
->inode_bitmap
, start
, metaend
))
121 ext4_warning(sb
, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 (unsigned long long)input
->inode_bitmap
,
124 else if (inside(input
->inode_table
, start
, metaend
) ||
125 inside(itend
- 1, start
, metaend
))
126 ext4_warning(sb
, "Inode table (%llu-%llu) overlaps GDT table "
128 (unsigned long long)input
->inode_table
,
129 itend
- 1, start
, metaend
- 1);
137 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
140 struct buffer_head
*bh
;
143 bh
= sb_getblk(sb
, blk
);
145 return ERR_PTR(-EIO
);
146 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
150 memset(bh
->b_data
, 0, sb
->s_blocksize
);
151 set_buffer_uptodate(bh
);
158 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
159 * If that fails, restart the transaction & regain write access for the
160 * buffer head which is used for block_bitmap modifications.
162 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
)
166 if (ext4_handle_has_enough_credits(handle
, thresh
))
169 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
173 err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
);
182 * Set up the block and inode bitmaps, and the inode table for the new group.
183 * This doesn't need to be part of the main transaction, since we are only
184 * changing blocks outside the actual filesystem. We still do journaling to
185 * ensure the recovery is correct in case of a failure just after resize.
186 * If any part of this fails, we simply abort the resize.
188 static int setup_new_group_blocks(struct super_block
*sb
,
189 struct ext4_new_group_data
*input
)
191 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
192 ext4_fsblk_t start
= ext4_group_first_block_no(sb
, input
->group
);
193 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
194 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
) : 0;
195 unsigned long gdblocks
= ext4_bg_num_gdb(sb
, input
->group
);
196 struct buffer_head
*bh
;
203 /* This transaction may be extended/restarted along the way */
204 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
207 return PTR_ERR(handle
);
209 BUG_ON(input
->group
!= sbi
->s_groups_count
);
211 /* Copy all of the GDT blocks into the backup in this group */
212 for (i
= 0, bit
= 1, block
= start
+ 1;
213 i
< gdblocks
; i
++, block
++, bit
++) {
214 struct buffer_head
*gdb
;
216 ext4_debug("update backup group %#04llx (+%d)\n", block
, bit
);
217 err
= extend_or_restart_transaction(handle
, 1);
221 gdb
= sb_getblk(sb
, block
);
226 if ((err
= ext4_journal_get_write_access(handle
, gdb
))) {
230 memcpy(gdb
->b_data
, sbi
->s_group_desc
[i
]->b_data
, gdb
->b_size
);
231 set_buffer_uptodate(gdb
);
232 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
240 /* Zero out all of the reserved backup group descriptor table blocks */
241 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
242 block
, sbi
->s_itb_per_group
);
243 err
= sb_issue_zeroout(sb
, gdblocks
+ start
+ 1, reserved_gdb
,
248 err
= extend_or_restart_transaction(handle
, 2);
252 bh
= bclean(handle
, sb
, input
->block_bitmap
);
258 if (ext4_bg_has_super(sb
, input
->group
)) {
259 ext4_debug("mark backup group tables %#04llx (+0)\n", start
);
260 ext4_set_bits(bh
->b_data
, 0, gdblocks
+ reserved_gdb
+ 1);
263 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input
->block_bitmap
,
264 input
->block_bitmap
- start
);
265 ext4_set_bit(input
->block_bitmap
- start
, bh
->b_data
);
266 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input
->inode_bitmap
,
267 input
->inode_bitmap
- start
);
268 ext4_set_bit(input
->inode_bitmap
- start
, bh
->b_data
);
270 /* Zero out all of the inode table blocks */
271 block
= input
->inode_table
;
272 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
273 block
, sbi
->s_itb_per_group
);
274 err
= sb_issue_zeroout(sb
, block
, sbi
->s_itb_per_group
, GFP_NOFS
);
277 ext4_set_bits(bh
->b_data
, input
->inode_table
- start
,
278 sbi
->s_itb_per_group
);
281 ext4_mark_bitmap_end(input
->blocks_count
, sb
->s_blocksize
* 8,
283 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
285 ext4_std_error(sb
, err
);
289 /* Mark unused entries in inode bitmap used */
290 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
291 input
->inode_bitmap
, input
->inode_bitmap
- start
);
292 if (IS_ERR(bh
= bclean(handle
, sb
, input
->inode_bitmap
))) {
297 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
299 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
301 ext4_std_error(sb
, err
);
306 if ((err2
= ext4_journal_stop(handle
)) && !err
)
313 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
314 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
315 * calling this for the first time. In a sparse filesystem it will be the
316 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
317 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
319 static unsigned ext4_list_backups(struct super_block
*sb
, unsigned *three
,
320 unsigned *five
, unsigned *seven
)
322 unsigned *min
= three
;
326 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
327 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
349 * Check that all of the backup GDT blocks are held in the primary GDT block.
