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))) {
151 memset(bh->b_data, 0, sb->s_blocksize);
152 set_buffer_uptodate(bh);
160 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
161 * If that fails, restart the transaction & regain write access for the
162 * buffer head which is used for block_bitmap modifications.
164 static int extend_or_restart_transaction(handle_t *handle, int thresh,
165 struct buffer_head *bh)
169 if (ext4_handle_has_enough_credits(handle, thresh))
172 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
176 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
178 if ((err = ext4_journal_get_write_access(handle, bh)))
186 * Set up the block and inode bitmaps, and the inode table for the new group.
187 * This doesn't need to be part of the main transaction, since we are only
188 * changing blocks outside the actual filesystem. We still do journaling to
189 * ensure the recovery is correct in case of a failure just after resize.
190 * If any part of this fails, we simply abort the resize.
192 static int setup_new_group_blocks(struct super_block *sb,
193 struct ext4_new_group_data *input)
195 struct ext4_sb_info *sbi = EXT4_SB(sb);
196 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
197 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
198 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
199 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
200 struct buffer_head *bh;
207 /* This transaction may be extended/restarted along the way */
208 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
211 return PTR_ERR(handle);
213 BUG_ON(input->group != sbi->s_groups_count);
215 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
220 /* Copy all of the GDT blocks into the backup in this group */
221 for (i = 0, bit = 1, block = start + 1;
222 i < gdblocks; i++, block++, bit++) {
223 struct buffer_head *gdb;
225 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
227 if ((err = extend_or_restart_transaction(handle, 1, bh)))
230 gdb = sb_getblk(sb, block);
235 if ((err = ext4_journal_get_write_access(handle, gdb))) {
240 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
241 set_buffer_uptodate(gdb);
243 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
251 /* Zero out all of the reserved backup group descriptor table blocks */
252 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
253 block, sbi->s_itb_per_group);
254 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
259 if (ext4_bg_has_super(sb, input->group)) {
260 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
261 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
264 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
265 input->block_bitmap - start);
266 ext4_set_bit(input->block_bitmap - start, bh->b_data);
267 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
268 input->inode_bitmap - start);
269 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
271 /* Zero out all of the inode table blocks */
272 block = input->inode_table;
273 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
274 block, sbi->s_itb_per_group);
275 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
278 ext4_set_bits(bh->b_data, input->inode_table - start,
279 sbi->s_itb_per_group);
281 if ((err = extend_or_restart_transaction(handle, 2, bh)))
284 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
286 err = ext4_handle_dirty_metadata(handle, NULL, bh);
288 ext4_std_error(sb, err);
292 /* Mark unused entries in inode bitmap used */
293 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
294 input->inode_bitmap, input->inode_bitmap - start);
295 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
300 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
302 err = ext4_handle_dirty_metadata(handle, NULL, bh);
304 ext4_std_error(sb, err);
309 if ((err2 = ext4_journal_stop(handle)) && !err)
316 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
317 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
318 * calling this for the first time. In a sparse filesystem it will be the
319 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
320 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
322 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
323 unsigned *five, unsigned *seven)
325 unsigned *min = three;
329 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
330 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
352 * Check that all of the backup GDT blocks are held in the primary GDT block.
353 * It is assumed that they are stored in group order. Returns the number of
354 * groups in current filesystem that have BACKUPS, or -ve error code.
356 static int verify_reserved_gdb(struct super_block *sb,
357 struct buffer_head *primary)
359 const ext4_fsblk_t blk = primary->b_blocknr;
360 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
365 __le32 *p = (__le32 *)primary->b_data;
368 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
369 if (le32_to_cpu(*p++) !=
370 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
371 ext4_warning(sb, "reserved GDT %llu"
372 " missing grp %d (%llu)",
375 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
379 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
387 * Called when we need to bring a reserved group descriptor table block into
388 * use from the resize inode. The primary copy of the new GDT block currently
389 * is an indirect block (under the double indirect block in the resize inode).
390 * The new backup GDT blocks will be stored as leaf blocks in this indirect
391 * block, in group order. Even though we know all the block numbers we need,
392 * we check to ensure that the resize inode has actually reserved these blocks.
