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 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
20 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
22 static int verify_group_input(struct super_block *sb,
23 struct ext4_new_group_data *input)
25 struct ext4_sb_info *sbi = EXT4_SB(sb);
26 struct ext4_super_block *es = sbi->s_es;
27 ext4_fsblk_t start = ext4_blocks_count(es);
28 ext4_fsblk_t end = start + input->blocks_count;
29 ext4_group_t group = input->group;
30 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
31 unsigned overhead = ext4_bg_has_super(sb, group) ?
32 (1 + ext4_bg_num_gdb(sb, group) +
33 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
34 ext4_fsblk_t metaend = start + overhead;
35 struct buffer_head *bh = NULL;
36 ext4_grpblk_t free_blocks_count, offset;
39 input->free_blocks_count = free_blocks_count =
40 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
42 if (test_opt(sb, DEBUG))
43 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
44 "(%d free, %u reserved)\n",
45 ext4_bg_has_super(sb, input->group) ? "normal" :
46 "no-super", input->group, input->blocks_count,
47 free_blocks_count, input->reserved_blocks);
49 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
50 if (group != sbi->s_groups_count)
51 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
52 input->group, sbi->s_groups_count);
54 ext4_warning(sb, "Last group not full");
55 else if (input->reserved_blocks > input->blocks_count / 5)
56 ext4_warning(sb, "Reserved blocks too high (%u)",
57 input->reserved_blocks);
58 else if (free_blocks_count < 0)
59 ext4_warning(sb, "Bad blocks count %u",
61 else if (!(bh = sb_bread(sb, end - 1)))
62 ext4_warning(sb, "Cannot read last block (%llu)",
64 else if (outside(input->block_bitmap, start, end))
65 ext4_warning(sb, "Block bitmap not in group (block %llu)",
66 (unsigned long long)input->block_bitmap);
67 else if (outside(input->inode_bitmap, start, end))
68 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
69 (unsigned long long)input->inode_bitmap);
70 else if (outside(input->inode_table, start, end) ||
71 outside(itend - 1, start, end))
72 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
73 (unsigned long long)input->inode_table, itend - 1);
74 else if (input->inode_bitmap == input->block_bitmap)
75 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
76 (unsigned long long)input->block_bitmap);
77 else if (inside(input->block_bitmap, input->inode_table, itend))
78 ext4_warning(sb, "Block bitmap (%llu) in inode table "
80 (unsigned long long)input->block_bitmap,
81 (unsigned long long)input->inode_table, itend - 1);
82 else if (inside(input->inode_bitmap, input->inode_table, itend))
83 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
85 (unsigned long long)input->inode_bitmap,
86 (unsigned long long)input->inode_table, itend - 1);
87 else if (inside(input->block_bitmap, start, metaend))
88 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
89 (unsigned long long)input->block_bitmap,
91 else if (inside(input->inode_bitmap, start, metaend))
92 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
93 (unsigned long long)input->inode_bitmap,
95 else if (inside(input->inode_table, start, metaend) ||
96 inside(itend - 1, start, metaend))
97 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
99 (unsigned long long)input->inode_table,
100 itend - 1, start, metaend - 1);
108 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
111 struct buffer_head *bh;
114 bh = sb_getblk(sb, blk);
116 return ERR_PTR(-EIO);
117 if ((err = ext4_journal_get_write_access(handle, bh))) {
122 memset(bh->b_data, 0, sb->s_blocksize);
123 set_buffer_uptodate(bh);
131 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
132 * If that fails, restart the transaction & regain write access for the
133 * buffer head which is used for block_bitmap modifications.
135 static int extend_or_restart_transaction(handle_t *handle, int thresh,
136 struct buffer_head *bh)
140 if (ext4_handle_has_enough_credits(handle, thresh))
143 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
147 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
149 if ((err = ext4_journal_get_write_access(handle, bh)))
157 * Set up the block and inode bitmaps, and the inode table for the new group.
158 * This doesn't need to be part of the main transaction, since we are only
159 * changing blocks outside the actual filesystem. We still do journaling to
160 * ensure the recovery is correct in case of a failure just after resize.
161 * If any part of this fails, we simply abort the resize.
