1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/resize.c
5 * Support for resizing an ext4 filesystem while it is mounted.
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
9 * This could probably be made into a module, because it is not often in use.
15 #include <linux/errno.h>
16 #include <linux/slab.h>
18 #include "ext4_jbd2.h"
25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
27 struct ext4_rcu_ptr *ptr;
29 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
34 void ext4_kvfree_array_rcu(void *to_free)
36 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
40 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
47 int ext4_resize_begin(struct super_block *sb)
49 struct ext4_sb_info *sbi = EXT4_SB(sb);
52 if (!capable(CAP_SYS_RESOURCE))
56 * If we are not using the primary superblock/GDT copy don't resize,
57 * because the user tools have no way of handling this. Probably a
58 * bad time to do it anyways.
60 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
61 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
62 ext4_warning(sb, "won't resize using backup superblock at %llu",
63 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
68 * We are not allowed to do online-resizing on a filesystem mounted
69 * with error, because it can destroy the filesystem easily.
71 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
72 ext4_warning(sb, "There are errors in the filesystem, "
73 "so online resizing is not allowed");
77 if (ext4_has_feature_sparse_super2(sb)) {
78 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
82 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
83 &EXT4_SB(sb)->s_ext4_flags))
89 void ext4_resize_end(struct super_block *sb)
91 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
92 smp_mb__after_atomic();
95 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
97 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
98 EXT4_DESC_PER_BLOCK_BITS(sb);
101 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
102 ext4_group_t group) {
103 group = ext4_meta_bg_first_group(sb, group);
104 return ext4_group_first_block_no(sb, group);
107 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
108 ext4_group_t group) {
109 ext4_grpblk_t overhead;
110 overhead = ext4_bg_num_gdb(sb, group);
111 if (ext4_bg_has_super(sb, group))
113 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
117 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
118 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
120 static int verify_group_input(struct super_block *sb,
121 struct ext4_new_group_data *input)
123 struct ext4_sb_info *sbi = EXT4_SB(sb);
124 struct ext4_super_block *es = sbi->s_es;
125 ext4_fsblk_t start = ext4_blocks_count(es);
126 ext4_fsblk_t end = start + input->blocks_count;
127 ext4_group_t group = input->group;
128 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
130 ext4_fsblk_t metaend;
131 struct buffer_head *bh = NULL;
132 ext4_grpblk_t free_blocks_count, offset;
135 if (group != sbi->s_groups_count) {
136 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
137 input->group, sbi->s_groups_count);
141 overhead = ext4_group_overhead_blocks(sb, group);
142 metaend = start + overhead;
143 input->free_clusters_count = free_blocks_count =
144 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
146 if (test_opt(sb, DEBUG))
147 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
148 "(%d free, %u reserved)\n",
149 ext4_bg_has_super(sb, input->group) ? "normal" :
150 "no-super", input->group, input->blocks_count,
151 free_blocks_count, input->reserved_blocks);
153 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
155 ext4_warning(sb, "Last group not full");
156 else if (input->reserved_blocks > input->blocks_count / 5)
157 ext4_warning(sb, "Reserved blocks too high (%u)",
158 input->reserved_blocks);
159 else if (free_blocks_count < 0)
160 ext4_warning(sb, "Bad blocks count %u",
161 input->blocks_count);
162 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
165 ext4_warning(sb, "Cannot read last block (%llu)",
167 } else if (outside(input->block_bitmap, start, end))
168 ext4_warning(sb, "Block bitmap not in group (block %llu)",
169 (unsigned long long)input->block_bitmap);
170 else if (outside(input->inode_bitmap, start, end))
171 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
172 (unsigned long long)input->inode_bitmap);
173 else if (outside(input->inode_table, start, end) ||
174 outside(itend - 1, start, end))
175 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
176 (unsigned long long)input->inode_table, itend - 1);
177 else if (input->inode_bitmap == input->block_bitmap)
178 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (inside(input->block_bitmap, input->inode_table, itend))
181 ext4_warning(sb, "Block bitmap (%llu) in inode table "
183 (unsigned long long)input->block_bitmap,
184 (unsigned long long)input->inode_table, itend - 1);
185 else if (inside(input->inode_bitmap, input->inode_table, itend))
186 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
188 (unsigned long long)input->inode_bitmap,
189 (unsigned long long)input->inode_table, itend - 1);
190 else if (inside(input->block_bitmap, start, metaend))
191 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
192 (unsigned long long)input->block_bitmap,
194 else if (inside(input->inode_bitmap, start, metaend))
195 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
196 (unsigned long long)input->inode_bitmap,
198 else if (inside(input->inode_table, start, metaend) ||
199 inside(itend - 1, start, metaend))
200 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
202 (unsigned long long)input->inode_table,
203 itend - 1, start, metaend - 1);
212 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
215 struct ext4_new_flex_group_data {
216 struct ext4_new_group_data *groups; /* new_group_data for groups
218 __u16 *bg_flags; /* block group flags of groups
220 ext4_group_t count; /* number of groups in @groups
225 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
228 * Returns NULL on failure otherwise address of the allocated structure.
230 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
232 struct ext4_new_flex_group_data *flex_gd;
234 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
238 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
240 flex_gd->count = flexbg_size;
242 flex_gd->groups = kmalloc_array(flexbg_size,
243 sizeof(struct ext4_new_group_data),
245 if (flex_gd->groups == NULL)
248 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
250 if (flex_gd->bg_flags == NULL)
256 kfree(flex_gd->groups);
263 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
265 kfree(flex_gd->bg_flags);
266 kfree(flex_gd->groups);
271 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
272 * and inode tables for a flex group.
274 * This function is used by 64bit-resize. Note that this function allocates
275 * group tables from the 1st group of groups contained by @flexgd, which may
276 * be a partial of a flex group.