394 * Don't need to update the block bitmaps because the blocks are still in use.
396 * We get all of the error cases out of the way, so that we are sure to not
397 * fail once we start modifying the data on disk, because JBD has no rollback.
399 static int add_new_gdb(handle_t *handle, struct inode *inode,
400 struct ext4_new_group_data *input,
401 struct buffer_head **primary)
403 struct super_block *sb = inode->i_sb;
404 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
405 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
406 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
407 struct buffer_head **o_group_desc, **n_group_desc;
408 struct buffer_head *dind;
410 struct ext4_iloc iloc;
414 if (test_opt(sb, DEBUG))
416 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
420 * If we are not using the primary superblock/GDT copy don't resize,
421 * because the user tools have no way of handling this. Probably a
422 * bad time to do it anyways.
424 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
425 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
426 ext4_warning(sb, "won't resize using backup superblock at %llu",
427 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
431 *primary = sb_bread(sb, gdblock);
435 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
440 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
441 dind = sb_bread(sb, le32_to_cpu(*data));
447 data = (__le32 *)dind->b_data;
448 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
449 ext4_warning(sb, "new group %u GDT block %llu not reserved",
450 input->group, gdblock);
455 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
459 err = ext4_journal_get_write_access(handle, *primary);
463 err = ext4_journal_get_write_access(handle, dind);
465 ext4_std_error(sb, err);
467 /* ext4_reserve_inode_write() gets a reference on the iloc */
468 err = ext4_reserve_inode_write(handle, inode, &iloc);
472 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
477 "not enough memory for %lu groups", gdb_num + 1);
482 * Finally, we have all of the possible failures behind us...
484 * Remove new GDT block from inode double-indirect block and clear out
485 * the new GDT block for use (which also "frees" the backup GDT blocks
486 * from the reserved inode). We don't need to change the bitmaps for
487 * these blocks, because they are marked as in-use from being in the
488 * reserved inode, and will become GDT blocks (primary and backup).
490 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
491 err = ext4_handle_dirty_metadata(handle, NULL, dind);
493 ext4_std_error(sb, err);
496 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
497 ext4_mark_iloc_dirty(handle, inode, &iloc);
498 memset((*primary)->b_data, 0, sb->s_blocksize);
499 err = ext4_handle_dirty_metadata(handle, NULL, *primary);
501 ext4_std_error(sb, err);
506 o_group_desc = EXT4_SB(sb)->s_group_desc;
507 memcpy(n_group_desc, o_group_desc,
508 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
509 n_group_desc[gdb_num] = *primary;
510 EXT4_SB(sb)->s_group_desc = n_group_desc;
511 EXT4_SB(sb)->s_gdb_count++;
514 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
515 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
517 ext4_std_error(sb, err);
522 /* ext4_handle_release_buffer(handle, iloc.bh); */
525 /* ext4_handle_release_buffer(handle, dind); */
527 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
533 ext4_debug("leaving with error %d\n", err);
538 * Called when we are adding a new group which has a backup copy of each of
539 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
540 * We need to add these reserved backup GDT blocks to the resize inode, so
541 * that they are kept for future resizing and not allocated to files.
543 * Each reserved backup GDT block will go into a different indirect block.
544 * The indirect blocks are actually the primary reserved GDT blocks,
545 * so we know in advance what their block numbers are. We only get the
546 * double-indirect block to verify it is pointing to the primary reserved
547 * GDT blocks so we don't overwrite a data block by accident. The reserved
548 * backup GDT blocks are stored in their reserved primary GDT block.