163 static int setup_new_group_blocks(struct super_block *sb,
164 struct ext4_new_group_data *input)
166 struct ext4_sb_info *sbi = EXT4_SB(sb);
167 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
168 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
169 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
170 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
171 struct buffer_head *bh;
178 /* This transaction may be extended/restarted along the way */
179 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
182 return PTR_ERR(handle);
184 mutex_lock(&sbi->s_resize_lock);
185 if (input->group != sbi->s_groups_count) {
190 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
195 if (ext4_bg_has_super(sb, input->group)) {
196 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
197 ext4_set_bit(0, bh->b_data);
200 /* Copy all of the GDT blocks into the backup in this group */
201 for (i = 0, bit = 1, block = start + 1;
202 i < gdblocks; i++, block++, bit++) {
203 struct buffer_head *gdb;
205 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
207 if ((err = extend_or_restart_transaction(handle, 1, bh)))
210 gdb = sb_getblk(sb, block);
215 if ((err = ext4_journal_get_write_access(handle, gdb))) {
220 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
221 set_buffer_uptodate(gdb);
223 ext4_handle_dirty_metadata(handle, NULL, gdb);
224 ext4_set_bit(bit, bh->b_data);
228 /* Zero out all of the reserved backup group descriptor table blocks */
229 ext4_debug("clear inode table blocks %#04llx -> %#04llx\n",
230 block, sbi->s_itb_per_group);
231 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
232 GFP_NOFS, BLKDEV_IFL_WAIT);
236 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
237 input->block_bitmap - start);
238 ext4_set_bit(input->block_bitmap - start, bh->b_data);
239 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
240 input->inode_bitmap - start);
241 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
243 /* Zero out all of the inode table blocks */
244 block = input->inode_table;
245 ext4_debug("clear inode table blocks %#04llx -> %#04llx\n",
246 block, sbi->s_itb_per_group);
247 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
248 GFP_NOFS, BLKDEV_IFL_WAIT);
252 if ((err = extend_or_restart_transaction(handle, 2, bh)))
255 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
257 ext4_handle_dirty_metadata(handle, NULL, bh);
259 /* Mark unused entries in inode bitmap used */
260 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
261 input->inode_bitmap, input->inode_bitmap - start);
262 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
267 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
269 ext4_handle_dirty_metadata(handle, NULL, bh);
274 mutex_unlock(&sbi->s_resize_lock);
275 if ((err2 = ext4_journal_stop(handle)) && !err)
282 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
283 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
284 * calling this for the first time. In a sparse filesystem it will be the
285 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
286 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
288 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
289 unsigned *five, unsigned *seven)
291 unsigned *min = three;
295 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
296 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
318 * Check that all of the backup GDT blocks are held in the primary GDT block.
319 * It is assumed that they are stored in group order. Returns the number of
320 * groups in current filesystem that have BACKUPS, or -ve error code.
322 static int verify_reserved_gdb(struct super_block *sb,
323 struct buffer_head *primary)
325 const ext4_fsblk_t blk = primary->b_blocknr;
326 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
331 __le32 *p = (__le32 *)primary->b_data;
334 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
335 if (le32_to_cpu(*p++) !=
336 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
337 ext4_warning(sb, "reserved GDT %llu"
338 " missing grp %d (%llu)",
341 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
345 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
353 * Called when we need to bring a reserved group descriptor table block into
354 * use from the resize inode. The primary copy of the new GDT block currently
355 * is an indirect block (under the double indirect block in the resize inode).
356 * The new backup GDT blocks will be stored as leaf blocks in this indirect
357 * block, in group order. Even though we know all the block numbers we need,
358 * we check to ensure that the resize inode has actually reserved these blocks.
360 * Don't need to update the block bitmaps because the blocks are still in use.
362 * We get all of the error cases out of the way, so that we are sure to not
363 * fail once we start modifying the data on disk, because JBD has no rollback.
365 static int add_new_gdb(handle_t *handle, struct inode *inode,
366 struct ext4_new_group_data *input,
367 struct buffer_head **primary)
369 struct super_block *sb = inode->i_sb;
370 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
371 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
372 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
373 struct buffer_head **o_group_desc, **n_group_desc;
374 struct buffer_head *dind;
376 struct ext4_iloc iloc;
380 if (test_opt(sb, DEBUG))
382 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
386 * If we are not using the primary superblock/GDT copy don't resize,
387 * because the user tools have no way of handling this. Probably a
388 * bad time to do it anyways.