278 * @sb: super block of fs to which the groups belongs
280 * Returns 0 on a successful allocation of the metadata blocks in the
283 static int ext4_alloc_group_tables(struct super_block *sb,
284 struct ext4_new_flex_group_data *flex_gd,
287 struct ext4_new_group_data *group_data = flex_gd->groups;
288 ext4_fsblk_t start_blk;
289 ext4_fsblk_t last_blk;
290 ext4_group_t src_group;
291 ext4_group_t bb_index = 0;
292 ext4_group_t ib_index = 0;
293 ext4_group_t it_index = 0;
295 ext4_group_t last_group;
297 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
300 BUG_ON(flex_gd->count == 0 || group_data == NULL);
302 src_group = group_data[0].group;
303 last_group = src_group + flex_gd->count - 1;
305 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
306 (last_group & ~(flexbg_size - 1))));
308 group = group_data[0].group;
309 if (src_group >= group_data[0].group + flex_gd->count)
311 start_blk = ext4_group_first_block_no(sb, src_group);
312 last_blk = start_blk + group_data[src_group - group].blocks_count;
314 overhead = ext4_group_overhead_blocks(sb, src_group);
316 start_blk += overhead;
318 /* We collect contiguous blocks as much as possible. */
320 for (; src_group <= last_group; src_group++) {
321 overhead = ext4_group_overhead_blocks(sb, src_group);
323 last_blk += group_data[src_group - group].blocks_count;
328 /* Allocate block bitmaps */
329 for (; bb_index < flex_gd->count; bb_index++) {
330 if (start_blk >= last_blk)
332 group_data[bb_index].block_bitmap = start_blk++;
333 group = ext4_get_group_number(sb, start_blk - 1);
334 group -= group_data[0].group;
335 group_data[group].mdata_blocks++;
336 flex_gd->bg_flags[group] &= uninit_mask;
339 /* Allocate inode bitmaps */
340 for (; ib_index < flex_gd->count; ib_index++) {
341 if (start_blk >= last_blk)
343 group_data[ib_index].inode_bitmap = start_blk++;
344 group = ext4_get_group_number(sb, start_blk - 1);
345 group -= group_data[0].group;
346 group_data[group].mdata_blocks++;
347 flex_gd->bg_flags[group] &= uninit_mask;
350 /* Allocate inode tables */
351 for (; it_index < flex_gd->count; it_index++) {
352 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
353 ext4_fsblk_t next_group_start;
355 if (start_blk + itb > last_blk)
357 group_data[it_index].inode_table = start_blk;
358 group = ext4_get_group_number(sb, start_blk);
359 next_group_start = ext4_group_first_block_no(sb, group + 1);
360 group -= group_data[0].group;
362 if (start_blk + itb > next_group_start) {
363 flex_gd->bg_flags[group + 1] &= uninit_mask;
364 overhead = start_blk + itb - next_group_start;
365 group_data[group + 1].mdata_blocks += overhead;
369 group_data[group].mdata_blocks += itb;
370 flex_gd->bg_flags[group] &= uninit_mask;
371 start_blk += EXT4_SB(sb)->s_itb_per_group;
374 /* Update free clusters count to exclude metadata blocks */
375 for (i = 0; i < flex_gd->count; i++) {
376 group_data[i].free_clusters_count -=
377 EXT4_NUM_B2C(EXT4_SB(sb),
378 group_data[i].mdata_blocks);
381 if (test_opt(sb, DEBUG)) {
383 group = group_data[0].group;
385 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
386 "%d groups, flexbg size is %d:\n", flex_gd->count,
389 for (i = 0; i < flex_gd->count; i++) {
391 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
392 ext4_bg_has_super(sb, group + i) ? "normal" :
393 "no-super", group + i,
394 group_data[i].blocks_count,
395 group_data[i].free_clusters_count,
396 group_data[i].mdata_blocks);
402 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
405 struct buffer_head *bh;
408 bh = sb_getblk(sb, blk);
410 return ERR_PTR(-ENOMEM);
411 BUFFER_TRACE(bh, "get_write_access");
412 if ((err = ext4_journal_get_write_access(handle, bh))) {
416 memset(bh->b_data, 0, sb->s_blocksize);
417 set_buffer_uptodate(bh);
423 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
425 return ext4_journal_ensure_credits_fn(handle, credits,
426 EXT4_MAX_TRANS_DATA, 0, 0);
430 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
432 * Helper function for ext4_setup_new_group_blocks() which set .
435 * @handle: journal handle
436 * @flex_gd: flex group data
438 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
439 struct ext4_new_flex_group_data *flex_gd,
440 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
442 struct ext4_sb_info *sbi = EXT4_SB(sb);
443 ext4_group_t count = last_cluster - first_cluster + 1;
446 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
448 for (count2 = count; count > 0;
449 count -= count2, first_cluster += count2) {
451 struct buffer_head *bh;
455 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
456 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
457 group -= flex_gd->groups[0].group;
459 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
463 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
464 BUG_ON(flex_gd->count > 1);
468 err = ext4_resize_ensure_credits_batch(handle, 1);
472 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
476 BUFFER_TRACE(bh, "get_write_access");
477 err = ext4_journal_get_write_access(handle, bh);
482 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
483 first_cluster, first_cluster - start, count2);
484 ext4_set_bits(bh->b_data, first_cluster - start, count2);
486 err = ext4_handle_dirty_metadata(handle, NULL, bh);
496 * Set up the block and inode bitmaps, and the inode table for the new groups.
497 * This doesn't need to be part of the main transaction, since we are only
498 * changing blocks outside the actual filesystem. We still do journaling to
499 * ensure the recovery is correct in case of a failure just after resize.
500 * If any part of this fails, we simply abort the resize.
502 * setup_new_flex_group_blocks handles a flex group as follow:
503 * 1. copy super block and GDT, and initialize group tables if necessary.
504 * In this step, we only set bits in blocks bitmaps for blocks taken by
505 * super block and GDT.
506 * 2. allocate group tables in block bitmaps, that is, set bits in block
507 * bitmap for blocks taken by group tables.
