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)
168 if (ext4_handle_has_enough_credits(handle, thresh))
171 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
175 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
184 * Set up the block and inode bitmaps, and the inode table for the new group.
185 * This doesn't need to be part of the main transaction, since we are only
186 * changing blocks outside the actual filesystem. We still do journaling to
187 * ensure the recovery is correct in case of a failure just after resize.
188 * If any part of this fails, we simply abort the resize.
190 static int setup_new_group_blocks(struct super_block *sb,
191 struct ext4_new_group_data *input)
193 struct ext4_sb_info *sbi = EXT4_SB(sb);
194 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
195 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
196 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
197 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
198 struct buffer_head *bh;
205 /* This transaction may be extended/restarted along the way */
206 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
209 return PTR_ERR(handle);
211 BUG_ON(input->group != sbi->s_groups_count);
213 /* Copy all of the GDT blocks into the backup in this group */
214 for (i = 0, bit = 1, block = start + 1;
215 i < gdblocks; i++, block++, bit++) {
216 struct buffer_head *gdb;
218 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
219 err = extend_or_restart_transaction(handle, 1);
223 gdb = sb_getblk(sb, block);
228 if ((err = ext4_journal_get_write_access(handle, gdb))) {
233 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
234 set_buffer_uptodate(gdb);
236 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
244 /* Zero out all of the reserved backup group descriptor table blocks */
245 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
246 block, sbi->s_itb_per_group);
247 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
252 err = extend_or_restart_transaction(handle, 2);
256 bh = bclean(handle, sb, input->block_bitmap);
262 if (ext4_bg_has_super(sb, input->group)) {
263 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
264 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
267 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
268 input->block_bitmap - start);
269 ext4_set_bit(input->block_bitmap - start, bh->b_data);
270 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
271 input->inode_bitmap - start);
272 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
274 /* Zero out all of the inode table blocks */
275 block = input->inode_table;
276 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
277 block, sbi->s_itb_per_group);
278 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
281 ext4_set_bits(bh->b_data, input->inode_table - start,
282 sbi->s_itb_per_group);
285 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
287 err = ext4_handle_dirty_metadata(handle, NULL, bh);
289 ext4_std_error(sb, err);
293 /* Mark unused entries in inode bitmap used */
294 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
295 input->inode_bitmap, input->inode_bitmap - start);
296 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
301 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
303 err = ext4_handle_dirty_metadata(handle, NULL, bh);
305 ext4_std_error(sb, err);
310 if ((err2 = ext4_journal_stop(handle)) && !err)
317 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
318 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
319 * calling this for the first time. In a sparse filesystem it will be the
320 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
321 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
323 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
324 unsigned *five, unsigned *seven)
326 unsigned *min = three;
330 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
331 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
353 * Check that all of the backup GDT blocks are held in the primary GDT block.
354 * It is assumed that they are stored in group order. Returns the number of
355 * groups in current filesystem that have BACKUPS, or -ve error code.
357 static int verify_reserved_gdb(struct super_block *sb,
358 struct buffer_head *primary)
360 const ext4_fsblk_t blk = primary->b_blocknr;
361 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
366 __le32 *p = (__le32 *)primary->b_data;
369 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
370 if (le32_to_cpu(*p++) !=
371 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
372 ext4_warning(sb, "reserved GDT %llu"
373 " missing grp %d (%llu)",
376 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
380 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
388 * Called when we need to bring a reserved group descriptor table block into
389 * use from the resize inode. The primary copy of the new GDT block currently
390 * is an indirect block (under the double indirect block in the resize inode).
391 * The new backup GDT blocks will be stored as leaf blocks in this indirect
392 * block, in group order. Even though we know all the block numbers we need,
393 * we check to ensure that the resize inode has actually reserved these blocks.
395 * Don't need to update the block bitmaps because the blocks are still in use.
397 * We get all of the error cases out of the way, so that we are sure to not
398 * fail once we start modifying the data on disk, because JBD has no rollback.