350 * It is assumed that they are stored in group order. Returns the number of
351 * groups in current filesystem that have BACKUPS, or -ve error code.
353 static int verify_reserved_gdb(struct super_block
*sb
,
354 struct buffer_head
*primary
)
356 const ext4_fsblk_t blk
= primary
->b_blocknr
;
357 const ext4_group_t end
= EXT4_SB(sb
)->s_groups_count
;
362 __le32
*p
= (__le32
*)primary
->b_data
;
365 while ((grp
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < end
) {
366 if (le32_to_cpu(*p
++) !=
367 grp
* EXT4_BLOCKS_PER_GROUP(sb
) + blk
){
368 ext4_warning(sb
, "reserved GDT %llu"
369 " missing grp %d (%llu)",
372 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
376 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
384 * Called when we need to bring a reserved group descriptor table block into
385 * use from the resize inode. The primary copy of the new GDT block currently
386 * is an indirect block (under the double indirect block in the resize inode).
387 * The new backup GDT blocks will be stored as leaf blocks in this indirect
388 * block, in group order. Even though we know all the block numbers we need,
389 * we check to ensure that the resize inode has actually reserved these blocks.
391 * Don't need to update the block bitmaps because the blocks are still in use.
393 * We get all of the error cases out of the way, so that we are sure to not
394 * fail once we start modifying the data on disk, because JBD has no rollback.
396 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
399 struct super_block
*sb
= inode
->i_sb
;
400 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
401 unsigned long gdb_num
= group
/ EXT4_DESC_PER_BLOCK(sb
);
402 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
403 struct buffer_head
**o_group_desc
, **n_group_desc
;
404 struct buffer_head
*dind
;
405 struct buffer_head
*gdb_bh
;
407 struct ext4_iloc iloc
;
411 if (test_opt(sb
, DEBUG
))
413 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
417 * If we are not using the primary superblock/GDT copy don't resize,
418 * because the user tools have no way of handling this. Probably a
419 * bad time to do it anyways.
421 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
422 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
423 ext4_warning(sb
, "won't resize using backup superblock at %llu",
424 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
428 gdb_bh
= sb_bread(sb
, gdblock
);
432 gdbackups
= verify_reserved_gdb(sb
, gdb_bh
);
438 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
439 dind
= sb_bread(sb
, le32_to_cpu(*data
));
445 data
= (__le32
*)dind
->b_data
;
446 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
447 ext4_warning(sb
, "new group %u GDT block %llu not reserved",
453 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
457 err
= ext4_journal_get_write_access(handle
, gdb_bh
);
461 err
= ext4_journal_get_write_access(handle
, dind
);
463 ext4_std_error(sb
, err
);
465 /* ext4_reserve_inode_write() gets a reference on the iloc */
466 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
470 n_group_desc
= kmalloc((gdb_num
+ 1) * sizeof(struct buffer_head
*),
475 "not enough memory for %lu groups", gdb_num
+ 1);
480 * Finally, we have all of the possible failures behind us...
482 * Remove new GDT block from inode double-indirect block and clear out
483 * the new GDT block for use (which also "frees" the backup GDT blocks
484 * from the reserved inode). We don't need to change the bitmaps for
485 * these blocks, because they are marked as in-use from being in the
486 * reserved inode, and will become GDT blocks (primary and backup).
