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"
20 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
21 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
23 static int verify_group_input(struct super_block
*sb
,
24 struct ext4_new_group_data
*input
)
26 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
27 struct ext4_super_block
*es
= sbi
->s_es
;
28 ext4_fsblk_t start
= ext4_blocks_count(es
);
29 ext4_fsblk_t end
= start
+ input
->blocks_count
;
30 ext4_group_t group
= input
->group
;
31 ext4_fsblk_t itend
= input
->inode_table
+ sbi
->s_itb_per_group
;
32 unsigned overhead
= ext4_bg_has_super(sb
, group
) ?
33 (1 + ext4_bg_num_gdb(sb
, group
) +
34 le16_to_cpu(es
->s_reserved_gdt_blocks
)) : 0;
35 ext4_fsblk_t metaend
= start
+ overhead
;
36 struct buffer_head
*bh
= NULL
;
37 ext4_grpblk_t free_blocks_count
, offset
;
40 input
->free_blocks_count
= free_blocks_count
=
41 input
->blocks_count
- 2 - overhead
- sbi
->s_itb_per_group
;
43 if (test_opt(sb
, DEBUG
))
44 printk(KERN_DEBUG
"EXT4-fs: adding %s group %u: %u blocks "
45 "(%d free, %u reserved)\n",
46 ext4_bg_has_super(sb
, input
->group
) ? "normal" :
47 "no-super", input
->group
, input
->blocks_count
,
48 free_blocks_count
, input
->reserved_blocks
);
50 ext4_get_group_no_and_offset(sb
, start
, NULL
, &offset
);
51 if (group
!= sbi
->s_groups_count
)
52 ext4_warning(sb
, __func__
,
53 "Cannot add at group %u (only %u groups)",
54 input
->group
, sbi
->s_groups_count
);
56 ext4_warning(sb
, __func__
, "Last group not full");
57 else if (input
->reserved_blocks
> input
->blocks_count
/ 5)
58 ext4_warning(sb
, __func__
, "Reserved blocks too high (%u)",
59 input
->reserved_blocks
);
60 else if (free_blocks_count
< 0)
61 ext4_warning(sb
, __func__
, "Bad blocks count %u",
63 else if (!(bh
= sb_bread(sb
, end
- 1)))
64 ext4_warning(sb
, __func__
,
65 "Cannot read last block (%llu)",
67 else if (outside(input
->block_bitmap
, start
, end
))
68 ext4_warning(sb
, __func__
,
69 "Block bitmap not in group (block %llu)",
70 (unsigned long long)input
->block_bitmap
);
71 else if (outside(input
->inode_bitmap
, start
, end
))
72 ext4_warning(sb
, __func__
,
73 "Inode bitmap not in group (block %llu)",
74 (unsigned long long)input
->inode_bitmap
);
75 else if (outside(input
->inode_table
, start
, end
) ||
76 outside(itend
- 1, start
, end
))
77 ext4_warning(sb
, __func__
,
78 "Inode table not in group (blocks %llu-%llu)",
79 (unsigned long long)input
->inode_table
, itend
- 1);
80 else if (input
->inode_bitmap
== input
->block_bitmap
)
81 ext4_warning(sb
, __func__
,
82 "Block bitmap same as inode bitmap (%llu)",
83 (unsigned long long)input
->block_bitmap
);
84 else if (inside(input
->block_bitmap
, input
->inode_table
, itend
))
85 ext4_warning(sb
, __func__
,
86 "Block bitmap (%llu) in inode table (%llu-%llu)",
87 (unsigned long long)input
->block_bitmap
,
88 (unsigned long long)input
->inode_table
, itend
- 1);
89 else if (inside(input
->inode_bitmap
, input
->inode_table
, itend
))
90 ext4_warning(sb
, __func__
,
91 "Inode bitmap (%llu) in inode table (%llu-%llu)",
92 (unsigned long long)input
->inode_bitmap
,
93 (unsigned long long)input
->inode_table
, itend
- 1);
94 else if (inside(input
->block_bitmap
, start
, metaend
))
95 ext4_warning(sb
, __func__
,
96 "Block bitmap (%llu) in GDT table"
98 (unsigned long long)input
->block_bitmap
,
100 else if (inside(input
->inode_bitmap
, start
, metaend
))
101 ext4_warning(sb
, __func__
,
102 "Inode bitmap (%llu) in GDT table"
104 (unsigned long long)input
->inode_bitmap
,
106 else if (inside(input
->inode_table
, start
, metaend
) ||
107 inside(itend
- 1, start
, metaend
))
108 ext4_warning(sb
, __func__
,
109 "Inode table (%llu-%llu) overlaps"
110 "GDT table (%llu-%llu)",
111 (unsigned long long)input
->inode_table
,
112 itend
- 1, start
, metaend
- 1);
120 static struct buffer_head
*bclean(handle_t
*handle
, struct super_block
*sb
,
123 struct buffer_head
*bh
;
126 bh
= sb_getblk(sb
, blk
);
128 return ERR_PTR(-EIO
);
129 if ((err
= ext4_journal_get_write_access(handle
, bh
))) {
134 memset(bh
->b_data
, 0, sb
->s_blocksize
);
135 set_buffer_uptodate(bh
);
143 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
144 * If that fails, restart the transaction & regain write access for the
145 * buffer head which is used for block_bitmap modifications.