400 static int add_new_gdb(handle_t *handle, struct inode *inode,
401 struct ext4_new_group_data *input,
402 struct buffer_head **primary)
404 struct super_block *sb = inode->i_sb;
405 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
406 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
407 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
408 struct buffer_head **o_group_desc, **n_group_desc;
409 struct buffer_head *dind;
411 struct ext4_iloc iloc;
415 if (test_opt(sb, DEBUG))
417 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
421 * If we are not using the primary superblock/GDT copy don't resize,
422 * because the user tools have no way of handling this. Probably a
423 * bad time to do it anyways.
425 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
426 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
427 ext4_warning(sb, "won't resize using backup superblock at %llu",
428 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
432 *primary = sb_bread(sb, gdblock);
436 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
441 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
442 dind = sb_bread(sb, le32_to_cpu(*data));
448 data = (__le32 *)dind->b_data;
449 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
450 ext4_warning(sb, "new group %u GDT block %llu not reserved",
451 input->group, gdblock);
456 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
460 err = ext4_journal_get_write_access(handle, *primary);
464 err = ext4_journal_get_write_access(handle, dind);
466 ext4_std_error(sb, err);
468 /* ext4_reserve_inode_write() gets a reference on the iloc */
469 err = ext4_reserve_inode_write(handle, inode, &iloc);
473 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
478 "not enough memory for %lu groups", gdb_num + 1);
483 * Finally, we have all of the possible failures behind us...
485 * Remove new GDT block from inode double-indirect block and clear out
486 * the new GDT block for use (which also "frees" the backup GDT blocks
487 * from the reserved inode). We don't need to change the bitmaps for
488 * these blocks, because they are marked as in-use from being in the
489 * reserved inode, and will become GDT blocks (primary and backup).
491 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
492 err = ext4_handle_dirty_metadata(handle, NULL, dind);
494 ext4_std_error(sb, err);
497 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
498 ext4_mark_iloc_dirty(handle, inode, &iloc);
499 memset((*primary)->b_data, 0, sb->s_blocksize);
500 err = ext4_handle_dirty_metadata(handle, NULL, *primary);
502 ext4_std_error(sb, err);
507 o_group_desc = EXT4_SB(sb)->s_group_desc;
508 memcpy(n_group_desc, o_group_desc,
509 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
510 n_group_desc[gdb_num] = *primary;
511 EXT4_SB(sb)->s_group_desc = n_group_desc;
512 EXT4_SB(sb)->s_gdb_count++;
515 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
516 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
518 ext4_std_error(sb, err);
523 /* ext4_handle_release_buffer(handle, iloc.bh); */
526 /* ext4_handle_release_buffer(handle, dind); */
528 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
534 ext4_debug("leaving with error %d\n", err);
539 * Called when we are adding a new group which has a backup copy of each of
540 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
541 * We need to add these reserved backup GDT blocks to the resize inode, so
542 * that they are kept for future resizing and not allocated to files.
544 * Each reserved backup GDT block will go into a different indirect block.
545 * The indirect blocks are actually the primary reserved GDT blocks,
546 * so we know in advance what their block numbers are. We only get the
547 * double-indirect block to verify it is pointing to the primary reserved
548 * GDT blocks so we don't overwrite a data block by accident. The reserved
549 * backup GDT blocks are stored in their reserved primary GDT block.
551 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
552 struct ext4_new_group_data *input)
554 struct super_block *sb = inode->i_sb;
555 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
556 struct buffer_head **primary;
557 struct buffer_head *dind;
558 struct ext4_iloc iloc;
565 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
569 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
570 dind = sb_bread(sb, le32_to_cpu(*data));
576 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
577 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
578 EXT4_ADDR_PER_BLOCK(sb));
579 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
581 /* Get each reserved primary GDT block and verify it holds backups */
582 for (res = 0; res < reserved_gdb; res++, blk++) {
583 if (le32_to_cpu(*data) != blk) {
584 ext4_warning(sb, "reserved block %llu"
585 " not at offset %ld",
587 (long)(data - (__le32 *)dind->b_data));
591 primary[res] = sb_bread(sb, blk);
596 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
597 brelse(primary[res]);
602 data = (__le32 *)dind->b_data;
605 for (i = 0; i < reserved_gdb; i++) {
606 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
609 for (j = 0; j < i; j++)
610 ext4_handle_release_buffer(handle, primary[j]);
616 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
620 * Finally we can add each of the reserved backup GDT blocks from
621 * the new group to its reserved primary GDT block.