550 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
551 struct ext4_new_group_data *input)
553 struct super_block *sb = inode->i_sb;
554 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
555 struct buffer_head **primary;
556 struct buffer_head *dind;
557 struct ext4_iloc iloc;
564 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
568 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
569 dind = sb_bread(sb, le32_to_cpu(*data));
575 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
576 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
577 EXT4_ADDR_PER_BLOCK(sb));
578 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
580 /* Get each reserved primary GDT block and verify it holds backups */
581 for (res = 0; res < reserved_gdb; res++, blk++) {
582 if (le32_to_cpu(*data) != blk) {
583 ext4_warning(sb, "reserved block %llu"
584 " not at offset %ld",
586 (long)(data - (__le32 *)dind->b_data));
590 primary[res] = sb_bread(sb, blk);
595 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
596 brelse(primary[res]);
601 data = (__le32 *)dind->b_data;
604 for (i = 0; i < reserved_gdb; i++) {
605 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
608 for (j = 0; j < i; j++)
609 ext4_handle_release_buffer(handle, primary[j]);
615 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
619 * Finally we can add each of the reserved backup GDT blocks from
620 * the new group to its reserved primary GDT block.
622 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
623 for (i = 0; i < reserved_gdb; i++) {
625 data = (__le32 *)primary[i]->b_data;
626 /* printk("reserving backup %lu[%u] = %lu\n",
627 primary[i]->b_blocknr, gdbackups,
628 blk + primary[i]->b_blocknr); */
629 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
630 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
634 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
635 ext4_mark_iloc_dirty(handle, inode, &iloc);
639 brelse(primary[res]);
649 * Update the backup copies of the ext4 metadata. These don't need to be part
650 * of the main resize transaction, because e2fsck will re-write them if there
651 * is a problem (basically only OOM will cause a problem). However, we
652 * _should_ update the backups if possible, in case the primary gets trashed
653 * for some reason and we need to run e2fsck from a backup superblock. The
654 * important part is that the new block and inode counts are in the backup
655 * superblocks, and the location of the new group metadata in the GDT backups.
657 * We do not need take the s_resize_lock for this, because these
658 * blocks are not otherwise touched by the filesystem code when it is
659 * mounted. We don't need to worry about last changing from
660 * sbi->s_groups_count, because the worst that can happen is that we
661 * do not copy the full number of backups at this time. The resize
662 * which changed s_groups_count will backup again.
664 static void update_backups(struct super_block *sb,
665 int blk_off, char *data, int size)
667 struct ext4_sb_info *sbi = EXT4_SB(sb);
668 const ext4_group_t last = sbi->s_groups_count;
669 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
674 int rest = sb->s_blocksize - size;
678 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
679 if (IS_ERR(handle)) {
681 err = PTR_ERR(handle);
685 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
686 struct buffer_head *bh;
688 /* Out of journal space, and can't get more - abort - so sad */
689 if (ext4_handle_valid(handle) &&
690 handle->h_buffer_credits == 0 &&
691 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
692 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
695 bh = sb_getblk(sb, group * bpg + blk_off);
700 ext4_debug("update metadata backup %#04lx\n",
701 (unsigned long)bh->b_blocknr);
702 if ((err = ext4_journal_get_write_access(handle, bh)))
705 memcpy(bh->b_data, data, size);
707 memset(bh->b_data + size, 0, rest);
708 set_buffer_uptodate(bh);
710 err = ext4_handle_dirty_metadata(handle, NULL, bh);
712 ext4_std_error(sb, err);
715 if ((err2 = ext4_journal_stop(handle)) && !err)
719 * Ugh! Need to have e2fsck write the backup copies. It is too
720 * late to revert the resize, we shouldn't fail just because of
721 * the backup copies (they are only needed in case of corruption).
723 * However, if we got here we have a journal problem too, so we
724 * can't really start a transaction to mark the superblock.
725 * Chicken out and just set the flag on the hope it will be written
726 * to disk, and if not - we will simply wait until next fsck.
730 ext4_warning(sb, "can't update backup for group %u (err %d), "
731 "forcing fsck on next reboot", group, err);
732 sbi->s_mount_state &= ~EXT4_VALID_FS;
733 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
734 mark_buffer_dirty(sbi->s_sbh);
738 /* Add group descriptor data to an existing or new group descriptor block.
739 * Ensure we handle all possible error conditions _before_ we start modifying
740 * the filesystem, because we cannot abort the transaction and not have it
741 * write the data to disk.
743 * If we are on a GDT block boundary, we need to get the reserved GDT block.