509 static int setup_new_flex_group_blocks(struct super_block *sb,
510 struct ext4_new_flex_group_data *flex_gd)
512 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
515 struct ext4_sb_info *sbi = EXT4_SB(sb);
516 struct ext4_super_block *es = sbi->s_es;
517 struct ext4_new_group_data *group_data = flex_gd->groups;
518 __u16 *bg_flags = flex_gd->bg_flags;
520 ext4_group_t group, count;
521 struct buffer_head *bh = NULL;
522 int reserved_gdb, i, j, err = 0, err2;
525 BUG_ON(!flex_gd->count || !group_data ||
526 group_data[0].group != sbi->s_groups_count);
528 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
529 meta_bg = ext4_has_feature_meta_bg(sb);
531 /* This transaction may be extended/restarted along the way */
532 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
534 return PTR_ERR(handle);
536 group = group_data[0].group;
537 for (i = 0; i < flex_gd->count; i++, group++) {
538 unsigned long gdblocks;
539 ext4_grpblk_t overhead;
541 gdblocks = ext4_bg_num_gdb(sb, group);
542 start = ext4_group_first_block_no(sb, group);
544 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
548 ext4_group_t first_group;
549 first_group = ext4_meta_bg_first_group(sb, group);
550 if (first_group != group + 1 &&
551 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
555 block = start + ext4_bg_has_super(sb, group);
556 /* Copy all of the GDT blocks into the backup in this group */
557 for (j = 0; j < gdblocks; j++, block++) {
558 struct buffer_head *gdb;
560 ext4_debug("update backup group %#04llx\n", block);
561 err = ext4_resize_ensure_credits_batch(handle, 1);
565 gdb = sb_getblk(sb, block);
566 if (unlikely(!gdb)) {
571 BUFFER_TRACE(gdb, "get_write_access");
572 err = ext4_journal_get_write_access(handle, gdb);
577 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
578 s_group_desc, j)->b_data, gdb->b_size);
579 set_buffer_uptodate(gdb);
581 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
589 /* Zero out all of the reserved backup group descriptor
592 if (ext4_bg_has_super(sb, group)) {
593 err = sb_issue_zeroout(sb, gdblocks + start + 1,
594 reserved_gdb, GFP_NOFS);
600 /* Initialize group tables of the grop @group */
601 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
604 /* Zero out all of the inode table blocks */
605 block = group_data[i].inode_table;
606 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
607 block, sbi->s_itb_per_group);
608 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
614 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
617 /* Initialize block bitmap of the @group */
618 block = group_data[i].block_bitmap;
619 err = ext4_resize_ensure_credits_batch(handle, 1);
623 bh = bclean(handle, sb, block);
628 overhead = ext4_group_overhead_blocks(sb, group);
630 ext4_debug("mark backup superblock %#04llx (+0)\n",
632 ext4_set_bits(bh->b_data, 0,
633 EXT4_NUM_B2C(sbi, overhead));
635 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
636 sb->s_blocksize * 8, bh->b_data);
637 err = ext4_handle_dirty_metadata(handle, NULL, bh);
643 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
646 /* Initialize inode bitmap of the @group */
647 block = group_data[i].inode_bitmap;
648 err = ext4_resize_ensure_credits_batch(handle, 1);
651 /* Mark unused entries in inode bitmap used */
652 bh = bclean(handle, sb, block);
658 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
659 sb->s_blocksize * 8, bh->b_data);
660 err = ext4_handle_dirty_metadata(handle, NULL, bh);
666 /* Mark group tables in block bitmap */
667 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
668 count = group_table_count[j];
669 start = (&group_data[0].block_bitmap)[j];
671 for (i = 1; i < flex_gd->count; i++) {
672 block += group_table_count[j];
673 if (block == (&group_data[i].block_bitmap)[j]) {
674 count += group_table_count[j];
677 err = set_flexbg_block_bitmap(sb, handle,
679 EXT4_B2C(sbi, start),
685 count = group_table_count[j];
686 start = (&group_data[i].block_bitmap)[j];
691 err = set_flexbg_block_bitmap(sb, handle,
693 EXT4_B2C(sbi, start),
703 err2 = ext4_journal_stop(handle);
711 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
712 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
713 * calling this for the first time. In a sparse filesystem it will be the
714 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
715 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
717 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
718 unsigned *five, unsigned *seven)
720 unsigned *min = three;
724 if (!ext4_has_feature_sparse_super(sb)) {
746 * Check that all of the backup GDT blocks are held in the primary GDT block.
747 * It is assumed that they are stored in group order. Returns the number of
748 * groups in current filesystem that have BACKUPS, or -ve error code.
750 static int verify_reserved_gdb(struct super_block *sb,
752 struct buffer_head *primary)
754 const ext4_fsblk_t blk = primary->b_blocknr;
759 __le32 *p = (__le32 *)primary->b_data;
762 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
763 if (le32_to_cpu(*p++) !=
764 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
765 ext4_warning(sb, "reserved GDT %llu"
766 " missing grp %d (%llu)",
769 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
773 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
781 * Called when we need to bring a reserved group descriptor table block into
782 * use from the resize inode. The primary copy of the new GDT block currently
783 * is an indirect block (under the double indirect block in the resize inode).
784 * The new backup GDT blocks will be stored as leaf blocks in this indirect
785 * block, in group order. Even though we know all the block numbers we need,
786 * we check to ensure that the resize inode has actually reserved these blocks.
788 * Don't need to update the block bitmaps because the blocks are still in use.
790 * We get all of the error cases out of the way, so that we are sure to not
791 * fail once we start modifying the data on disk, because JBD has no rollback.
