2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
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,
235 for (i = 0, bit = gdblocks + 1; i < reserved_gdb; i++, bit++)
236 ext4_set_bit(bit, bh->b_data);
238 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
239 input->block_bitmap - start);
240 ext4_set_bit(input->block_bitmap - start, bh->b_data);
241 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
242 input->inode_bitmap - start);
243 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
245 /* Zero out all of the inode table blocks */
246 block = input->inode_table;
247 ext4_debug("clear inode table blocks %#04llx -> %#04llx\n",
248 block, sbi->s_itb_per_group);
249 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
252 for (i = 0, bit = input->inode_table - start;
253 i < sbi->s_itb_per_group; i++, bit++)
254 ext4_set_bit(bit, bh->b_data);
256 if ((err = extend_or_restart_transaction(handle, 2, bh)))
259 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
261 ext4_handle_dirty_metadata(handle, NULL, bh);
263 /* Mark unused entries in inode bitmap used */
264 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
265 input->inode_bitmap, input->inode_bitmap - start);
266 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
271 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
273 ext4_handle_dirty_metadata(handle, NULL, bh);
278 mutex_unlock(&sbi->s_resize_lock);
279 if ((err2 = ext4_journal_stop(handle)) && !err)
286 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
287 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
288 * calling this for the first time. In a sparse filesystem it will be the
289 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
290 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
292 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
293 unsigned *five, unsigned *seven)
295 unsigned *min = three;
299 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
300 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
322 * Check that all of the backup GDT blocks are held in the primary GDT block.
323 * It is assumed that they are stored in group order. Returns the number of
324 * groups in current filesystem that have BACKUPS, or -ve error code.
326 static int verify_reserved_gdb(struct super_block *sb,
327 struct buffer_head *primary)
329 const ext4_fsblk_t blk = primary->b_blocknr;
330 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
335 __le32 *p = (__le32 *)primary->b_data;
338 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
339 if (le32_to_cpu(*p++) !=
340 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
341 ext4_warning(sb, "reserved GDT %llu"
342 " missing grp %d (%llu)",
345 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
349 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
357 * Called when we need to bring a reserved group descriptor table block into
358 * use from the resize inode. The primary copy of the new GDT block currently
359 * is an indirect block (under the double indirect block in the resize inode).
360 * The new backup GDT blocks will be stored as leaf blocks in this indirect
361 * block, in group order. Even though we know all the block numbers we need,
362 * we check to ensure that the resize inode has actually reserved these blocks.
364 * Don't need to update the block bitmaps because the blocks are still in use.
366 * We get all of the error cases out of the way, so that we are sure to not
367 * fail once we start modifying the data on disk, because JBD has no rollback.
369 static int add_new_gdb(handle_t *handle, struct inode *inode,
370 struct ext4_new_group_data *input,
371 struct buffer_head **primary)
373 struct super_block *sb = inode->i_sb;
374 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
375 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
376 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
377 struct buffer_head **o_group_desc, **n_group_desc;
378 struct buffer_head *dind;
380 struct ext4_iloc iloc;
384 if (test_opt(sb, DEBUG))
386 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
390 * If we are not using the primary superblock/GDT copy don't resize,
391 * because the user tools have no way of handling this. Probably a
392 * bad time to do it anyways.
394 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
395 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
396 ext4_warning(sb, "won't resize using backup superblock at %llu",
397 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
401 *primary = sb_bread(sb, gdblock);
405 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
410 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
411 dind = sb_bread(sb, le32_to_cpu(*data));
417 data = (__le32 *)dind->b_data;
418 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
419 ext4_warning(sb, "new group %u GDT block %llu not reserved",
420 input->group, gdblock);
425 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
428 if ((err = ext4_journal_get_write_access(handle, *primary)))
431 if ((err = ext4_journal_get_write_access(handle, dind)))
434 /* ext4_reserve_inode_write() gets a reference on the iloc */
435 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
438 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
443 "not enough memory for %lu groups", gdb_num + 1);
448 * Finally, we have all of the possible failures behind us...