744 * Otherwise, we may need to add backup GDT blocks for a sparse group.
746 * We only need to hold the superblock lock while we are actually adding
747 * in the new group's counts to the superblock. Prior to that we have
748 * not really "added" the group at all. We re-check that we are still
749 * adding in the last group in case things have changed since verifying.
751 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
753 struct ext4_sb_info *sbi = EXT4_SB(sb);
754 struct ext4_super_block *es = sbi->s_es;
755 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
756 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
757 struct buffer_head *primary = NULL;
758 struct ext4_group_desc *gdp;
759 struct inode *inode = NULL;
761 int gdb_off, gdb_num;
764 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
765 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
767 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
768 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
769 ext4_warning(sb, "Can't resize non-sparse filesystem further");
773 if (ext4_blocks_count(es) + input->blocks_count <
774 ext4_blocks_count(es)) {
775 ext4_warning(sb, "blocks_count overflow");
779 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
780 le32_to_cpu(es->s_inodes_count)) {
781 ext4_warning(sb, "inodes_count overflow");
785 if (reserved_gdb || gdb_off == 0) {
786 if (!EXT4_HAS_COMPAT_FEATURE(sb,
787 EXT4_FEATURE_COMPAT_RESIZE_INODE)
788 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
790 "No reserved GDT blocks, can't resize");
793 inode = ext4_iget(sb, EXT4_RESIZE_INO);
795 ext4_warning(sb, "Error opening resize inode");
796 return PTR_ERR(inode);
801 if ((err = verify_group_input(sb, input)))
804 if ((err = setup_new_group_blocks(sb, input)))
808 * We will always be modifying at least the superblock and a GDT
809 * block. If we are adding a group past the last current GDT block,
810 * we will also modify the inode and the dindirect block. If we
811 * are adding a group with superblock/GDT backups we will also
812 * modify each of the reserved GDT dindirect blocks.
814 handle = ext4_journal_start_sb(sb,
815 ext4_bg_has_super(sb, input->group) ?
816 3 + reserved_gdb : 4);
817 if (IS_ERR(handle)) {
818 err = PTR_ERR(handle);
822 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
826 * We will only either add reserved group blocks to a backup group
827 * or remove reserved blocks for the first group in a new group block.
828 * Doing both would be mean more complex code, and sane people don't
829 * use non-sparse filesystems anymore. This is already checked above.
832 primary = sbi->s_group_desc[gdb_num];
833 if ((err = ext4_journal_get_write_access(handle, primary)))
836 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
837 (err = reserve_backup_gdb(handle, inode, input)))
839 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
843 * OK, now we've set up the new group. Time to make it active.
845 * so we have to be safe wrt. concurrent accesses the group
846 * data. So we need to be careful to set all of the relevant
847 * group descriptor data etc. *before* we enable the group.
849 * The key field here is sbi->s_groups_count: as long as
850 * that retains its old value, nobody is going to access the new
853 * So first we update all the descriptor metadata for the new
854 * group; then we update the total disk blocks count; then we
855 * update the groups count to enable the group; then finally we
856 * update the free space counts so that the system can start
857 * using the new disk blocks.
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);
882 * Make the new blocks and inodes valid next. We do this before
883 * increasing the group count so that once the group is enabled,
884 * all of its blocks and inodes are already valid.
886 * We always allocate group-by-group, then block-by-block or
887 * inode-by-inode within a group, so enabling these
888 * blocks/inodes before the group is live won't actually let us
889 * allocate the new space yet.
891 ext4_blocks_count_set(es, ext4_blocks_count(es) +
892 input->blocks_count);
893 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
896 * We need to protect s_groups_count against other CPUs seeing
897 * inconsistent state in the superblock.
899 * The precise rules we use are:
901 * * Writers must perform a smp_wmb() after updating all dependent
902 * data and before modifying the groups count
904 * * Readers must perform an smp_rmb() after reading the groups count
905 * and before reading any dependent data.