623 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
624 for (i = 0; i < reserved_gdb; i++) {
626 data = (__le32 *)primary[i]->b_data;
627 /* printk("reserving backup %lu[%u] = %lu\n",
628 primary[i]->b_blocknr, gdbackups,
629 blk + primary[i]->b_blocknr); */
630 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
631 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
635 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
636 ext4_mark_iloc_dirty(handle, inode, &iloc);
640 brelse(primary[res]);
650 * Update the backup copies of the ext4 metadata. These don't need to be part
651 * of the main resize transaction, because e2fsck will re-write them if there
652 * is a problem (basically only OOM will cause a problem). However, we
653 * _should_ update the backups if possible, in case the primary gets trashed
654 * for some reason and we need to run e2fsck from a backup superblock. The
655 * important part is that the new block and inode counts are in the backup
656 * superblocks, and the location of the new group metadata in the GDT backups.
658 * We do not need take the s_resize_lock for this, because these
659 * blocks are not otherwise touched by the filesystem code when it is
660 * mounted. We don't need to worry about last changing from
661 * sbi->s_groups_count, because the worst that can happen is that we
662 * do not copy the full number of backups at this time. The resize
663 * which changed s_groups_count will backup again.
665 static void update_backups(struct super_block *sb,
666 int blk_off, char *data, int size)
668 struct ext4_sb_info *sbi = EXT4_SB(sb);
669 const ext4_group_t last = sbi->s_groups_count;
670 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
675 int rest = sb->s_blocksize - size;
679 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
680 if (IS_ERR(handle)) {
682 err = PTR_ERR(handle);
686 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
687 struct buffer_head *bh;
689 /* Out of journal space, and can't get more - abort - so sad */
690 if (ext4_handle_valid(handle) &&
691 handle->h_buffer_credits == 0 &&
692 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
693 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
696 bh = sb_getblk(sb, group * bpg + blk_off);
701 ext4_debug("update metadata backup %#04lx\n",
702 (unsigned long)bh->b_blocknr);
703 if ((err = ext4_journal_get_write_access(handle, bh)))
706 memcpy(bh->b_data, data, size);
708 memset(bh->b_data + size, 0, rest);
709 set_buffer_uptodate(bh);
711 err = ext4_handle_dirty_metadata(handle, NULL, bh);
713 ext4_std_error(sb, err);
716 if ((err2 = ext4_journal_stop(handle)) && !err)
720 * Ugh! Need to have e2fsck write the backup copies. It is too
721 * late to revert the resize, we shouldn't fail just because of
722 * the backup copies (they are only needed in case of corruption).
724 * However, if we got here we have a journal problem too, so we
725 * can't really start a transaction to mark the superblock.
726 * Chicken out and just set the flag on the hope it will be written
727 * to disk, and if not - we will simply wait until next fsck.
731 ext4_warning(sb, "can't update backup for group %u (err %d), "
732 "forcing fsck on next reboot", group, err);
733 sbi->s_mount_state &= ~EXT4_VALID_FS;
734 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
735 mark_buffer_dirty(sbi->s_sbh);
739 /* Add group descriptor data to an existing or new group descriptor block.
740 * Ensure we handle all possible error conditions _before_ we start modifying
741 * the filesystem, because we cannot abort the transaction and not have it
742 * write the data to disk.
744 * If we are on a GDT block boundary, we need to get the reserved GDT block.
745 * Otherwise, we may need to add backup GDT blocks for a sparse group.
747 * We only need to hold the superblock lock while we are actually adding
748 * in the new group's counts to the superblock. Prior to that we have
749 * not really "added" the group at all. We re-check that we are still
750 * adding in the last group in case things have changed since verifying.