793 static int add_new_gdb(handle_t *handle, struct inode *inode,
796 struct super_block *sb = inode->i_sb;
797 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
798 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
799 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
800 struct buffer_head **o_group_desc, **n_group_desc = NULL;
801 struct buffer_head *dind = NULL;
802 struct buffer_head *gdb_bh = NULL;
804 struct ext4_iloc iloc = { .bh = NULL };
808 if (test_opt(sb, DEBUG))
810 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
813 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
815 return PTR_ERR(gdb_bh);
817 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
823 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
824 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
831 data = (__le32 *)dind->b_data;
832 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
833 ext4_warning(sb, "new group %u GDT block %llu not reserved",
839 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
840 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
844 BUFFER_TRACE(gdb_bh, "get_write_access");
845 err = ext4_journal_get_write_access(handle, gdb_bh);
849 BUFFER_TRACE(dind, "get_write_access");
850 err = ext4_journal_get_write_access(handle, dind);
852 ext4_std_error(sb, err);
856 /* ext4_reserve_inode_write() gets a reference on the iloc */
857 err = ext4_reserve_inode_write(handle, inode, &iloc);
861 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
865 ext4_warning(sb, "not enough memory for %lu groups",
871 * Finally, we have all of the possible failures behind us...
873 * Remove new GDT block from inode double-indirect block and clear out
874 * the new GDT block for use (which also "frees" the backup GDT blocks
875 * from the reserved inode). We don't need to change the bitmaps for
876 * these blocks, because they are marked as in-use from being in the
877 * reserved inode, and will become GDT blocks (primary and backup).
879 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
880 err = ext4_handle_dirty_metadata(handle, NULL, dind);
882 ext4_std_error(sb, err);
885 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
886 (9 - EXT4_SB(sb)->s_cluster_bits);
887 ext4_mark_iloc_dirty(handle, inode, &iloc);
888 memset(gdb_bh->b_data, 0, sb->s_blocksize);
889 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
891 ext4_std_error(sb, err);
898 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
899 memcpy(n_group_desc, o_group_desc,
900 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
902 n_group_desc[gdb_num] = gdb_bh;
903 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
904 EXT4_SB(sb)->s_gdb_count++;
905 ext4_kvfree_array_rcu(o_group_desc);
907 lock_buffer(EXT4_SB(sb)->s_sbh);
908 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
909 ext4_superblock_csum_set(sb);
910 unlock_buffer(EXT4_SB(sb)->s_sbh);
911 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
913 ext4_std_error(sb, err);
916 kvfree(n_group_desc);
921 ext4_debug("leaving with error %d\n", err);
926 * add_new_gdb_meta_bg is the sister of add_new_gdb.
928 static int add_new_gdb_meta_bg(struct super_block *sb,
929 handle_t *handle, ext4_group_t group) {
930 ext4_fsblk_t gdblock;
931 struct buffer_head *gdb_bh;
932 struct buffer_head **o_group_desc, **n_group_desc;
933 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
936 gdblock = ext4_meta_bg_first_block_no(sb, group) +
937 ext4_bg_has_super(sb, group);
938 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
940 return PTR_ERR(gdb_bh);
941 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
946 ext4_warning(sb, "not enough memory for %lu groups",
952 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
953 memcpy(n_group_desc, o_group_desc,
954 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
956 n_group_desc[gdb_num] = gdb_bh;
958 BUFFER_TRACE(gdb_bh, "get_write_access");
959 err = ext4_journal_get_write_access(handle, gdb_bh);
961 kvfree(n_group_desc);
966 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
967 EXT4_SB(sb)->s_gdb_count++;
968 ext4_kvfree_array_rcu(o_group_desc);
973 * Called when we are adding a new group which has a backup copy of each of
974 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
975 * We need to add these reserved backup GDT blocks to the resize inode, so
976 * that they are kept for future resizing and not allocated to files.
978 * Each reserved backup GDT block will go into a different indirect block.
979 * The indirect blocks are actually the primary reserved GDT blocks,
980 * so we know in advance what their block numbers are. We only get the
981 * double-indirect block to verify it is pointing to the primary reserved
982 * GDT blocks so we don't overwrite a data block by accident. The reserved
983 * backup GDT blocks are stored in their reserved primary GDT block.
985 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
988 struct super_block *sb = inode->i_sb;
989 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
990 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
991 struct buffer_head **primary;
992 struct buffer_head *dind;
993 struct ext4_iloc iloc;
1000 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1004 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1005 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1007 err = PTR_ERR(dind);
1012 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1013 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1014 EXT4_ADDR_PER_BLOCK(sb));
1015 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1017 /* Get each reserved primary GDT block and verify it holds backups */
1018 for (res = 0; res < reserved_gdb; res++, blk++) {
1019 if (le32_to_cpu(*data) != blk) {
1020 ext4_warning(sb, "reserved block %llu"
1021 " not at offset %ld",
1023 (long)(data - (__le32 *)dind->b_data));
1027 primary[res] = ext4_sb_bread(sb, blk, 0);
1028 if (IS_ERR(primary[res])) {
1029 err = PTR_ERR(primary[res]);
1030 primary[res] = NULL;
1033 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1034 if (gdbackups < 0) {
1035 brelse(primary[res]);
1040 data = (__le32 *)dind->b_data;
1043 for (i = 0; i < reserved_gdb; i++) {
1044 BUFFER_TRACE(primary[i], "get_write_access");
1045 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1049 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1053 * Finally we can add each of the reserved backup GDT blocks from
1054 * the new group to its reserved primary GDT block.
1056 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1057 for (i = 0; i < reserved_gdb; i++) {
1059 data = (__le32 *)primary[i]->b_data;
1060 /* printk("reserving backup %lu[%u] = %lu\n",
1061 primary[i]->b_blocknr, gdbackups,
1062 blk + primary[i]->b_blocknr); */
1063 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1064 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1069 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1070 ext4_mark_iloc_dirty(handle, inode, &iloc);
1074 brelse(primary[res]);
1084 * Update the backup copies of the ext4 metadata. These don't need to be part
1085 * of the main resize transaction, because e2fsck will re-write them if there
1086 * is a problem (basically only OOM will cause a problem). However, we
1087 * _should_ update the backups if possible, in case the primary gets trashed
1088 * for some reason and we need to run e2fsck from a backup superblock. The
1089 * important part is that the new block and inode counts are in the backup
1090 * superblocks, and the location of the new group metadata in the GDT backups.