907 * NB. These rules can be relaxed when checking the group count
908 * while freeing data, as we can only allocate from a block
909 * group after serialising against the group count, and we can
910 * only then free after serialising in turn against that
915 /* Update the global fs size fields */
916 sbi->s_groups_count++;
918 err = ext4_handle_dirty_metadata(handle, NULL, primary);
920 ext4_std_error(sb, err);
924 /* Update the reserved block counts only once the new group is
926 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
927 input->reserved_blocks);
929 /* Update the free space counts */
930 percpu_counter_add(&sbi->s_freeblocks_counter,
931 input->free_blocks_count);
932 percpu_counter_add(&sbi->s_freeinodes_counter,
933 EXT4_INODES_PER_GROUP(sb));
935 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
936 sbi->s_log_groups_per_flex) {
937 ext4_group_t flex_group;
938 flex_group = ext4_flex_group(sbi, input->group);
939 atomic_add(input->free_blocks_count,
940 &sbi->s_flex_groups[flex_group].free_blocks);
941 atomic_add(EXT4_INODES_PER_GROUP(sb),
942 &sbi->s_flex_groups[flex_group].free_inodes);
945 ext4_handle_dirty_super(handle, sb);
948 if ((err2 = ext4_journal_stop(handle)) && !err)
951 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
952 sizeof(struct ext4_super_block));
953 update_backups(sb, primary->b_blocknr, primary->b_data,
959 } /* ext4_group_add */
962 * Extend the filesystem to the new number of blocks specified. This entry
963 * point is only used to extend the current filesystem to the end of the last
964 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
965 * for emergencies (because it has no dependencies on reserved blocks).
967 * If we _really_ wanted, we could use default values to call ext4_group_add()
968 * allow the "remount" trick to work for arbitrary resizing, assuming enough
969 * GDT blocks are reserved to grow to the desired size.
971 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
972 ext4_fsblk_t n_blocks_count)
974 ext4_fsblk_t o_blocks_count;
977 struct buffer_head *bh;
982 o_blocks_count = ext4_blocks_count(es);
984 if (test_opt(sb, DEBUG))
985 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
986 o_blocks_count, n_blocks_count);
988 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
991 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
992 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
993 " too large to resize to %llu blocks safely\n",
994 sb->s_id, n_blocks_count);
995 if (sizeof(sector_t) < 8)
996 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1000 if (n_blocks_count < o_blocks_count) {
1001 ext4_warning(sb, "can't shrink FS - resize aborted");
1005 /* Handle the remaining blocks in the last group only. */
1006 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1009 ext4_warning(sb, "need to use ext2online to resize further");
1013 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1015 if (o_blocks_count + add < o_blocks_count) {
1016 ext4_warning(sb, "blocks_count overflow");
1020 if (o_blocks_count + add > n_blocks_count)
1021 add = n_blocks_count - o_blocks_count;
1023 if (o_blocks_count + add < n_blocks_count)
1024 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1025 o_blocks_count + add, add);
1027 /* See if the device is actually as big as what was requested */
1028 bh = sb_bread(sb, o_blocks_count + add - 1);
1030 ext4_warning(sb, "can't read last block, resize aborted");
1035 /* We will update the superblock, one block bitmap, and
1036 * one group descriptor via ext4_free_blocks().
1038 handle = ext4_journal_start_sb(sb, 3);
1039 if (IS_ERR(handle)) {
1040 err = PTR_ERR(handle);
1041 ext4_warning(sb, "error %d on journal start", err);
1045 if ((err = ext4_journal_get_write_access(handle,
1046 EXT4_SB(sb)->s_sbh))) {
1047 ext4_warning(sb, "error %d on journal write access", err);
1048 ext4_journal_stop(handle);
1051 ext4_blocks_count_set(es, o_blocks_count + add);
1052 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1053 o_blocks_count + add);
1054 /* We add the blocks to the bitmap and set the group need init bit */
1055 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1056 ext4_handle_dirty_super(handle, sb);
1057 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1058 o_blocks_count + add);
1059 err2 = ext4_journal_stop(handle);
1066 if (test_opt(sb, DEBUG))
1067 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1068 ext4_blocks_count(es));
1069 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1070 sizeof(struct ext4_super_block));
1073 } /* ext4_group_extend */