450 * Remove new GDT block from inode double-indirect block and clear out
451 * the new GDT block for use (which also "frees" the backup GDT blocks
452 * from the reserved inode). We don't need to change the bitmaps for
453 * these blocks, because they are marked as in-use from being in the
454 * reserved inode, and will become GDT blocks (primary and backup).
456 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
457 ext4_handle_dirty_metadata(handle, NULL, dind);
459 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
460 ext4_mark_iloc_dirty(handle, inode, &iloc);
461 memset((*primary)->b_data, 0, sb->s_blocksize);
462 ext4_handle_dirty_metadata(handle, NULL, *primary);
464 o_group_desc = EXT4_SB(sb)->s_group_desc;
465 memcpy(n_group_desc, o_group_desc,
466 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
467 n_group_desc[gdb_num] = *primary;
468 EXT4_SB(sb)->s_group_desc = n_group_desc;
469 EXT4_SB(sb)->s_gdb_count++;
472 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
473 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
478 /* ext4_journal_release_buffer(handle, iloc.bh); */
481 /* ext4_journal_release_buffer(handle, dind); */
483 /* ext4_journal_release_buffer(handle, *primary); */
485 /* ext4_journal_release_buffer(handle, *primary); */
491 ext4_debug("leaving with error %d\n", err);
496 * Called when we are adding a new group which has a backup copy of each of
497 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
498 * We need to add these reserved backup GDT blocks to the resize inode, so
499 * that they are kept for future resizing and not allocated to files.
501 * Each reserved backup GDT block will go into a different indirect block.
502 * The indirect blocks are actually the primary reserved GDT blocks,
503 * so we know in advance what their block numbers are. We only get the
504 * double-indirect block to verify it is pointing to the primary reserved
505 * GDT blocks so we don't overwrite a data block by accident. The reserved
506 * backup GDT blocks are stored in their reserved primary GDT block.
508 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
509 struct ext4_new_group_data *input)
511 struct super_block *sb = inode->i_sb;
512 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
513 struct buffer_head **primary;
514 struct buffer_head *dind;
515 struct ext4_iloc iloc;
522 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
526 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
527 dind = sb_bread(sb, le32_to_cpu(*data));
533 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
534 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
535 EXT4_ADDR_PER_BLOCK(sb));
536 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
538 /* Get each reserved primary GDT block and verify it holds backups */
539 for (res = 0; res < reserved_gdb; res++, blk++) {
540 if (le32_to_cpu(*data) != blk) {
541 ext4_warning(sb, "reserved block %llu"
542 " not at offset %ld",
544 (long)(data - (__le32 *)dind->b_data));
548 primary[res] = sb_bread(sb, blk);
553 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
554 brelse(primary[res]);
559 data = (__le32 *)dind->b_data;
562 for (i = 0; i < reserved_gdb; i++) {
563 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
566 for (j = 0; j < i; j++)
567 ext4_journal_release_buffer(handle, primary[j]);
573 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
577 * Finally we can add each of the reserved backup GDT blocks from
578 * the new group to its reserved primary GDT block.
580 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
581 for (i = 0; i < reserved_gdb; i++) {
583 data = (__le32 *)primary[i]->b_data;
584 /* printk("reserving backup %lu[%u] = %lu\n",
585 primary[i]->b_blocknr, gdbackups,
586 blk + primary[i]->b_blocknr); */
587 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
588 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
592 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
593 ext4_mark_iloc_dirty(handle, inode, &iloc);
597 brelse(primary[res]);
607 * Update the backup copies of the ext4 metadata. These don't need to be part
608 * of the main resize transaction, because e2fsck will re-write them if there
609 * is a problem (basically only OOM will cause a problem). However, we
610 * _should_ update the backups if possible, in case the primary gets trashed
611 * for some reason and we need to run e2fsck from a backup superblock. The
612 * important part is that the new block and inode counts are in the backup
613 * superblocks, and the location of the new group metadata in the GDT backups.