390 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
391 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
392 ext4_warning(sb, "won't resize using backup superblock at %llu",
393 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
397 *primary = sb_bread(sb, gdblock);
401 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
406 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
407 dind = sb_bread(sb, le32_to_cpu(*data));
413 data = (__le32 *)dind->b_data;
414 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
415 ext4_warning(sb, "new group %u GDT block %llu not reserved",
416 input->group, gdblock);
421 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
424 if ((err = ext4_journal_get_write_access(handle, *primary)))
427 if ((err = ext4_journal_get_write_access(handle, dind)))
430 /* ext4_reserve_inode_write() gets a reference on the iloc */
431 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
434 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
439 "not enough memory for %lu groups", gdb_num + 1);
444 * Finally, we have all of the possible failures behind us...
446 * Remove new GDT block from inode double-indirect block and clear out
447 * the new GDT block for use (which also "frees" the backup GDT blocks
448 * from the reserved inode). We don't need to change the bitmaps for
449 * these blocks, because they are marked as in-use from being in the
450 * reserved inode, and will become GDT blocks (primary and backup).
452 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
453 ext4_handle_dirty_metadata(handle, NULL, dind);
455 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
456 ext4_mark_iloc_dirty(handle, inode, &iloc);
457 memset((*primary)->b_data, 0, sb->s_blocksize);
458 ext4_handle_dirty_metadata(handle, NULL, *primary);
460 o_group_desc = EXT4_SB(sb)->s_group_desc;
461 memcpy(n_group_desc, o_group_desc,
462 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
463 n_group_desc[gdb_num] = *primary;
464 EXT4_SB(sb)->s_group_desc = n_group_desc;
465 EXT4_SB(sb)->s_gdb_count++;
468 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
469 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
474 /* ext4_journal_release_buffer(handle, iloc.bh); */
477 /* ext4_journal_release_buffer(handle, dind); */
479 /* ext4_journal_release_buffer(handle, *primary); */
481 /* ext4_journal_release_buffer(handle, *primary); */
487 ext4_debug("leaving with error %d\n", err);
492 * Called when we are adding a new group which has a backup copy of each of
493 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
494 * We need to add these reserved backup GDT blocks to the resize inode, so
495 * that they are kept for future resizing and not allocated to files.
497 * Each reserved backup GDT block will go into a different indirect block.
498 * The indirect blocks are actually the primary reserved GDT blocks,
499 * so we know in advance what their block numbers are. We only get the
500 * double-indirect block to verify it is pointing to the primary reserved
501 * GDT blocks so we don't overwrite a data block by accident. The reserved
502 * backup GDT blocks are stored in their reserved primary GDT block.
504 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
505 struct ext4_new_group_data *input)
507 struct super_block *sb = inode->i_sb;
508 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
509 struct buffer_head **primary;
510 struct buffer_head *dind;
511 struct ext4_iloc iloc;
518 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
522 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
523 dind = sb_bread(sb, le32_to_cpu(*data));
529 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
530 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
531 EXT4_ADDR_PER_BLOCK(sb));
532 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
534 /* Get each reserved primary GDT block and verify it holds backups */
535 for (res = 0; res < reserved_gdb; res++, blk++) {
536 if (le32_to_cpu(*data) != blk) {
537 ext4_warning(sb, "reserved block %llu"
538 " not at offset %ld",
540 (long)(data - (__le32 *)dind->b_data));
544 primary[res] = sb_bread(sb, blk);
549 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
550 brelse(primary[res]);
555 data = (__le32 *)dind->b_data;
558 for (i = 0; i < reserved_gdb; i++) {
559 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
562 for (j = 0; j < i; j++)
563 ext4_journal_release_buffer(handle, primary[j]);
569 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
573 * Finally we can add each of the reserved backup GDT blocks from
574 * the new group to its reserved primary GDT block.