752 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
754 struct ext4_sb_info *sbi = EXT4_SB(sb);
755 struct ext4_super_block *es = sbi->s_es;
756 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
757 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
758 struct buffer_head *primary = NULL;
759 struct ext4_group_desc *gdp;
760 struct inode *inode = NULL;
762 int gdb_off, gdb_num;
765 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
766 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
768 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
769 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
770 ext4_warning(sb, "Can't resize non-sparse filesystem further");
774 if (ext4_blocks_count(es) + input->blocks_count <
775 ext4_blocks_count(es)) {
776 ext4_warning(sb, "blocks_count overflow");
780 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
781 le32_to_cpu(es->s_inodes_count)) {
782 ext4_warning(sb, "inodes_count overflow");
786 if (reserved_gdb || gdb_off == 0) {
787 if (!EXT4_HAS_COMPAT_FEATURE(sb,
788 EXT4_FEATURE_COMPAT_RESIZE_INODE)
789 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
791 "No reserved GDT blocks, can't resize");
794 inode = ext4_iget(sb, EXT4_RESIZE_INO);
796 ext4_warning(sb, "Error opening resize inode");
797 return PTR_ERR(inode);
802 if ((err = verify_group_input(sb, input)))
805 if ((err = setup_new_group_blocks(sb, input)))
809 * We will always be modifying at least the superblock and a GDT
810 * block. If we are adding a group past the last current GDT block,
811 * we will also modify the inode and the dindirect block. If we
812 * are adding a group with superblock/GDT backups we will also
813 * modify each of the reserved GDT dindirect blocks.
815 handle = ext4_journal_start_sb(sb,
816 ext4_bg_has_super(sb, input->group) ?
817 3 + reserved_gdb : 4);
818 if (IS_ERR(handle)) {
819 err = PTR_ERR(handle);
823 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
827 * We will only either add reserved group blocks to a backup group
828 * or remove reserved blocks for the first group in a new group block.
829 * Doing both would be mean more complex code, and sane people don't
830 * use non-sparse filesystems anymore. This is already checked above.
833 primary = sbi->s_group_desc[gdb_num];
834 if ((err = ext4_journal_get_write_access(handle, primary)))
837 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
838 (err = reserve_backup_gdb(handle, inode, input)))
840 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
844 * OK, now we've set up the new group. Time to make it active.
846 * so we have to be safe wrt. concurrent accesses the group
847 * data. So we need to be careful to set all of the relevant
848 * group descriptor data etc. *before* we enable the group.
850 * The key field here is sbi->s_groups_count: as long as
851 * that retains its old value, nobody is going to access the new
854 * So first we update all the descriptor metadata for the new
855 * group; then we update the total disk blocks count; then we
856 * update the groups count to enable the group; then finally we
857 * update the free space counts so that the system can start
858 * using the new disk blocks.
861 /* Update group descriptor block for new group */
862 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
863 gdb_off * EXT4_DESC_SIZE(sb));
865 memset(gdp, 0, EXT4_DESC_SIZE(sb));
866 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
867 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
868 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
869 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
870 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
871 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
872 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
875 * We can allocate memory for mb_alloc based on the new group
878 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
883 * Make the new blocks and inodes valid next. We do this before
884 * increasing the group count so that once the group is enabled,
885 * all of its blocks and inodes are already valid.
887 * We always allocate group-by-group, then block-by-block or
888 * inode-by-inode within a group, so enabling these
889 * blocks/inodes before the group is live won't actually let us
890 * allocate the new space yet.
892 ext4_blocks_count_set(es, ext4_blocks_count(es) +
893 input->blocks_count);
894 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
897 * We need to protect s_groups_count against other CPUs seeing
898 * inconsistent state in the superblock.
900 * The precise rules we use are:
902 * * Writers must perform a smp_wmb() after updating all dependent
903 * data and before modifying the groups count
905 * * Readers must perform an smp_rmb() after reading the groups count
906 * and before reading any dependent data.