147 static int extend_or_restart_transaction(handle_t
*handle
, int thresh
,
148 struct buffer_head
*bh
)
152 if (ext4_handle_has_enough_credits(handle
, thresh
))
155 err
= ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
);
159 if ((err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
161 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
169 * Set up the block and inode bitmaps, and the inode table for the new group.
170 * This doesn't need to be part of the main transaction, since we are only
171 * changing blocks outside the actual filesystem. We still do journaling to
172 * ensure the recovery is correct in case of a failure just after resize.
173 * If any part of this fails, we simply abort the resize.
175 static int setup_new_group_blocks(struct super_block
*sb
,
176 struct ext4_new_group_data
*input
)
178 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
179 ext4_fsblk_t start
= ext4_group_first_block_no(sb
, input
->group
);
180 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
181 le16_to_cpu(sbi
->s_es
->s_reserved_gdt_blocks
) : 0;
182 unsigned long gdblocks
= ext4_bg_num_gdb(sb
, input
->group
);
183 struct buffer_head
*bh
;
190 /* This transaction may be extended/restarted along the way */
191 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
194 return PTR_ERR(handle
);
197 if (input
->group
!= sbi
->s_groups_count
) {
202 if (IS_ERR(bh
= bclean(handle
, sb
, input
->block_bitmap
))) {
207 if (ext4_bg_has_super(sb
, input
->group
)) {
208 ext4_debug("mark backup superblock %#04llx (+0)\n", start
);
209 ext4_set_bit(0, bh
->b_data
);
212 /* Copy all of the GDT blocks into the backup in this group */
213 for (i
= 0, bit
= 1, block
= start
+ 1;
214 i
< gdblocks
; i
++, block
++, bit
++) {
215 struct buffer_head
*gdb
;
217 ext4_debug("update backup group %#04llx (+%d)\n", block
, bit
);
219 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
222 gdb
= sb_getblk(sb
, block
);
227 if ((err
= ext4_journal_get_write_access(handle
, gdb
))) {
232 memcpy(gdb
->b_data
, sbi
->s_group_desc
[i
]->b_data
, gdb
->b_size
);
233 set_buffer_uptodate(gdb
);
235 ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
236 ext4_set_bit(bit
, bh
->b_data
);
240 /* Zero out all of the reserved backup group descriptor table blocks */
241 for (i
= 0, bit
= gdblocks
+ 1, block
= start
+ bit
;
242 i
< reserved_gdb
; i
++, block
++, bit
++) {
243 struct buffer_head
*gdb
;
245 ext4_debug("clear reserved block %#04llx (+%d)\n", block
, bit
);
247 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
250 if (IS_ERR(gdb
= bclean(handle
, sb
, block
))) {
254 ext4_handle_dirty_metadata(handle
, NULL
, gdb
);
255 ext4_set_bit(bit
, bh
->b_data
);
258 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input
->block_bitmap
,
259 input
->block_bitmap
- start
);
260 ext4_set_bit(input
->block_bitmap
- start
, bh
->b_data
);
261 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input
->inode_bitmap
,
262 input
->inode_bitmap
- start
);
263 ext4_set_bit(input
->inode_bitmap
- start
, bh
->b_data
);
265 /* Zero out all of the inode table blocks */
266 for (i
= 0, block
= input
->inode_table
, bit
= block
- start
;
267 i
< sbi
->s_itb_per_group
; i
++, bit
++, block
++) {
268 struct buffer_head
*it
;
270 ext4_debug("clear inode block %#04llx (+%d)\n", block
, bit
);
272 if ((err
= extend_or_restart_transaction(handle
, 1, bh
)))
275 if (IS_ERR(it
= bclean(handle
, sb
, block
))) {
279 ext4_handle_dirty_metadata(handle
, NULL
, it
);
281 ext4_set_bit(bit
, bh
->b_data