1092 * We do not need take the s_resize_lock for this, because these
1093 * blocks are not otherwise touched by the filesystem code when it is
1094 * mounted. We don't need to worry about last changing from
1095 * sbi->s_groups_count, because the worst that can happen is that we
1096 * do not copy the full number of backups at this time. The resize
1097 * which changed s_groups_count will backup again.
1099 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1100 int size, int meta_bg)
1102 struct ext4_sb_info *sbi = EXT4_SB(sb);
1104 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1108 ext4_group_t group = 0;
1109 int rest = sb->s_blocksize - size;
1113 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1114 if (IS_ERR(handle)) {
1116 err = PTR_ERR(handle);
1121 group = ext4_list_backups(sb, &three, &five, &seven);
1122 last = sbi->s_groups_count;
1124 group = ext4_get_group_number(sb, blk_off) + 1;
1125 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1128 while (group < sbi->s_groups_count) {
1129 struct buffer_head *bh;
1130 ext4_fsblk_t backup_block;
1132 /* Out of journal space, and can't get more - abort - so sad */
1133 err = ext4_resize_ensure_credits_batch(handle, 1);
1138 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1140 backup_block = (ext4_group_first_block_no(sb, group) +
1141 ext4_bg_has_super(sb, group));
1143 bh = sb_getblk(sb, backup_block);
1144 if (unlikely(!bh)) {
1148 ext4_debug("update metadata backup %llu(+%llu)\n",
1149 backup_block, backup_block -
1150 ext4_group_first_block_no(sb, group));
1151 BUFFER_TRACE(bh, "get_write_access");
1152 if ((err = ext4_journal_get_write_access(handle, bh))) {
1157 memcpy(bh->b_data, data, size);
1159 memset(bh->b_data + size, 0, rest);
1160 set_buffer_uptodate(bh);
1162 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1164 ext4_std_error(sb, err);
1168 group = ext4_list_backups(sb, &three, &five, &seven);
1169 else if (group == last)
1174 if ((err2 = ext4_journal_stop(handle)) && !err)
1178 * Ugh! Need to have e2fsck write the backup copies. It is too
1179 * late to revert the resize, we shouldn't fail just because of
1180 * the backup copies (they are only needed in case of corruption).
1182 * However, if we got here we have a journal problem too, so we
1183 * can't really start a transaction to mark the superblock.
1184 * Chicken out and just set the flag on the hope it will be written
1185 * to disk, and if not - we will simply wait until next fsck.
1189 ext4_warning(sb, "can't update backup for group %u (err %d), "
1190 "forcing fsck on next reboot", group, err);
1191 sbi->s_mount_state &= ~EXT4_VALID_FS;
1192 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1193 mark_buffer_dirty(sbi->s_sbh);
1198 * ext4_add_new_descs() adds @count group descriptor of groups
1199 * starting at @group
1201 * @handle: journal handle
1203 * @group: the group no. of the first group desc to be added
1204 * @resize_inode: the resize inode
1205 * @count: number of group descriptors to be added
1207 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1208 ext4_group_t group, struct inode *resize_inode,
1211 struct ext4_sb_info *sbi = EXT4_SB(sb);
1212 struct ext4_super_block *es = sbi->s_es;
1213 struct buffer_head *gdb_bh;
1214 int i, gdb_off, gdb_num, err = 0;
1217 meta_bg = ext4_has_feature_meta_bg(sb);
1218 for (i = 0; i < count; i++, group++) {
1219 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1220 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1222 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1223 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1226 * We will only either add reserved group blocks to a backup group
1227 * or remove reserved blocks for the first group in a new group block.
1228 * Doing both would be mean more complex code, and sane people don't
1229 * use non-sparse filesystems anymore. This is already checked above.
1232 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1234 BUFFER_TRACE(gdb_bh, "get_write_access");
1235 err = ext4_journal_get_write_access(handle, gdb_bh);
1237 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1238 err = reserve_backup_gdb(handle, resize_inode, group);
1239 } else if (meta_bg != 0) {
1240 err = add_new_gdb_meta_bg(sb, handle, group);
1242 err = add_new_gdb(handle, resize_inode, group);
1250 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1252 struct buffer_head *bh = sb_getblk(sb, block);
1255 if (!bh_uptodate_or_lock(bh)) {
1256 if (ext4_read_bh(bh, 0, NULL) < 0) {
1265 static int ext4_set_bitmap_checksums(struct super_block *sb,
1267 struct ext4_group_desc *gdp,
1268 struct ext4_new_group_data *group_data)
1270 struct buffer_head *bh;
1272 if (!ext4_has_metadata_csum(sb))
1275 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1278 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1279 EXT4_INODES_PER_GROUP(sb) / 8);
1282 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1285 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1292 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1294 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1295 struct ext4_new_flex_group_data *flex_gd)
1297 struct ext4_new_group_data *group_data = flex_gd->groups;
1298 struct ext4_group_desc *gdp;
1299 struct ext4_sb_info *sbi = EXT4_SB(sb);
1300 struct buffer_head *gdb_bh;
1302 __u16 *bg_flags = flex_gd->bg_flags;
1303 int i, gdb_off, gdb_num, err = 0;
1306 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1307 group = group_data->group;
1309 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1310 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1313 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1315 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1316 /* Update group descriptor block for new group */
1317 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1318 gdb_off * EXT4_DESC_SIZE(sb));
1320 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1321 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1322 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1323 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1325 ext4_std_error(sb, err);
1329 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1330 ext4_free_group_clusters_set(sb, gdp,
1331 group_data->free_clusters_count);
1332 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1333 if (ext4_has_group_desc_csum(sb))
1334 ext4_itable_unused_set(sb, gdp,
1335 EXT4_INODES_PER_GROUP(sb));
1336 gdp->bg_flags = cpu_to_le16(*bg_flags);
1337 ext4_group_desc_csum_set(sb, group, gdp);
1339 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1340 if (unlikely(err)) {
1341 ext4_std_error(sb, err);
1346 * We can allocate memory for mb_alloc based on the new group
1349 err = ext4_mb_add_groupinfo(sb, group, gdp);
1357 * ext4_update_super() updates the super block so that the newly added
1358 * groups can be seen by the filesystem.