576 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
577 for (i = 0; i < reserved_gdb; i++) {
579 data = (__le32 *)primary[i]->b_data;
580 /* printk("reserving backup %lu[%u] = %lu\n",
581 primary[i]->b_blocknr, gdbackups,
582 blk + primary[i]->b_blocknr); */
583 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
584 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
588 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
589 ext4_mark_iloc_dirty(handle, inode, &iloc);
593 brelse(primary[res]);
603 * Update the backup copies of the ext4 metadata. These don't need to be part
604 * of the main resize transaction, because e2fsck will re-write them if there
605 * is a problem (basically only OOM will cause a problem). However, we
606 * _should_ update the backups if possible, in case the primary gets trashed
607 * for some reason and we need to run e2fsck from a backup superblock. The
608 * important part is that the new block and inode counts are in the backup
609 * superblocks, and the location of the new group metadata in the GDT backups.
611 * We do not need take the s_resize_lock for this, because these
612 * blocks are not otherwise touched by the filesystem code when it is
613 * mounted. We don't need to worry about last changing from
614 * sbi->s_groups_count, because the worst that can happen is that we
615 * do not copy the full number of backups at this time. The resize
616 * which changed s_groups_count will backup again.
618 static void update_backups(struct super_block *sb,
619 int blk_off, char *data, int size)
621 struct ext4_sb_info *sbi = EXT4_SB(sb);
622 const ext4_group_t last = sbi->s_groups_count;
623 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
628 int rest = sb->s_blocksize - size;
632 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
633 if (IS_ERR(handle)) {
635 err = PTR_ERR(handle);
639 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
640 struct buffer_head *bh;
642 /* Out of journal space, and can't get more - abort - so sad */
643 if (ext4_handle_valid(handle) &&
644 handle->h_buffer_credits == 0 &&
645 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
646 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
649 bh = sb_getblk(sb, group * bpg + blk_off);
654 ext4_debug("update metadata backup %#04lx\n",
655 (unsigned long)bh->b_blocknr);
656 if ((err = ext4_journal_get_write_access(handle, bh)))
659 memcpy(bh->b_data, data, size);
661 memset(bh->b_data + size, 0, rest);
662 set_buffer_uptodate(bh);
664 ext4_handle_dirty_metadata(handle, NULL, bh);
667 if ((err2 = ext4_journal_stop(handle)) && !err)
671 * Ugh! Need to have e2fsck write the backup copies. It is too
672 * late to revert the resize, we shouldn't fail just because of
673 * the backup copies (they are only needed in case of corruption).
675 * However, if we got here we have a journal problem too, so we
676 * can't really start a transaction to mark the superblock.
677 * Chicken out and just set the flag on the hope it will be written
678 * to disk, and if not - we will simply wait until next fsck.
682 ext4_warning(sb, "can't update backup for group %u (err %d), "
683 "forcing fsck on next reboot", group, err);
684 sbi->s_mount_state &= ~EXT4_VALID_FS;
685 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
686 mark_buffer_dirty(sbi->s_sbh);
690 /* Add group descriptor data to an existing or new group descriptor block.
691 * Ensure we handle all possible error conditions _before_ we start modifying
692 * the filesystem, because we cannot abort the transaction and not have it
693 * write the data to disk.
695 * If we are on a GDT block boundary, we need to get the reserved GDT block.
696 * Otherwise, we may need to add backup GDT blocks for a sparse group.
698 * We only need to hold the superblock lock while we are actually adding
699 * in the new group's counts to the superblock. Prior to that we have
700 * not really "added" the group at all. We re-check that we are still
701 * adding in the last group in case things have changed since verifying.