);
284 if ((err
= extend_or_restart_transaction(handle
, 2, bh
)))
287 mark_bitmap_end(input
->blocks_count
, sb
->s_blocksize
* 8, bh
->b_data
);
288 ext4_handle_dirty_metadata(handle
, NULL
, bh
);
290 /* Mark unused entries in inode bitmap used */
291 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
292 input
->inode_bitmap
, input
->inode_bitmap
- start
);
293 if (IS_ERR(bh
= bclean(handle
, sb
, input
->inode_bitmap
))) {
298 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb
), sb
->s_blocksize
* 8,
300 ext4_handle_dirty_metadata(handle
, NULL
, bh
);
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
, __func__
,
370 " missing grp %d (%llu)",
373 (ext4_fsblk_t
)EXT4_BLOCKS_PER_GROUP(sb
) +
377 if (++gdbackups
> EXT4_ADDR_PER_BLOCK(sb
))
385 * Called when we need to bring a reserved group descriptor table block into
386 * use from the resize inode. The primary copy of the new GDT block currently
387 * is an indirect block (under the double indirect block in the resize inode).
388 * The new backup GDT blocks will be stored as leaf blocks in this indirect
389 * block, in group order. Even though we know all the block numbers we need,
390 * we check to ensure that the resize inode has actually reserved these blocks.
392 * Don't need to update the block bitmaps because the blocks are still in use.
394 * We get all of the error cases out of the way, so that we are sure to not
395 * fail once we start modifying the data on disk, because JBD has no rollback.
397 static int add_new_gdb(handle_t
*handle
, struct inode
*inode
,
398 struct ext4_new_group_data
*input
,
399 struct buffer_head
**primary
)
401 struct super_block
*sb
= inode
->i_sb
;
402 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
403 unsigned long gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
404 ext4_fsblk_t gdblock
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + gdb_num
;
405 struct buffer_head
**o_group_desc
, **n_group_desc
;
406 struct buffer_head
*dind
;
408 struct ext4_iloc iloc
;
412 if (test_opt(sb
, DEBUG
))
414 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
418 * If we are not using the primary superblock/GDT copy don't resize,
419 * because the user tools have no way of handling this. Probably a
420 * bad time to do it anyways.
422 if (EXT4_SB(sb
)->s_sbh
->b_blocknr
!=
423 le32_to_cpu(EXT4_SB(sb
)->s_es
->s_first_data_block
)) {
424 ext4_warning(sb
, __func__
,
425 "won't resize using backup superblock at %llu",
426 (unsigned long long)EXT4_SB(sb
)->s_sbh
->b_blocknr
);
430 *primary
= sb_bread(sb
, gdblock
);
434 if ((gdbackups
= verify_reserved_gdb(sb
, *primary
)) < 0) {
439 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
440 dind
= sb_bread(sb
, le32_to_cpu(*data
));
446 data
= (__le32
*)dind
->b_data
;
447 if (le32_to_cpu(data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)]) != gdblock
) {
448 ext4_warning(sb
, __func__
,
449 "new group %u GDT block %llu not reserved",
450 input
->group
, gdblock
);
455 if ((err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
)))
458 if ((err
= ext4_journal_get_write_access(handle
, *primary
)))
461 if ((err
= ext4_journal_get_write_access(handle
, dind
)))
464 /* ext4_reserve_inode_write() gets a reference on the iloc */
465 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
468 n_group_desc
= kmalloc((gdb_num
+ 1) * sizeof(struct buffer_head
*),
472 ext4_warning(sb
, __func__
,
473 "not enough memory for %lu groups", gdb_num
+ 1);
478 * Finally, we have all of the possible failures behind us...