1361 * @flex_gd: new added groups
1363 static void ext4_update_super(struct super_block *sb,
1364 struct ext4_new_flex_group_data *flex_gd)
1366 ext4_fsblk_t blocks_count = 0;
1367 ext4_fsblk_t free_blocks = 0;
1368 ext4_fsblk_t reserved_blocks = 0;
1369 struct ext4_new_group_data *group_data = flex_gd->groups;
1370 struct ext4_sb_info *sbi = EXT4_SB(sb);
1371 struct ext4_super_block *es = sbi->s_es;
1374 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1376 * Make the new blocks and inodes valid next. We do this before
1377 * increasing the group count so that once the group is enabled,
1378 * all of its blocks and inodes are already valid.
1380 * We always allocate group-by-group, then block-by-block or
1381 * inode-by-inode within a group, so enabling these
1382 * blocks/inodes before the group is live won't actually let us
1383 * allocate the new space yet.
1385 for (i = 0; i < flex_gd->count; i++) {
1386 blocks_count += group_data[i].blocks_count;
1387 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1390 reserved_blocks = ext4_r_blocks_count(es) * 100;
1391 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1392 reserved_blocks *= blocks_count;
1393 do_div(reserved_blocks, 100);
1395 lock_buffer(sbi->s_sbh);
1396 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1397 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1398 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1400 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1403 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1405 * We need to protect s_groups_count against other CPUs seeing
1406 * inconsistent state in the superblock.
1408 * The precise rules we use are:
1410 * * Writers must perform a smp_wmb() after updating all
1411 * dependent data and before modifying the groups count
1413 * * Readers must perform an smp_rmb() after reading the groups
1414 * count and before reading any dependent data.
1416 * NB. These rules can be relaxed when checking the group count
1417 * while freeing data, as we can only allocate from a block
1418 * group after serialising against the group count, and we can
1419 * only then free after serialising in turn against that
1424 /* Update the global fs size fields */
1425 sbi->s_groups_count += flex_gd->count;
1426 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1427 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1429 /* Update the reserved block counts only once the new group is
1431 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1433 ext4_superblock_csum_set(sb);
1434 unlock_buffer(sbi->s_sbh);
1436 /* Update the free space counts */
1437 percpu_counter_add(&sbi->s_freeclusters_counter,
1438 EXT4_NUM_B2C(sbi, free_blocks));
1439 percpu_counter_add(&sbi->s_freeinodes_counter,
1440 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1442 ext4_debug("free blocks count %llu",
1443 percpu_counter_read(&sbi->s_freeclusters_counter));
1444 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1445 ext4_group_t flex_group;
1446 struct flex_groups *fg;
1448 flex_group = ext4_flex_group(sbi, group_data[0].group);
1449 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1450 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1451 &fg->free_clusters);
1452 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1457 * Update the fs overhead information
1459 ext4_calculate_overhead(sb);
1461 if (test_opt(sb, DEBUG))
1462 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1463 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1464 blocks_count, free_blocks, reserved_blocks);
1467 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1468 * _before_ we start modifying the filesystem, because we cannot abort the
1469 * transaction and not have it write the data to disk.
1471 static int ext4_flex_group_add(struct super_block *sb,
1472 struct inode *resize_inode,
1473 struct ext4_new_flex_group_data *flex_gd)
1475 struct ext4_sb_info *sbi = EXT4_SB(sb);
1476 struct ext4_super_block *es = sbi->s_es;
1477 ext4_fsblk_t o_blocks_count;
1481 unsigned reserved_gdb;
1482 int err = 0, err2 = 0, credit;
1484 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1486 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1487 o_blocks_count = ext4_blocks_count(es);
1488 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1491 err = setup_new_flex_group_blocks(sb, flex_gd);
1495 * We will always be modifying at least the superblock and GDT
1496 * blocks. If we are adding a group past the last current GDT block,
1497 * we will also modify the inode and the dindirect block. If we
1498 * are adding a group with superblock/GDT backups we will also
1499 * modify each of the reserved GDT dindirect blocks.
1501 credit = 3; /* sb, resize inode, resize inode dindirect */
1503 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1504 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1505 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1506 if (IS_ERR(handle)) {
1507 err = PTR_ERR(handle);
1511 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1512 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1516 group = flex_gd->groups[0].group;
1517 BUG_ON(group != sbi->s_groups_count);
1518 err = ext4_add_new_descs(handle, sb, group,
1519 resize_inode, flex_gd->count);
1523 err = ext4_setup_new_descs(handle, sb, flex_gd);
1527 ext4_update_super(sb, flex_gd);
1529 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1532 err2 = ext4_journal_stop(handle);
1537 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1538 int gdb_num_end = ((group + flex_gd->count - 1) /
1539 EXT4_DESC_PER_BLOCK(sb));
1540 int meta_bg = ext4_has_feature_meta_bg(sb);
1541 sector_t old_gdb = 0;
1543 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1544 sizeof(struct ext4_super_block), 0);
1545 for (; gdb_num <= gdb_num_end; gdb_num++) {
1546 struct buffer_head *gdb_bh;
1548 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1550 if (old_gdb == gdb_bh->b_blocknr)
1552 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1553 gdb_bh->b_size, meta_bg);
1554 old_gdb = gdb_bh->b_blocknr;
1561 static int ext4_setup_next_flex_gd(struct super_block *sb,
1562 struct ext4_new_flex_group_data *flex_gd,
1563 ext4_fsblk_t n_blocks_count,
1564 unsigned long flexbg_size)
1566 struct ext4_sb_info *sbi = EXT4_SB(sb);
1567 struct ext4_super_block *es = sbi->s_es;
1568 struct ext4_new_group_data *group_data = flex_gd->groups;
1569 ext4_fsblk_t o_blocks_count;
1570 ext4_group_t n_group;
1572 ext4_group_t last_group;
1574 ext4_grpblk_t clusters_per_group;
1577 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1579 o_blocks_count = ext4_blocks_count(es);
1581 if (o_blocks_count == n_blocks_count)
1584 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1586 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1588 last_group = group | (flexbg_size - 1);
1589 if (last_group > n_group)
1590 last_group = n_group;
1592 flex_gd->count = last_group - group + 1;
1594 for (i = 0; i < flex_gd->count; i++) {
1597 group_data[i].group = group + i;
1598 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1599 overhead = ext4_group_overhead_blocks(sb, group + i);
1600 group_data[i].mdata_blocks = overhead;
1601 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1602 if (ext4_has_group_desc_csum(sb)) {
1603 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1604 EXT4_BG_INODE_UNINIT;
1605 if (!test_opt(sb, INIT_INODE_TABLE))
1606 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1608 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1611 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1612 /* We need to initialize block bitmap of last group. */
1613 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1615 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1616 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1617 group_data[i - 1].free_clusters_count -= clusters_per_group -
1624 /* Add group descriptor data to an existing or new group descriptor block.