703 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
705 struct ext4_sb_info *sbi = EXT4_SB(sb);
706 struct ext4_super_block *es = sbi->s_es;
707 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
708 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
709 struct buffer_head *primary = NULL;
710 struct ext4_group_desc *gdp;
711 struct inode *inode = NULL;
713 int gdb_off, gdb_num;
716 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
717 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
719 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
720 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
721 ext4_warning(sb, "Can't resize non-sparse filesystem further");
725 if (ext4_blocks_count(es) + input->blocks_count <
726 ext4_blocks_count(es)) {
727 ext4_warning(sb, "blocks_count overflow");
731 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
732 le32_to_cpu(es->s_inodes_count)) {
733 ext4_warning(sb, "inodes_count overflow");
737 if (reserved_gdb || gdb_off == 0) {
738 if (!EXT4_HAS_COMPAT_FEATURE(sb,
739 EXT4_FEATURE_COMPAT_RESIZE_INODE)
740 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
742 "No reserved GDT blocks, can't resize");
745 inode = ext4_iget(sb, EXT4_RESIZE_INO);
747 ext4_warning(sb, "Error opening resize inode");
748 return PTR_ERR(inode);
753 if ((err = verify_group_input(sb, input)))
756 if ((err = setup_new_group_blocks(sb, input)))
760 * We will always be modifying at least the superblock and a GDT
761 * block. If we are adding a group past the last current GDT block,
762 * we will also modify the inode and the dindirect block. If we
763 * are adding a group with superblock/GDT backups we will also
764 * modify each of the reserved GDT dindirect blocks.
766 handle = ext4_journal_start_sb(sb,
767 ext4_bg_has_super(sb, input->group) ?
768 3 + reserved_gdb : 4);
769 if (IS_ERR(handle)) {
770 err = PTR_ERR(handle);
774 mutex_lock(&sbi->s_resize_lock);
775 if (input->group != sbi->s_groups_count) {
776 ext4_warning(sb, "multiple resizers run on filesystem!");
781 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
785 * We will only either add reserved group blocks to a backup group
786 * or remove reserved blocks for the first group in a new group block.
787 * Doing both would be mean more complex code, and sane people don't
788 * use non-sparse filesystems anymore. This is already checked above.
791 primary = sbi->s_group_desc[gdb_num];
792 if ((err = ext4_journal_get_write_access(handle, primary)))
795 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
796 (err = reserve_backup_gdb(handle, inode, input)))
798 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
802 * OK, now we've set up the new group. Time to make it active.
804 * We do not lock all allocations via s_resize_lock
805 * so we have to be safe wrt. concurrent accesses the group
806 * data. So we need to be careful to set all of the relevant
807 * group descriptor data etc. *before* we enable the group.
809 * The key field here is sbi->s_groups_count: as long as
810 * that retains its old value, nobody is going to access the new
813 * So first we update all the descriptor metadata for the new
814 * group; then we update the total disk blocks count; then we
815 * update the groups count to enable the group; then finally we
816 * update the free space counts so that the system can start
817 * using the new disk blocks.
820 /* Update group descriptor block for new group */
821 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
822 gdb_off * EXT4_DESC_SIZE(sb));
824 memset(gdp, 0, EXT4_DESC_SIZE(sb));
825 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
826 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
827 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
828 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
829 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
830 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
831 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
834 * We can allocate memory for mb_alloc based on the new group
837 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
842 * Make the new blocks and inodes valid next. We do this before
843 * increasing the group count so that once the group is enabled,
844 * all of its blocks and inodes are already valid.
846 * We always allocate group-by-group, then block-by-block or
847 * inode-by-inode within a group, so enabling these
848 * blocks/inodes before the group is live won't actually let us
849 * allocate the new space yet.
851 ext4_blocks_count_set(es, ext4_blocks_count(es) +
852 input->blocks_count);
853 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
856 * We need to protect s_groups_count against other CPUs seeing
857 * inconsistent state in the superblock.
859 * The precise rules we use are:
861 * * Writers of s_groups_count *must* hold s_resize_lock
863 * * Writers must perform a smp_wmb() after updating all dependent
864 * data and before modifying the groups count
866 * * Readers must hold s_resize_lock over the access
868 * * Readers must perform an smp_rmb() after reading the groups count
869 * and before reading any dependent data.