480 * Remove new GDT block from inode double-indirect block and clear out
481 * the new GDT block for use (which also "frees" the backup GDT blocks
482 * from the reserved inode). We don't need to change the bitmaps for
483 * these blocks, because they are marked as in-use from being in the
484 * reserved inode, and will become GDT blocks (primary and backup).
486 data
[gdb_num
% EXT4_ADDR_PER_BLOCK(sb
)] = 0;
487 ext4_handle_dirty_metadata(handle
, NULL
, dind
);
489 inode
->i_blocks
-= (gdbackups
+ 1) * sb
->s_blocksize
>> 9;
490 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
491 memset((*primary
)->b_data
, 0, sb
->s_blocksize
);
492 ext4_handle_dirty_metadata(handle
, NULL
, *primary
);
494 o_group_desc
= EXT4_SB(sb
)->s_group_desc
;
495 memcpy(n_group_desc
, o_group_desc
,
496 EXT4_SB(sb
)->s_gdb_count
* sizeof(struct buffer_head
*));
497 n_group_desc
[gdb_num
] = *primary
;
498 EXT4_SB(sb
)->s_group_desc
= n_group_desc
;
499 EXT4_SB(sb
)->s_gdb_count
++;
502 le16_add_cpu(&es
->s_reserved_gdt_blocks
, -1);
503 ext4_handle_dirty_metadata(handle
, NULL
, EXT4_SB(sb
)->s_sbh
);
508 /* ext4_journal_release_buffer(handle, iloc.bh); */
511 /* ext4_journal_release_buffer(handle, dind); */
513 /* ext4_journal_release_buffer(handle, *primary); */
515 /* ext4_journal_release_buffer(handle, *primary); */
521 ext4_debug("leaving with error %d\n", err
);
526 * Called when we are adding a new group which has a backup copy of each of
527 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
528 * We need to add these reserved backup GDT blocks to the resize inode, so
529 * that they are kept for future resizing and not allocated to files.
531 * Each reserved backup GDT block will go into a different indirect block.
532 * The indirect blocks are actually the primary reserved GDT blocks,
533 * so we know in advance what their block numbers are. We only get the
534 * double-indirect block to verify it is pointing to the primary reserved
535 * GDT blocks so we don't overwrite a data block by accident. The reserved
536 * backup GDT blocks are stored in their reserved primary GDT block.
538 static int reserve_backup_gdb(handle_t
*handle
, struct inode
*inode
,
539 struct ext4_new_group_data
*input
)
541 struct super_block
*sb
= inode
->i_sb
;
542 int reserved_gdb
=le16_to_cpu(EXT4_SB(sb
)->s_es
->s_reserved_gdt_blocks
);
543 struct buffer_head
**primary
;
544 struct buffer_head
*dind
;
545 struct ext4_iloc iloc
;
552 primary
= kmalloc(reserved_gdb
* sizeof(*primary
), GFP_NOFS
);
556 data
= EXT4_I(inode
)->i_data
+ EXT4_DIND_BLOCK
;
557 dind
= sb_bread(sb
, le32_to_cpu(*data
));
563 blk
= EXT4_SB(sb
)->s_sbh
->b_blocknr
+ 1 + EXT4_SB(sb
)->s_gdb_count
;
564 data
= (__le32
*)dind
->b_data
+ (EXT4_SB(sb
)->s_gdb_count
%
565 EXT4_ADDR_PER_BLOCK(sb
));
566 end
= (__le32
*)dind
->b_data
+ EXT4_ADDR_PER_BLOCK(sb
);
568 /* Get each reserved primary GDT block and verify it holds backups */
569 for (res
= 0; res
< reserved_gdb
; res
++, blk
++) {
570 if (le32_to_cpu(*data
) != blk
) {
571 ext4_warning(sb
, __func__
,
572 "reserved block %llu"
573 " not at offset %ld",
575 (long)(data
- (__le32
*)dind
->b_data
));
579 primary
[res
] = sb_bread(sb
, blk
);
584 if ((gdbackups
= verify_reserved_gdb(sb
, primary
[res
])) < 0) {
585 brelse(primary
[res
]);
590 data
= (__le32
*)dind
->b_data
;
593 for (i
= 0; i
< reserved_gdb
; i
++) {
594 if ((err
= ext4_journal_get_write_access(handle
, primary
[i
]))) {
597 for (j = 0; j < i; j++)
598 ext4_journal_release_buffer(handle, primary[j]);
604 if ((err
= ext4_reserve_inode_write(handle
, inode
, &iloc
)))
608 * Finally we can add each of the reserved backup GDT blocks from
609 * the new group to its reserved primary GDT block.