1625 * Ensure we handle all possible error conditions _before_ we start modifying
1626 * the filesystem, because we cannot abort the transaction and not have it
1627 * write the data to disk.
1629 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1630 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1632 * We only need to hold the superblock lock while we are actually adding
1633 * in the new group's counts to the superblock. Prior to that we have
1634 * not really "added" the group at all. We re-check that we are still
1635 * adding in the last group in case things have changed since verifying.
1637 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1639 struct ext4_new_flex_group_data flex_gd;
1640 struct ext4_sb_info *sbi = EXT4_SB(sb);
1641 struct ext4_super_block *es = sbi->s_es;
1642 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1643 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1644 struct inode *inode = NULL;
1649 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1651 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1652 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1656 if (ext4_blocks_count(es) + input->blocks_count <
1657 ext4_blocks_count(es)) {
1658 ext4_warning(sb, "blocks_count overflow");
1662 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1663 le32_to_cpu(es->s_inodes_count)) {
1664 ext4_warning(sb, "inodes_count overflow");
1668 if (reserved_gdb || gdb_off == 0) {
1669 if (!ext4_has_feature_resize_inode(sb) ||
1670 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1672 "No reserved GDT blocks, can't resize");
1675 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1676 if (IS_ERR(inode)) {
1677 ext4_warning(sb, "Error opening resize inode");
1678 return PTR_ERR(inode);
1683 err = verify_group_input(sb, input);
1687 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1691 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1696 flex_gd.groups = input;
1697 flex_gd.bg_flags = &bg_flags;
1698 err = ext4_flex_group_add(sb, inode, &flex_gd);
1702 } /* ext4_group_add */
1705 * extend a group without checking assuming that checking has been done.
1707 static int ext4_group_extend_no_check(struct super_block *sb,
1708 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1710 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1714 /* We will update the superblock, one block bitmap, and
1715 * one group descriptor via ext4_group_add_blocks().
1717 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1718 if (IS_ERR(handle)) {
1719 err = PTR_ERR(handle);
1720 ext4_warning(sb, "error %d on journal start", err);
1724 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1725 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1727 ext4_warning(sb, "error %d on journal write access", err);
1731 lock_buffer(EXT4_SB(sb)->s_sbh);
1732 ext4_blocks_count_set(es, o_blocks_count + add);
1733 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1734 ext4_superblock_csum_set(sb);
1735 unlock_buffer(EXT4_SB(sb)->s_sbh);
1736 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1737 o_blocks_count + add);
1738 /* We add the blocks to the bitmap and set the group need init bit */
1739 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1742 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1743 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1744 o_blocks_count + add);
1746 err2 = ext4_journal_stop(handle);
1751 if (test_opt(sb, DEBUG))
1752 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1753 "blocks\n", ext4_blocks_count(es));
1754 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1755 (char *)es, sizeof(struct ext4_super_block), 0);
1761 * Extend the filesystem to the new number of blocks specified. This entry
1762 * point is only used to extend the current filesystem to the end of the last
1763 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1764 * for emergencies (because it has no dependencies on reserved blocks).
1766 * If we _really_ wanted, we could use default values to call ext4_group_add()
1767 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1768 * GDT blocks are reserved to grow to the desired size.