488 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
489 err
= ext4_handle_dirty_metadata(handle
, NULL
, dind
);
491 ext4_std_error(sb
, err
);
494 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
495 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
496 memset(gdb_bh
->b_data
, 0, sb
->s_blocksize
);
497 err
= ext4_handle_dirty_metadata(handle
, NULL
, gdb_bh
);
499 ext4_std_error(sb
, err
);
504 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
505 memcpy(n_group_desc
, o_group_desc
,
506 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
507 n_group_desc
[gdb_num
] = gdb_bh
;
508 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
509 EXT4_SB(sb
)->s_gdb_count
++;
512 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
513 err
= ext4_handle_dirty_metadata(handle
, NULL
, EXT4_SB(sb
)->s_sbh
);
515 ext4_std_error(sb
, err
);
521 /* ext4_handle_release_buffer(handle, iloc.bh); */
524 /* ext4_handle_release_buffer(handle, dind); */
526 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
532 ext4_debug("leaving with error %d\n", err
);
537 * Called when we are adding a new group which has a backup copy of each of
538 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
539 * We need to add these reserved backup GDT blocks to the resize inode, so
540 * that they are kept for future resizing and not allocated to files.
542 * Each reserved backup GDT block will go into a different indirect block.
543 * The indirect blocks are actually the primary reserved GDT blocks,
544 * so we know in advance what their block numbers are. We only get the
545 * double-indirect block to verify it is pointing to the primary reserved
546 * GDT blocks so we don't overwrite a data block by accident. The reserved
547 * backup GDT blocks are stored in their reserved primary GDT block.
549 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
552 struct super_block
*sb
= inode
->i_sb
;
553 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
554 struct buffer_head
**primary
;
555 struct buffer_head
*dind
;
556 struct ext4_iloc iloc
;
563 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
567 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
568 dind
= sb_bread(sb
, le32_to_cpu(*data
));
574 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
575 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
576 EXT4_ADDR_PER_BLOCK(sb
));
577 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
579 /* Get each reserved primary GDT block and verify it holds backups */
580 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
581 if (le32_to_cpu(*data
) != blk
) {
582 ext4_warning(sb
, "reserved block %llu"
583 " not at offset %ld",
585 (long)(data
- (__le32
*)dind
->b_data
));
589 primary
[res
] = sb_bread(sb
, blk
);
594 if ((gdbackups
= verify_reserved_gdb(sb
, primary
[res
])) < 0) {
595 brelse(primary
[res
]);
600 data
= (__le32
*)dind
->b_data
;
603 for (i
= 0; i
< reserved_gdb
; i
++) {
604 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
607 for (j = 0; j < i; j++)
608 ext4_handle_release_buffer(handle, primary[j]);
614 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
618 * Finally we can add each of the reserved backup GDT blocks from
619 * the new group to its reserved primary GDT block.
621 blk
= group
* EXT4_BLOCKS_PER_GROUP(sb
);
622 for (i
= 0; i
< reserved_gdb
; i
++) {
624 data
= (__le32
*)primary
[i
]->b_data
;
625 /* printk("reserving backup %lu[%u] = %lu\n",
626 primary[i]->b_blocknr, gdbackups,
627 blk + primary[i]->b_blocknr); */
628 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
629 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
633 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
634 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
638 brelse(primary
[res
]);
648 * Update the backup copies of the ext4 metadata. These don't need to be part
649 * of the main resize transaction, because e2fsck will re-write them if there
650 * is a problem (basically only OOM will cause a problem). However, we
651 * _should_ update the backups if possible, in case the primary gets trashed
652 * for some reason and we need to run e2fsck from a backup superblock. The
653 * important part is that the new block and inode counts are in the backup
654 * superblocks, and the location of the new group metadata in the GDT backups.
656 * We do not need take the s_resize_lock for this, because these
657 * blocks are not otherwise touched by the filesystem code when it is
658 * mounted. We don't need to worry about last changing from
659 * sbi->s_groups_count, because the worst that can happen is that we
660 * do not copy the full number of backups at this time. The resize
661 * which changed s_groups_count will backup again.