615 * We do not need take the s_resize_lock for this, because these
616 * blocks are not otherwise touched by the filesystem code when it is
617 * mounted. We don't need to worry about last changing from
618 * sbi->s_groups_count, because the worst that can happen is that we
619 * do not copy the full number of backups at this time. The resize
620 * which changed s_groups_count will backup again.
622 static void update_backups(struct super_block *sb,
623 int blk_off, char *data, int size)
625 struct ext4_sb_info *sbi = EXT4_SB(sb);
626 const ext4_group_t last = sbi->s_groups_count;
627 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
632 int rest = sb->s_blocksize - size;
636 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
637 if (IS_ERR(handle)) {
639 err = PTR_ERR(handle);
643 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
644 struct buffer_head *bh;
646 /* Out of journal space, and can't get more - abort - so sad */
647 if (ext4_handle_valid(handle) &&
648 handle->h_buffer_credits == 0 &&
649 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
650 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
653 bh = sb_getblk(sb, group * bpg + blk_off);
658 ext4_debug("update metadata backup %#04lx\n",
659 (unsigned long)bh->b_blocknr);
660 if ((err = ext4_journal_get_write_access(handle, bh)))
663 memcpy(bh->b_data, data, size);
665 memset(bh->b_data + size, 0, rest);
666 set_buffer_uptodate(bh);
668 ext4_handle_dirty_metadata(handle, NULL, bh);
671 if ((err2 = ext4_journal_stop(handle)) && !err)
675 * Ugh! Need to have e2fsck write the backup copies. It is too
676 * late to revert the resize, we shouldn't fail just because of
677 * the backup copies (they are only needed in case of corruption).
679 * However, if we got here we have a journal problem too, so we
680 * can't really start a transaction to mark the superblock.
681 * Chicken out and just set the flag on the hope it will be written
682 * to disk, and if not - we will simply wait until next fsck.
686 ext4_warning(sb, "can't update backup for group %u (err %d), "
687 "forcing fsck on next reboot", group, err);
688 sbi->s_mount_state &= ~EXT4_VALID_FS;
689 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
690 mark_buffer_dirty(sbi->s_sbh);
694 /* Add group descriptor data to an existing or new group descriptor block.
695 * Ensure we handle all possible error conditions _before_ we start modifying
696 * the filesystem, because we cannot abort the transaction and not have it
697 * write the data to disk.
699 * If we are on a GDT block boundary, we need to get the reserved GDT block.
700 * Otherwise, we may need to add backup GDT blocks for a sparse group.
702 * We only need to hold the superblock lock while we are actually adding
703 * in the new group's counts to the superblock. Prior to that we have
704 * not really "added" the group at all. We re-check that we are still
705 * adding in the last group in case things have changed since verifying.
707 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
709 struct ext4_sb_info *sbi = EXT4_SB(sb);
710 struct ext4_super_block *es = sbi->s_es;
711 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
712 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
713 struct buffer_head *primary = NULL;
714 struct ext4_group_desc *gdp;
715 struct inode *inode = NULL;
717 int gdb_off, gdb_num;
720 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
721 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
723 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
724 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
725 ext4_warning(sb, "Can't resize non-sparse filesystem further");
729 if (ext4_blocks_count(es) + input->blocks_count <
730 ext4_blocks_count(es)) {
731 ext4_warning(sb, "blocks_count overflow");
735 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
736 le32_to_cpu(es->s_inodes_count)) {
737 ext4_warning(sb, "inodes_count overflow");
741 if (reserved_gdb || gdb_off == 0) {
742 if (!EXT4_HAS_COMPAT_FEATURE(sb,
743 EXT4_FEATURE_COMPAT_RESIZE_INODE)
744 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
746 "No reserved GDT blocks, can't resize");
749 inode = ext4_iget(sb, EXT4_RESIZE_INO);
751 ext4_warning(sb, "Error opening resize inode");
752 return PTR_ERR(inode);
757 if ((err = verify_group_input(sb, input)))
760 if ((err = setup_new_group_blocks(sb, input)))
764 * We will always be modifying at least the superblock and a GDT
765 * block. If we are adding a group past the last current GDT block,
766 * we will also modify the inode and the dindirect block. If we
767 * are adding a group with superblock/GDT backups we will also
768 * modify each of the reserved GDT dindirect blocks.