611 blk
= input
->group
* EXT4_BLOCKS_PER_GROUP(sb
);
612 for (i
= 0; i
< reserved_gdb
; i
++) {
614 data
= (__le32
*)primary
[i
]->b_data
;
615 /* printk("reserving backup %lu[%u] = %lu\n",
616 primary[i]->b_blocknr, gdbackups,
617 blk + primary[i]->b_blocknr); */
618 data
[gdbackups
] = cpu_to_le32(blk
+ primary
[i
]->b_blocknr
);
619 err2
= ext4_handle_dirty_metadata(handle
, NULL
, primary
[i
]);
623 inode
->i_blocks
+= reserved_gdb
* sb
->s_blocksize
>> 9;
624 ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
628 brelse(primary
[res
]);
638 * Update the backup copies of the ext4 metadata. These don't need to be part
639 * of the main resize transaction, because e2fsck will re-write them if there
640 * is a problem (basically only OOM will cause a problem). However, we
641 * _should_ update the backups if possible, in case the primary gets trashed
642 * for some reason and we need to run e2fsck from a backup superblock. The
643 * important part is that the new block and inode counts are in the backup
644 * superblocks, and the location of the new group metadata in the GDT backups.
646 * We do not need lock_super() for this, because these blocks are not
647 * otherwise touched by the filesystem code when it is mounted. We don't
648 * need to worry about last changing from sbi->s_groups_count, because the
649 * worst that can happen is that we do not copy the full number of backups
650 * at this time. The resize which changed s_groups_count will backup again.
652 static void update_backups(struct super_block
*sb
,
653 int blk_off
, char *data
, int size
)
655 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
656 const ext4_group_t last
= sbi
->s_groups_count
;
657 const int bpg
= EXT4_BLOCKS_PER_GROUP(sb
);
662 int rest
= sb
->s_blocksize
- size
;
666 handle
= ext4_journal_start_sb(sb
, EXT4_MAX_TRANS_DATA
);
667 if (IS_ERR(handle
)) {
669 err
= PTR_ERR(handle
);
673 while ((group
= ext4_list_backups(sb
, &three
, &five
, &seven
)) < last
) {
674 struct buffer_head
*bh
;
676 /* Out of journal space, and can't get more - abort - so sad */
677 if (ext4_handle_valid(handle
) &&
678 handle
->h_buffer_credits
== 0 &&
679 ext4_journal_extend(handle
, EXT4_MAX_TRANS_DATA
) &&
680 (err
= ext4_journal_restart(handle
, EXT4_MAX_TRANS_DATA
)))
683 bh
= sb_getblk(sb
, group
* bpg
+ blk_off
);
688 ext4_debug("update metadata backup %#04lx\n",
689 (unsigned long)bh
->b_blocknr
);
690 if ((err
= ext4_journal_get_write_access(handle
, bh
)))
693 memcpy(bh
->b_data
, data
, size
);
695 memset(bh
->b_data
+ size
, 0, rest
);
696 set_buffer_uptodate(bh
);
698 ext4_handle_dirty_metadata(handle
, NULL
, bh
);
701 if ((err2
= ext4_journal_stop(handle
)) && !err
)
705 * Ugh! Need to have e2fsck write the backup copies. It is too
706 * late to revert the resize, we shouldn't fail just because of
707 * the backup copies (they are only needed in case of corruption).
709 * However, if we got here we have a journal problem too, so we
710 * can't really start a transaction to mark the superblock.
711 * Chicken out and just set the flag on the hope it will be written
712 * to disk, and if not - we will simply wait until next fsck.
716 ext4_warning(sb
, __func__
,
717 "can't update backup for group %u (err %d), "
718 "forcing fsck on next reboot", group
, err
);
719 sbi
->s_mount_state
&= ~EXT4_VALID_FS
;
720 sbi
->s_es
->s_state
&= cpu_to_le16(~EXT4_VALID_FS
);
721 mark_buffer_dirty(sbi
->s_sbh
);
725 /* Add group descriptor data to an existing or new group descriptor block.
726 * Ensure we handle all possible error conditions _before_ we start modifying
727 * the filesystem, because we cannot abort the transaction and not have it
728 * write the data to disk.