663 static void update_backups(struct super_block
*sb
,
664 int blk_off
, char *data
, int size
)
666 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
667 const ext4_group_t last
= sbi
->s_groups_count
;
668 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
673 int rest
= sb
->s_blocksize
- size
;
677 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
678 if (IS_ERR(handle
)) {
680 err
= PTR_ERR(handle
);
684 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
685 struct buffer_head
*bh
;
687 /* Out of journal space, and can't get more - abort - so sad */
688 if (ext4_handle_valid(handle
) &&
689 handle
->h_buffer_credits
== 0 &&
690 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
691 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
694 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
699 ext4_debug("update metadata backup %#04lx\n",
700 (unsigned long)bh
->b_blocknr
);
701 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
704 memcpy(bh
->b_data
, data
, size
);
706 memset(bh
->b_data
+ size
, 0, rest
);
707 set_buffer_uptodate(bh
);
709 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
711 ext4_std_error(sb
, err
);
714 if ((err2
= ext4_journal_stop(handle
)) && !err
)
718 * Ugh! Need to have e2fsck write the backup copies. It is too
719 * late to revert the resize, we shouldn't fail just because of
720 * the backup copies (they are only needed in case of corruption).
722 * However, if we got here we have a journal problem too, so we
723 * can't really start a transaction to mark the superblock.
724 * Chicken out and just set the flag on the hope it will be written
725 * to disk, and if not - we will simply wait until next fsck.
729 ext4_warning(sb
, "can't update backup for group %u (err %d), "
730 "forcing fsck on next reboot", group
, err
);
731 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
732 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
733 mark_buffer_dirty(sbi
->s_sbh
);
737 /* Add group descriptor data to an existing or new group descriptor block.
738 * Ensure we handle all possible error conditions _before_ we start modifying
739 * the filesystem, because we cannot abort the transaction and not have it
740 * write the data to disk.
742 * If we are on a GDT block boundary, we need to get the reserved GDT block.
743 * Otherwise, we may need to add backup GDT blocks for a sparse group.
745 * We only need to hold the superblock lock while we are actually adding
746 * in the new group's counts to the superblock. Prior to that we have
747 * not really "added" the group at all. We re-check that we are still
748 * adding in the last group in case things have changed since verifying.
750 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
752 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
753 struct ext4_super_block
*es
= sbi
->s_es
;
754 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
755 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
756 struct buffer_head
*primary
= NULL
;
757 struct ext4_group_desc
*gdp
;
758 struct inode
*inode
= NULL
;
760 int gdb_off
, gdb_num
;
763 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
764 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
766 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
767 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
768 ext4_warning(sb
, "Can't resize non-sparse filesystem further");
772 if (ext4_blocks_count(es
) + input
->blocks_count
<
773 ext4_blocks_count(es
)) {
774 ext4_warning(sb
, "blocks_count overflow");
778 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
779 le32_to_cpu(es
->s_inodes_count
)) {
780 ext4_warning(sb
, "inodes_count overflow");
784 if (reserved_gdb
|| gdb_off
== 0) {
785 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
786 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
787 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
789 "No reserved GDT blocks, can't resize");
792 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
794 ext4_warning(sb
, "Error opening resize inode");
795 return PTR_ERR(inode
);
800 if ((err
= verify_group_input(sb
, input
)))
803 if ((err
= setup_new_group_blocks(sb
, input
)))
807 * We will always be modifying at least the superblock and a GDT
808 * block. If we are adding a group past the last current GDT block,
809 * we will also modify the inode and the dindirect block. If we
810 * are adding a group with superblock/GDT backups we will also
811 * modify each of the reserved GDT dindirect blocks.
813 handle
= ext4_journal_start_sb(sb
,
814 ext4_bg_has_super(sb
, input
->group
) ?
815 3 + reserved_gdb
: 4);
816 if (IS_ERR(handle
)) {
817 err
= PTR_ERR(handle
);
821 if ((err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
)))
825 * We will only either add reserved group blocks to a backup group
826 * or remove reserved blocks for the first group in a new group block.