908 * NB. These rules can be relaxed when checking the group count
909 * while freeing data, as we can only allocate from a block
910 * group after serialising against the group count, and we can
911 * only then free after serialising in turn against that
916 /* Update the global fs size fields */
917 sbi->s_groups_count++;
919 err = ext4_handle_dirty_metadata(handle, NULL, primary);
921 ext4_std_error(sb, err);
925 /* Update the reserved block counts only once the new group is
927 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
928 input->reserved_blocks);
930 /* Update the free space counts */
931 percpu_counter_add(&sbi->s_freeblocks_counter,
932 input->free_blocks_count);
933 percpu_counter_add(&sbi->s_freeinodes_counter,
934 EXT4_INODES_PER_GROUP(sb));
936 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
937 sbi->s_log_groups_per_flex) {
938 ext4_group_t flex_group;
939 flex_group = ext4_flex_group(sbi, input->group);
940 atomic_add(input->free_blocks_count,
941 &sbi->s_flex_groups[flex_group].free_blocks);
942 atomic_add(EXT4_INODES_PER_GROUP(sb),
943 &sbi->s_flex_groups[flex_group].free_inodes);
946 ext4_handle_dirty_super(handle, sb);
949 if ((err2 = ext4_journal_stop(handle)) && !err)
952 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
953 sizeof(struct ext4_super_block));
954 update_backups(sb, primary->b_blocknr, primary->b_data,
960 } /* ext4_group_add */
963 * Extend the filesystem to the new number of blocks specified. This entry
964 * point is only used to extend the current filesystem to the end of the last
965 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
966 * for emergencies (because it has no dependencies on reserved blocks).
968 * If we _really_ wanted, we could use default values to call ext4_group_add()
969 * allow the "remount" trick to work for arbitrary resizing, assuming enough
970 * GDT blocks are reserved to grow to the desired size.
972 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
973 ext4_fsblk_t n_blocks_count)
975 ext4_fsblk_t o_blocks_count;
978 struct buffer_head *bh;
983 o_blocks_count = ext4_blocks_count(es);
985 if (test_opt(sb, DEBUG))
986 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
987 o_blocks_count, n_blocks_count);
989 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
992 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
993 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
994 " too large to resize to %llu blocks safely\n",
995 sb->s_id, n_blocks_count);
996 if (sizeof(sector_t) < 8)
997 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1001 if (n_blocks_count < o_blocks_count) {
1002 ext4_warning(sb, "can't shrink FS - resize aborted");
1006 /* Handle the remaining blocks in the last group only. */
1007 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1010 ext4_warning(sb, "need to use ext2online to resize further");
1014 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1016 if (o_blocks_count + add < o_blocks_count) {
1017 ext4_warning(sb, "blocks_count overflow");
1021 if (o_blocks_count + add > n_blocks_count)
1022 add = n_blocks_count - o_blocks_count;
1024 if (o_blocks_count + add < n_blocks_count)
1025 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1026 o_blocks_count + add, add);
1028 /* See if the device is actually as big as what was requested */
1029 bh = sb_bread(sb, o_blocks_count + add - 1);
1031 ext4_warning(sb, "can't read last block, resize aborted");
1036 /* We will update the superblock, one block bitmap, and
1037 * one group descriptor via ext4_free_blocks().
1039 handle = ext4_journal_start_sb(sb, 3);
1040 if (IS_ERR(handle)) {
1041 err = PTR_ERR(handle);
1042 ext4_warning(sb, "error %d on journal start", err);
1046 if ((err = ext4_journal_get_write_access(handle,
1047 EXT4_SB(sb)->s_sbh))) {
1048 ext4_warning(sb, "error %d on journal write access", err);
1049 ext4_journal_stop(handle);
1052 ext4_blocks_count_set(es, o_blocks_count + add);
1053 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1054 o_blocks_count + add);
1055 /* We add the blocks to the bitmap and set the group need init bit */
1056 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1057 ext4_handle_dirty_super(handle, sb);
1058 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1059 o_blocks_count + add);
1060 err2 = ext4_journal_stop(handle);
1067 if (test_opt(sb, DEBUG))
1068 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1069 ext4_blocks_count(es));
1070 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1071 sizeof(struct ext4_super_block));
1074 } /* ext4_group_extend */