730 * If we are on a GDT block boundary, we need to get the reserved GDT block.
731 * Otherwise, we may need to add backup GDT blocks for a sparse group.
733 * We only need to hold the superblock lock while we are actually adding
734 * in the new group's counts to the superblock. Prior to that we have
735 * not really "added" the group at all. We re-check that we are still
736 * adding in the last group in case things have changed since verifying.
738 int ext4_group_add(struct super_block
*sb
, struct ext4_new_group_data
*input
)
740 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
741 struct ext4_super_block
*es
= sbi
->s_es
;
742 int reserved_gdb
= ext4_bg_has_super(sb
, input
->group
) ?
743 le16_to_cpu(es
->s_reserved_gdt_blocks
) : 0;
744 struct buffer_head
*primary
= NULL
;
745 struct ext4_group_desc
*gdp
;
746 struct inode
*inode
= NULL
;
748 int gdb_off
, gdb_num
;
749 int num_grp_locked
= 0;
752 gdb_num
= input
->group
/ EXT4_DESC_PER_BLOCK(sb
);
753 gdb_off
= input
->group
% EXT4_DESC_PER_BLOCK(sb
);
755 if (gdb_off
== 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb
,
756 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER
)) {
757 ext4_warning(sb
, __func__
,
758 "Can't resize non-sparse filesystem further");
762 if (ext4_blocks_count(es
) + input
->blocks_count
<
763 ext4_blocks_count(es
)) {
764 ext4_warning(sb
, __func__
, "blocks_count overflow");
768 if (le32_to_cpu(es
->s_inodes_count
) + EXT4_INODES_PER_GROUP(sb
) <
769 le32_to_cpu(es
->s_inodes_count
)) {
770 ext4_warning(sb
, __func__
, "inodes_count overflow");
774 if (reserved_gdb
|| gdb_off
== 0) {
775 if (!EXT4_HAS_COMPAT_FEATURE(sb
,
776 EXT4_FEATURE_COMPAT_RESIZE_INODE
)
777 || !le16_to_cpu(es
->s_reserved_gdt_blocks
)) {
778 ext4_warning(sb
, __func__
,
779 "No reserved GDT blocks, can't resize");
782 inode
= ext4_iget(sb
, EXT4_RESIZE_INO
);
784 ext4_warning(sb
, __func__
,
785 "Error opening resize inode");
786 return PTR_ERR(inode
);
791 if ((err
= verify_group_input(sb
, input
)))
794 if ((err
= setup_new_group_blocks(sb
, input
)))
798 * We will always be modifying at least the superblock and a GDT
799 * block. If we are adding a group past the last current GDT block,
800 * we will also modify the inode and the dindirect block. If we
801 * are adding a group with superblock/GDT backups we will also
802 * modify each of the reserved GDT dindirect blocks.
804 handle
= ext4_journal_start_sb(sb
,
805 ext4_bg_has_super(sb
, input
->group
) ?
806 3 + reserved_gdb
: 4);
807 if (IS_ERR(handle
)) {
808 err
= PTR_ERR(handle
);
813 if (input
->group
!= sbi
->s_groups_count
) {
814 ext4_warning(sb
, __func__
,
815 "multiple resizers run on filesystem!");
820 if ((err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
)))
824 * We will only either add reserved group blocks to a backup group
825 * or remove reserved blocks for the first group in a new group block.
826 * Doing both would be mean more complex code, and sane people don't
827 * use non-sparse filesystems anymore. This is already checked above.
830 primary
= sbi
->s_group_desc
[gdb_num
];
831 if ((err
= ext4_journal_get_write_access(handle
, primary
)))
834 if (reserved_gdb
&& ext4_bg_num_gdb(sb
, input
->group
) &&
835 (err
= reserve_backup_gdb(handle
, inode
, input
)))
837 } else if ((err
= add_new_gdb(handle
, inode
, input
, &primary
)))
841 * OK, now we've set up the new group. Time to make it active.
843 * Current kernels don't lock all allocations via lock_super(),
844 * so we have to be safe wrt. concurrent accesses the group
845 * data. So we need to be careful to set all of the relevant
846 * group descriptor data etc. *before* we enable the group.