827 * Doing both would be mean more complex code, and sane people don't
828 * use non-sparse filesystems anymore. This is already checked above.
831 primary
= sbi
->s_group_desc
[gdb_num
];
832 if ((err
= ext4_journal_get_write_access(handle
, primary
)))
835 if (reserved_gdb
&& ext4_bg_num_gdb(sb
, input
->group
)) {
836 err
= reserve_backup_gdb(handle
, inode
, input
->group
);
842 * Note that we can access new group descriptor block safely
843 * only if add_new_gdb() succeeds.
845 err
= add_new_gdb(handle
, inode
, input
->group
);
848 primary
= sbi
->s_group_desc
[gdb_num
];
852 * OK, now we've set up the new group. Time to make it active.
854 * so we have to be safe wrt. concurrent accesses the group
855 * data. So we need to be careful to set all of the relevant
856 * group descriptor data etc. *before* we enable the group.
858 * The key field here is sbi->s_groups_count: as long as
859 * that retains its old value, nobody is going to access the new
862 * So first we update all the descriptor metadata for the new
863 * group; then we update the total disk blocks count; then we
864 * update the groups count to enable the group; then finally we
865 * update the free space counts so that the system can start
866 * using the new disk blocks.
869 /* Update group descriptor block for new group */
870 gdp
= (struct ext4_group_desc
*)((char *)primary
->b_data
+
871 gdb_off
* EXT4_DESC_SIZE(sb
));
873 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
874 ext4_block_bitmap_set(sb
, gdp
, input
->block_bitmap
); /* LV FIXME */
875 ext4_inode_bitmap_set(sb
, gdp
, input
->inode_bitmap
); /* LV FIXME */
876 ext4_inode_table_set(sb
, gdp
, input
->inode_table
); /* LV FIXME */
877 ext4_free_blks_set(sb
, gdp
, input
->free_blocks_count
);
878 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
879 gdp
->bg_flags
= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
880 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, input
->group
, gdp
);
883 * We can allocate memory for mb_alloc based on the new group
886 err
= ext4_mb_add_groupinfo(sb
, input
->group
, gdp
);
891 * Make the new blocks and inodes valid next. We do this before
892 * increasing the group count so that once the group is enabled,
893 * all of its blocks and inodes are already valid.
895 * We always allocate group-by-group, then block-by-block or
896 * inode-by-inode within a group, so enabling these
897 * blocks/inodes before the group is live won't actually let us
898 * allocate the new space yet.
900 ext4_blocks_count_set(es
, ext4_blocks_count(es
) +
901 input
->blocks_count
);
902 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
));
905 * We need to protect s_groups_count against other CPUs seeing
906 * inconsistent state in the superblock.
908 * The precise rules we use are:
910 * * Writers must perform a smp_wmb() after updating all dependent
911 * data and before modifying the groups count
913 * * Readers must perform an smp_rmb() after reading the groups count
914 * and before reading any dependent data.
916 * NB. These rules can be relaxed when checking the group count
917 * while freeing data, as we can only allocate from a block
918 * group after serialising against the group count, and we can
919 * only then free after serialising in turn against that
924 /* Update the global fs size fields */
925 sbi
->s_groups_count
++;
927 err
= ext4_handle_dirty_metadata(handle
, NULL
, primary
);
929 ext4_std_error(sb
, err
);
933 /* Update the reserved block counts only once the new group is
935 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
936 input
->reserved_blocks
);
938 /* Update the free space counts */
939 percpu_counter_add(&sbi
->s_freeblocks_counter
,
940 input
->free_blocks_count
);
941 percpu_counter_add(&sbi
->s_freeinodes_counter
,
942 EXT4_INODES_PER_GROUP(sb
));
944 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
) &&
945 sbi
->s_log_groups_per_flex
) {
946 ext4_group_t flex_group
;
947 flex_group
= ext4_flex_group(sbi
, input
->group
);
948 atomic_add(input
->free_blocks_count
,
949 &sbi
->s_flex_groups
[flex_group
].free_blocks
);
950 atomic_add(EXT4_INODES_PER_GROUP(sb
),
951 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
954 ext4_handle_dirty_super(handle
, sb
);
957 if ((err2
= ext4_journal_stop(handle
)) && !err
)
959 if (!err
&& primary
) {
960 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
961 sizeof(struct ext4_super_block
));
962 update_backups(sb
, primary
->b_blocknr
, primary
->b_data
,
968 } /* ext4_group_add */
971 * Extend the filesystem to the new number of blocks specified. This entry
972 * point is only used to extend the current filesystem to the end of the last
973 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
974 * for emergencies (because it has no dependencies on reserved blocks).