770 handle = ext4_journal_start_sb(sb,
771 ext4_bg_has_super(sb, input->group) ?
772 3 + reserved_gdb : 4);
773 if (IS_ERR(handle)) {
774 err = PTR_ERR(handle);
778 mutex_lock(&sbi->s_resize_lock);
779 if (input->group != sbi->s_groups_count) {
780 ext4_warning(sb, "multiple resizers run on filesystem!");
785 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
789 * We will only either add reserved group blocks to a backup group
790 * or remove reserved blocks for the first group in a new group block.
791 * Doing both would be mean more complex code, and sane people don't
792 * use non-sparse filesystems anymore. This is already checked above.
795 primary = sbi->s_group_desc[gdb_num];
796 if ((err = ext4_journal_get_write_access(handle, primary)))
799 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
800 (err = reserve_backup_gdb(handle, inode, input)))
802 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
806 * OK, now we've set up the new group. Time to make it active.
808 * We do not lock all allocations via s_resize_lock
809 * so we have to be safe wrt. concurrent accesses the group
810 * data. So we need to be careful to set all of the relevant
811 * group descriptor data etc. *before* we enable the group.
813 * The key field here is sbi->s_groups_count: as long as
814 * that retains its old value, nobody is going to access the new
817 * So first we update all the descriptor metadata for the new
818 * group; then we update the total disk blocks count; then we
819 * update the groups count to enable the group; then finally we
820 * update the free space counts so that the system can start
821 * using the new disk blocks.
824 /* Update group descriptor block for new group */
825 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
826 gdb_off * EXT4_DESC_SIZE(sb));
828 memset(gdp, 0, EXT4_DESC_SIZE(sb));
829 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
830 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
831 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
832 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
833 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
834 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
835 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
838 * We can allocate memory for mb_alloc based on the new group
841 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
846 * Make the new blocks and inodes valid next. We do this before
847 * increasing the group count so that once the group is enabled,
848 * all of its blocks and inodes are already valid.
850 * We always allocate group-by-group, then block-by-block or
851 * inode-by-inode within a group, so enabling these
852 * blocks/inodes before the group is live won't actually let us
853 * allocate the new space yet.
855 ext4_blocks_count_set(es, ext4_blocks_count(es) +
856 input->blocks_count);
857 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
860 * We need to protect s_groups_count against other CPUs seeing
861 * inconsistent state in the superblock.
863 * The precise rules we use are:
865 * * Writers of s_groups_count *must* hold s_resize_lock
867 * * Writers must perform a smp_wmb() after updating all dependent
868 * data and before modifying the groups count
870 * * Readers must hold s_resize_lock over the access
872 * * Readers must perform an smp_rmb() after reading the groups count
873 * and before reading any dependent data.
875 * NB. These rules can be relaxed when checking the group count
876 * while freeing data, as we can only allocate from a block
877 * group after serialising against the group count, and we can
878 * only then free after serialising in turn against that
883 /* Update the global fs size fields */
884 sbi->s_groups_count++;
886 ext4_handle_dirty_metadata(handle, NULL, primary);
888 /* Update the reserved block counts only once the new group is
890 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
891 input->reserved_blocks);
893 /* Update the free space counts */
894 percpu_counter_add(&sbi->s_freeblocks_counter,
895 input->free_blocks_count);
896 percpu_counter_add(&sbi->s_freeinodes_counter,
897 EXT4_INODES_PER_GROUP(sb));
899 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
900 sbi->s_log_groups_per_flex) {
901 ext4_group_t flex_group;
902 flex_group = ext4_flex_group(sbi, input->group);
903 atomic_add(input->free_blocks_count,
904 &sbi->s_flex_groups[flex_group].free_blocks);
905 atomic_add(EXT4_INODES_PER_GROUP(sb),
906 &sbi->s_flex_groups[flex_group].free_inodes);
909 ext4_handle_dirty_super(handle, sb);
912 mutex_unlock(&sbi->s_resize_lock);
913 if ((err2 = ext4_journal_stop(handle)) && !err)
916 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
917 sizeof(struct ext4_super_block));
918 update_backups(sb, primary->b_blocknr, primary->b_data,
924 } /* ext4_group_add */
927 * Extend the filesystem to the new number of blocks specified. This entry
928 * point is only used to extend the current filesystem to the end of the last
929 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
930 * for emergencies (because it has no dependencies on reserved blocks).