871 * NB. These rules can be relaxed when checking the group count
872 * while freeing data, as we can only allocate from a block
873 * group after serialising against the group count, and we can
874 * only then free after serialising in turn against that
879 /* Update the global fs size fields */
880 sbi->s_groups_count++;
882 ext4_handle_dirty_metadata(handle, NULL, primary);
884 /* Update the reserved block counts only once the new group is
886 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
887 input->reserved_blocks);
889 /* Update the free space counts */
890 percpu_counter_add(&sbi->s_freeblocks_counter,
891 input->free_blocks_count);
892 percpu_counter_add(&sbi->s_freeinodes_counter,
893 EXT4_INODES_PER_GROUP(sb));
895 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
896 sbi->s_log_groups_per_flex) {
897 ext4_group_t flex_group;
898 flex_group = ext4_flex_group(sbi, input->group);
899 atomic_add(input->free_blocks_count,
900 &sbi->s_flex_groups[flex_group].free_blocks);
901 atomic_add(EXT4_INODES_PER_GROUP(sb),
902 &sbi->s_flex_groups[flex_group].free_inodes);
905 ext4_handle_dirty_super(handle, sb);
908 mutex_unlock(&sbi->s_resize_lock);
909 if ((err2 = ext4_journal_stop(handle)) && !err)
912 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
913 sizeof(struct ext4_super_block));
914 update_backups(sb, primary->b_blocknr, primary->b_data,
920 } /* ext4_group_add */
923 * Extend the filesystem to the new number of blocks specified. This entry
924 * point is only used to extend the current filesystem to the end of the last
925 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
926 * for emergencies (because it has no dependencies on reserved blocks).
928 * If we _really_ wanted, we could use default values to call ext4_group_add()
929 * allow the "remount" trick to work for arbitrary resizing, assuming enough
930 * GDT blocks are reserved to grow to the desired size.
932 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
933 ext4_fsblk_t n_blocks_count)
935 ext4_fsblk_t o_blocks_count;
938 struct buffer_head *bh;
943 /* We don't need to worry about locking wrt other resizers just
944 * yet: we're going to revalidate es->s_blocks_count after
945 * taking the s_resize_lock below. */
946 o_blocks_count = ext4_blocks_count(es);
948 if (test_opt(sb, DEBUG))
949 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
950 o_blocks_count, n_blocks_count);
952 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
955 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
956 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
957 " too large to resize to %llu blocks safely\n",
958 sb->s_id, n_blocks_count);
959 if (sizeof(sector_t) < 8)
960 ext4_warning(sb, "CONFIG_LBDAF not enabled");
964 if (n_blocks_count < o_blocks_count) {
965 ext4_warning(sb, "can't shrink FS - resize aborted");
969 /* Handle the remaining blocks in the last group only. */
970 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
973 ext4_warning(sb, "need to use ext2online to resize further");
977 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
979 if (o_blocks_count + add < o_blocks_count) {
980 ext4_warning(sb, "blocks_count overflow");
984 if (o_blocks_count + add > n_blocks_count)
985 add = n_blocks_count - o_blocks_count;
987 if (o_blocks_count + add < n_blocks_count)
988 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
989 o_blocks_count + add, add);
991 /* See if the device is actually as big as what was requested */
992 bh = sb_bread(sb, o_blocks_count + add - 1);
994 ext4_warning(sb, "can't read last block, resize aborted");
999 /* We will update the superblock, one block bitmap, and
1000 * one group descriptor via ext4_free_blocks().
1002 handle = ext4_journal_start_sb(sb, 3);
1003 if (IS_ERR(handle)) {
1004 err = PTR_ERR(handle);
1005 ext4_warning(sb, "error %d on journal start", err);
1009 mutex_lock(&EXT4_SB(sb)->s_resize_lock);
1010 if (o_blocks_count != ext4_blocks_count(es)) {
1011 ext4_warning(sb, "multiple resizers run on filesystem!");
1012 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1013 ext4_journal_stop(handle);
1018 if ((err = ext4_journal_get_write_access(handle,
1019 EXT4_SB(sb)->s_sbh))) {
1020 ext4_warning(sb, "error %d on journal write access", err);
1021 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1022 ext4_journal_stop(handle);
1025 ext4_blocks_count_set(es, o_blocks_count + add);
1026 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1027 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1028 o_blocks_count + add);
1029 /* We add the blocks to the bitmap and set the group need init bit */
1030 ext4_add_groupblocks(handle, sb, o_blocks_count, add);
1031 ext4_handle_dirty_super(handle, sb);
1032 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1033 o_blocks_count + add);
1034 if ((err = ext4_journal_stop(handle)))
1037 if (test_opt(sb, DEBUG))
1038 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1039 ext4_blocks_count(es));
1040 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1041 sizeof(struct ext4_super_block));
1044 } /* ext4_group_extend */