848 * The key field here is sbi->s_groups_count: as long as
849 * that retains its old value, nobody is going to access the new
852 * So first we update all the descriptor metadata for the new
853 * group; then we update the total disk blocks count; then we
854 * update the groups count to enable the group; then finally we
855 * update the free space counts so that the system can start
856 * using the new disk blocks.
859 num_grp_locked
= ext4_mb_get_buddy_cache_lock(sb
, input
->group
);
860 /* Update group descriptor block for new group */
861 gdp
= (struct ext4_group_desc
*)((char *)primary
->b_data
+
862 gdb_off
* EXT4_DESC_SIZE(sb
));
864 memset(gdp
, 0, EXT4_DESC_SIZE(sb
));
865 ext4_block_bitmap_set(sb
, gdp
, input
->block_bitmap
); /* LV FIXME */
866 ext4_inode_bitmap_set(sb
, gdp
, input
->inode_bitmap
); /* LV FIXME */
867 ext4_inode_table_set(sb
, gdp
, input
->inode_table
); /* LV FIXME */
868 ext4_free_blks_set(sb
, gdp
, input
->free_blocks_count
);
869 ext4_free_inodes_set(sb
, gdp
, EXT4_INODES_PER_GROUP(sb
));
870 gdp
->bg_flags
= cpu_to_le16(EXT4_BG_INODE_ZEROED
);
871 gdp
->bg_checksum
= ext4_group_desc_csum(sbi
, input
->group
, gdp
);
874 * We can allocate memory for mb_alloc based on the new group
877 err
= ext4_mb_add_groupinfo(sb
, input
->group
, gdp
);
879 ext4_mb_put_buddy_cache_lock(sb
, input
->group
, num_grp_locked
);
884 * Make the new blocks and inodes valid next. We do this before
885 * increasing the group count so that once the group is enabled,
886 * all of its blocks and inodes are already valid.
888 * We always allocate group-by-group, then block-by-block or
889 * inode-by-inode within a group, so enabling these
890 * blocks/inodes before the group is live won't actually let us
891 * allocate the new space yet.
893 ext4_blocks_count_set(es
, ext4_blocks_count(es
) +
894 input
->blocks_count
);
895 le32_add_cpu(&es
->s_inodes_count
, EXT4_INODES_PER_GROUP(sb
));
898 * We need to protect s_groups_count against other CPUs seeing
899 * inconsistent state in the superblock.
901 * The precise rules we use are:
903 * * Writers of s_groups_count *must* hold lock_super
905 * * Writers must perform a smp_wmb() after updating all dependent
906 * data and before modifying the groups count
908 * * Readers must hold lock_super() over the access
910 * * Readers must perform an smp_rmb() after reading the groups count
911 * and before reading any dependent data.
913 * NB. These rules can be relaxed when checking the group count
914 * while freeing data, as we can only allocate from a block
915 * group after serialising against the group count, and we can
916 * only then free after serialising in turn against that
921 /* Update the global fs size fields */
922 sbi
->s_groups_count
++;
923 ext4_mb_put_buddy_cache_lock(sb
, input
->group
, num_grp_locked
);
925 ext4_handle_dirty_metadata(handle
, NULL
, primary
);
927 /* Update the reserved block counts only once the new group is
929 ext4_r_blocks_count_set(es
, ext4_r_blocks_count(es
) +
930 input
->reserved_blocks
);
932 /* Update the free space counts */
933 percpu_counter_add(&sbi
->s_freeblocks_counter
,
934 input
->free_blocks_count
);
935 percpu_counter_add(&sbi
->s_freeinodes_counter
,
936 EXT4_INODES_PER_GROUP(sb
));
938 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_FLEX_BG
)) {
939 ext4_group_t flex_group
;
940 flex_group
= ext4_flex_group(sbi
, input
->group
);
941 atomic_add(input
->free_blocks_count
,
942 &sbi
->s_flex_groups
[flex_group
].free_blocks
);
943 atomic_add(EXT4_INODES_PER_GROUP(sb
),
944 &sbi
->s_flex_groups
[flex_group
].free_inodes
);
947 ext4_handle_dirty_metadata(handle
, NULL
, sbi
->s_sbh
);
952 if ((err2
= ext4_journal_stop(handle
)) && !err
)
955 update_backups(sb
, sbi
->s_sbh
->b_blocknr
, (char *)es
,
956 sizeof(struct ext4_super_block
));
957 update_backups(sb
, primary
->b_blocknr
, primary
->b_data
,
963 } /* ext4_group_add */
966 * Extend the filesystem to the new number of blocks specified. This entry
967 * point is only used to extend the current filesystem to the end of the last
968 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
969 * for emergencies (because it has no dependencies on reserved blocks).