932 * If we _really_ wanted, we could use default values to call ext4_group_add()
933 * allow the "remount" trick to work for arbitrary resizing, assuming enough
934 * GDT blocks are reserved to grow to the desired size.
936 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
937 ext4_fsblk_t n_blocks_count)
939 ext4_fsblk_t o_blocks_count;
942 struct buffer_head *bh;
947 /* We don't need to worry about locking wrt other resizers just
948 * yet: we're going to revalidate es->s_blocks_count after
949 * taking the s_resize_lock below. */
950 o_blocks_count = ext4_blocks_count(es);
952 if (test_opt(sb, DEBUG))
953 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
954 o_blocks_count, n_blocks_count);
956 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
959 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
960 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
961 " too large to resize to %llu blocks safely\n",
962 sb->s_id, n_blocks_count);
963 if (sizeof(sector_t) < 8)
964 ext4_warning(sb, "CONFIG_LBDAF not enabled");
968 if (n_blocks_count < o_blocks_count) {
969 ext4_warning(sb, "can't shrink FS - resize aborted");
973 /* Handle the remaining blocks in the last group only. */
974 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
977 ext4_warning(sb, "need to use ext2online to resize further");
981 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
983 if (o_blocks_count + add < o_blocks_count) {
984 ext4_warning(sb, "blocks_count overflow");
988 if (o_blocks_count + add > n_blocks_count)
989 add = n_blocks_count - o_blocks_count;
991 if (o_blocks_count + add < n_blocks_count)
992 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
993 o_blocks_count + add, add);
995 /* See if the device is actually as big as what was requested */
996 bh = sb_bread(sb, o_blocks_count + add - 1);
998 ext4_warning(sb, "can't read last block, resize aborted");
1003 /* We will update the superblock, one block bitmap, and
1004 * one group descriptor via ext4_free_blocks().
1006 handle = ext4_journal_start_sb(sb, 3);
1007 if (IS_ERR(handle)) {
1008 err = PTR_ERR(handle);
1009 ext4_warning(sb, "error %d on journal start", err);
1013 mutex_lock(&EXT4_SB(sb)->s_resize_lock);
1014 if (o_blocks_count != ext4_blocks_count(es)) {
1015 ext4_warning(sb, "multiple resizers run on filesystem!");
1016 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1017 ext4_journal_stop(handle);
1022 if ((err = ext4_journal_get_write_access(handle,
1023 EXT4_SB(sb)->s_sbh))) {
1024 ext4_warning(sb, "error %d on journal write access", err);
1025 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1026 ext4_journal_stop(handle);
1029 ext4_blocks_count_set(es, o_blocks_count + add);
1030 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1031 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1032 o_blocks_count + add);
1033 /* We add the blocks to the bitmap and set the group need init bit */
1034 ext4_add_groupblocks(handle, sb, o_blocks_count, add);
1035 ext4_handle_dirty_super(handle, sb);
1036 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1037 o_blocks_count + add);
1038 if ((err = ext4_journal_stop(handle)))
1041 if (test_opt(sb, DEBUG))
1042 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1043 ext4_blocks_count(es));
1044 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1045 sizeof(struct ext4_super_block));
1048 } /* ext4_group_extend */