1770 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1771 ext4_fsblk_t n_blocks_count)
1773 ext4_fsblk_t o_blocks_count;
1776 struct buffer_head *bh;
1780 o_blocks_count = ext4_blocks_count(es);
1782 if (test_opt(sb, DEBUG))
1783 ext4_msg(sb, KERN_DEBUG,
1784 "extending last group from %llu to %llu blocks",
1785 o_blocks_count, n_blocks_count);
1787 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1790 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1791 ext4_msg(sb, KERN_ERR,
1792 "filesystem too large to resize to %llu blocks safely",
1797 if (n_blocks_count < o_blocks_count) {
1798 ext4_warning(sb, "can't shrink FS - resize aborted");
1802 /* Handle the remaining blocks in the last group only. */
1803 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1806 ext4_warning(sb, "need to use ext2online to resize further");
1810 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1812 if (o_blocks_count + add < o_blocks_count) {
1813 ext4_warning(sb, "blocks_count overflow");
1817 if (o_blocks_count + add > n_blocks_count)
1818 add = n_blocks_count - o_blocks_count;
1820 if (o_blocks_count + add < n_blocks_count)
1821 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1822 o_blocks_count + add, add);
1824 /* See if the device is actually as big as what was requested */
1825 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1827 ext4_warning(sb, "can't read last block, resize aborted");
1832 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1834 } /* ext4_group_extend */
1837 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1839 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1843 * Release the resize inode and drop the resize_inode feature if there
1844 * are no more reserved gdt blocks, and then convert the file system
1847 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1850 struct ext4_sb_info *sbi = EXT4_SB(sb);
1851 struct ext4_super_block *es = sbi->s_es;
1852 struct ext4_inode_info *ei = EXT4_I(inode);
1854 int i, ret, err = 0;
1857 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1859 if (es->s_reserved_gdt_blocks) {
1860 ext4_error(sb, "Unexpected non-zero "
1861 "s_reserved_gdt_blocks");
1865 /* Do a quick sanity check of the resize inode */
1866 if (inode->i_blocks != 1 << (inode->i_blkbits -
1867 (9 - sbi->s_cluster_bits)))
1868 goto invalid_resize_inode;
1869 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1870 if (i == EXT4_DIND_BLOCK) {
1874 goto invalid_resize_inode;
1877 goto invalid_resize_inode;
1879 credits += 3; /* block bitmap, bg descriptor, resize inode */
1882 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1884 return PTR_ERR(handle);
1886 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1887 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1891 lock_buffer(sbi->s_sbh);
1892 ext4_clear_feature_resize_inode(sb);
1893 ext4_set_feature_meta_bg(sb);
1894 sbi->s_es->s_first_meta_bg =
1895 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1896 ext4_superblock_csum_set(sb);
1897 unlock_buffer(sbi->s_sbh);
1899 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1901 ext4_std_error(sb, err);
1906 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1907 ext4_free_blocks(handle, inode, NULL, nr, 1,
1908 EXT4_FREE_BLOCKS_METADATA |
1909 EXT4_FREE_BLOCKS_FORGET);
1910 ei->i_data[EXT4_DIND_BLOCK] = 0;
1911 inode->i_blocks = 0;
1913 err = ext4_mark_inode_dirty(handle, inode);
1915 ext4_std_error(sb, err);
1919 ret = ext4_journal_stop(handle);
1924 invalid_resize_inode:
1925 ext4_error(sb, "corrupted/inconsistent resize inode");
1930 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1932 * @sb: super block of the fs to be resized
1933 * @n_blocks_count: the number of blocks resides in the resized fs
1935 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1937 struct ext4_new_flex_group_data *flex_gd = NULL;
1938 struct ext4_sb_info *sbi = EXT4_SB(sb);
1939 struct ext4_super_block *es = sbi->s_es;
1940 struct buffer_head *bh;
1941 struct inode *resize_inode = NULL;
1942 ext4_grpblk_t add, offset;
1943 unsigned long n_desc_blocks;
1944 unsigned long o_desc_blocks;
1945 ext4_group_t o_group;
1946 ext4_group_t n_group;
1947 ext4_fsblk_t o_blocks_count;
1948 ext4_fsblk_t n_blocks_count_retry = 0;
1949 unsigned long last_update_time = 0;
1950 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1953 /* See if the device is actually as big as what was requested */
1954 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1956 ext4_warning(sb, "can't read last block, resize aborted");
1962 o_blocks_count = ext4_blocks_count(es);
1964 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1965 "to %llu blocks", o_blocks_count, n_blocks_count);
1967 if (n_blocks_count < o_blocks_count) {
1968 /* On-line shrinking not supported */
1969 ext4_warning(sb, "can't shrink FS - resize aborted");
1973 if (n_blocks_count == o_blocks_count)
1974 /* Nothing need to do */
1977 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1978 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1979 ext4_warning(sb, "resize would cause inodes_count overflow");
1982 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1984 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1985 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1987 meta_bg = ext4_has_feature_meta_bg(sb);
1989 if (ext4_has_feature_resize_inode(sb)) {
1991 ext4_error(sb, "resize_inode and meta_bg enabled "
1995 if (n_desc_blocks > o_desc_blocks +
1996 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1997 n_blocks_count_retry = n_blocks_count;
1998 n_desc_blocks = o_desc_blocks +
1999 le16_to_cpu(es->s_reserved_gdt_blocks);
2000 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2001 n_blocks_count = (ext4_fsblk_t)n_group *
2002 EXT4_BLOCKS_PER_GROUP(sb) +
2003 le32_to_cpu(es->s_first_data_block);
2004 n_group--; /* set to last group number */
2008 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2010 if (IS_ERR(resize_inode)) {
2011 ext4_warning(sb, "Error opening resize inode");
2012 return PTR_ERR(resize_inode);
2016 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2017 err = ext4_convert_meta_bg(sb, resize_inode);
2022 resize_inode = NULL;
2024 if (n_blocks_count_retry) {
2025 n_blocks_count = n_blocks_count_retry;
2026 n_blocks_count_retry = 0;
2032 * Make sure the last group has enough space so that it's
2033 * guaranteed to have enough space for all metadata blocks
2034 * that it might need to hold. (We might not need to store
2035 * the inode table blocks in the last block group, but there
2036 * will be cases where this might be needed.)
2038 if ((ext4_group_first_block_no(sb, n_group) +
2039 ext4_group_overhead_blocks(sb, n_group) + 2 +
2040 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2041 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2043 n_blocks_count_retry = 0;
2046 resize_inode = NULL;
2051 /* extend the last group */
2052 if (n_group == o_group)
2053 add = n_blocks_count - o_blocks_count;
2055 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2057 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2062 if (ext4_blocks_count(es) == n_blocks_count)
2065 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2069 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2073 flex_gd = alloc_flex_gd(flexbg_size);
2074 if (flex_gd == NULL) {
2079 /* Add flex groups. Note that a regular group is a
2080 * flex group with 1 group.
2082 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2084 if (jiffies - last_update_time > HZ * 10) {
2085 if (last_update_time)
2086 ext4_msg(sb, KERN_INFO,
2087 "resized to %llu blocks",
2088 ext4_blocks_count(es));
2089 last_update_time = jiffies;
2091 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2093 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2098 if (!err && n_blocks_count_retry) {
2099 n_blocks_count = n_blocks_count_retry;
2100 n_blocks_count_retry = 0;
2101 free_flex_gd(flex_gd);
2105 resize_inode = NULL;
2112 free_flex_gd(flex_gd);
2113 if (resize_inode != NULL)
2116 ext4_warning(sb, "error (%d) occurred during "
2117 "file system resize", err);
2118 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2119 ext4_blocks_count(es));