976 * If we _really_ wanted, we could use default values to call ext4_group_add()
977 * allow the "remount" trick to work for arbitrary resizing, assuming enough
978 * GDT blocks are reserved to grow to the desired size.
980 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
981 ext4_fsblk_t n_blocks_count
)
983 ext4_fsblk_t o_blocks_count
;
986 struct buffer_head
*bh
;
991 o_blocks_count
= ext4_blocks_count(es
);
993 if (test_opt(sb
, DEBUG
))
994 printk(KERN_DEBUG
"EXT4-fs: extending last group from %llu to %llu blocks\n",
995 o_blocks_count
, n_blocks_count
);
997 if (n_blocks_count
== 0 || n_blocks_count
== o_blocks_count
)
1000 if (n_blocks_count
> (sector_t
)(~0ULL) >> (sb
->s_blocksize_bits
- 9)) {
1001 printk(KERN_ERR
"EXT4-fs: filesystem on %s:"
1002 " too large to resize to %llu blocks safely\n",
1003 sb
->s_id
, n_blocks_count
);
1004 if (sizeof(sector_t
) < 8)
1005 ext4_warning(sb
, "CONFIG_LBDAF not enabled");
1009 if (n_blocks_count
< o_blocks_count
) {
1010 ext4_warning(sb
, "can't shrink FS - resize aborted");
1014 /* Handle the remaining blocks in the last group only. */
1015 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1018 ext4_warning(sb
, "need to use ext2online to resize further");
1022 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1024 if (o_blocks_count
+ add
< o_blocks_count
) {
1025 ext4_warning(sb
, "blocks_count overflow");
1029 if (o_blocks_count
+ add
> n_blocks_count
)
1030 add
= n_blocks_count
- o_blocks_count
;
1032 if (o_blocks_count
+ add
< n_blocks_count
)
1033 ext4_warning(sb
, "will only finish group (%llu blocks, %u new)",
1034 o_blocks_count
+ add
, add
);
1036 /* See if the device is actually as big as what was requested */
1037 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1039 ext4_warning(sb
, "can't read last block, resize aborted");
1044 /* We will update the superblock, one block bitmap, and
1045 * one group descriptor via ext4_free_blocks().
1047 handle
= ext4_journal_start_sb(sb
, 3);
1048 if (IS_ERR(handle
)) {
1049 err
= PTR_ERR(handle
);
1050 ext4_warning(sb
, "error %d on journal start", err
);
1054 if ((err
= ext4_journal_get_write_access(handle
,
1055 EXT4_SB(sb
)->s_sbh
))) {
1056 ext4_warning(sb
, "error %d on journal write access", err
);
1057 ext4_journal_stop(handle
);
1060 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1061 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1062 o_blocks_count
+ add
);
1063 /* We add the blocks to the bitmap and set the group need init bit */
1064 err
= ext4_group_add_blocks(handle
, sb
, o_blocks_count
, add
);
1065 ext4_handle_dirty_super(handle
, sb
);
1066 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1067 o_blocks_count
+ add
);
1068 err2
= ext4_journal_stop(handle
);
1075 if (test_opt(sb
, DEBUG
))
1076 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu blocks\n",
1077 ext4_blocks_count(es
));
1078 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1079 sizeof(struct ext4_super_block
));
1082 } /* ext4_group_extend */