971 * If we _really_ wanted, we could use default values to call ext4_group_add()
972 * allow the "remount" trick to work for arbitrary resizing, assuming enough
973 * GDT blocks are reserved to grow to the desired size.
975 int ext4_group_extend(struct super_block
*sb
, struct ext4_super_block
*es
,
976 ext4_fsblk_t n_blocks_count
)
978 ext4_fsblk_t o_blocks_count
;
979 ext4_group_t o_groups_count
;
982 struct buffer_head
*bh
;
987 /* We don't need to worry about locking wrt other resizers just
988 * yet: we're going to revalidate es->s_blocks_count after
989 * taking lock_super() below. */
990 o_blocks_count
= ext4_blocks_count(es
);
991 o_groups_count
= EXT4_SB(sb
)->s_groups_count
;
993 if (test_opt(sb
, DEBUG
))
994 printk(KERN_DEBUG
"EXT4-fs: extending last group from %llu uto %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
, __func__
, "CONFIG_LBD not enabled");
1009 if (n_blocks_count
< o_blocks_count
) {
1010 ext4_warning(sb
, __func__
,
1011 "can't shrink FS - resize aborted");
1015 /* Handle the remaining blocks in the last group only. */
1016 ext4_get_group_no_and_offset(sb
, o_blocks_count
, &group
, &last
);
1019 ext4_warning(sb
, __func__
,
1020 "need to use ext2online to resize further");
1024 add
= EXT4_BLOCKS_PER_GROUP(sb
) - last
;
1026 if (o_blocks_count
+ add
< o_blocks_count
) {
1027 ext4_warning(sb
, __func__
, "blocks_count overflow");
1031 if (o_blocks_count
+ add
> n_blocks_count
)
1032 add
= n_blocks_count
- o_blocks_count
;
1034 if (o_blocks_count
+ add
< n_blocks_count
)
1035 ext4_warning(sb
, __func__
,
1036 "will only finish group (%llu"
1038 o_blocks_count
+ add
, add
);
1040 /* See if the device is actually as big as what was requested */
1041 bh
= sb_bread(sb
, o_blocks_count
+ add
- 1);
1043 ext4_warning(sb
, __func__
,
1044 "can't read last block, resize aborted");
1049 /* We will update the superblock, one block bitmap, and
1050 * one group descriptor via ext4_free_blocks().
1052 handle
= ext4_journal_start_sb(sb
, 3);
1053 if (IS_ERR(handle
)) {
1054 err
= PTR_ERR(handle
);
1055 ext4_warning(sb
, __func__
, "error %d on journal start", err
);
1060 if (o_blocks_count
!= ext4_blocks_count(es
)) {
1061 ext4_warning(sb
, __func__
,
1062 "multiple resizers run on filesystem!");
1064 ext4_journal_stop(handle
);
1069 if ((err
= ext4_journal_get_write_access(handle
,
1070 EXT4_SB(sb
)->s_sbh
))) {
1071 ext4_warning(sb
, __func__
,
1072 "error %d on journal write access", err
);
1074 ext4_journal_stop(handle
);
1077 ext4_blocks_count_set(es
, o_blocks_count
+ add
);
1078 ext4_handle_dirty_metadata(handle
, NULL
, EXT4_SB(sb
)->s_sbh
);
1081 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count
,
1082 o_blocks_count
+ add
);
1083 /* We add the blocks to the bitmap and set the group need init bit */
1084 ext4_add_groupblocks(handle
, sb
, o_blocks_count
, add
);
1085 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count
,
1086 o_blocks_count
+ add
);
1087 if ((err
= ext4_journal_stop(handle
)))
1090 if (test_opt(sb
, DEBUG
))
1091 printk(KERN_DEBUG
"EXT4-fs: extended group to %llu blocks\n",
1092 ext4_blocks_count(es
));
1093 update_backups(sb
, EXT4_SB(sb
)->s_sbh
->b_blocknr
, (char *)es
,
1094 sizeof(struct ext4_super_block
));
1097 } /* ext4_group_extend */