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 the reserved GDT blocks is non-zero, the resize_inode feature
57 * should always be set.
59 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
60 !ext4_has_feature_resize_inode(sb)) {
61 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
66 * If we are not using the primary superblock/GDT copy don't resize,
67 * because the user tools have no way of handling this. Probably a
68 * bad time to do it anyways.
70 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
71 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
72 ext4_warning(sb, "won't resize using backup superblock at %llu",
73 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
78 * We are not allowed to do online-resizing on a filesystem mounted
79 * with error, because it can destroy the filesystem easily.
81 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
82 ext4_warning(sb, "There are errors in the filesystem, "
83 "so online resizing is not allowed");
87 if (ext4_has_feature_sparse_super2(sb)) {
88 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
92 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
93 &EXT4_SB(sb)->s_ext4_flags))
99 void ext4_resize_end(struct super_block *sb)
101 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
102 smp_mb__after_atomic();
105 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
106 ext4_group_t group) {
107 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
108 EXT4_DESC_PER_BLOCK_BITS(sb);
111 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
112 ext4_group_t group) {
113 group = ext4_meta_bg_first_group(sb, group);
114 return ext4_group_first_block_no(sb, group);
117 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
118 ext4_group_t group) {
119 ext4_grpblk_t overhead;
120 overhead = ext4_bg_num_gdb(sb, group);
121 if (ext4_bg_has_super(sb, group))
123 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
127 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
128 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
130 static int verify_group_input(struct super_block *sb,
131 struct ext4_new_group_data *input)
133 struct ext4_sb_info *sbi = EXT4_SB(sb);
134 struct ext4_super_block *es = sbi->s_es;
135 ext4_fsblk_t start = ext4_blocks_count(es);
136 ext4_fsblk_t end = start + input->blocks_count;
137 ext4_group_t group = input->group;
138 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
140 ext4_fsblk_t metaend;
141 struct buffer_head *bh = NULL;
142 ext4_grpblk_t free_blocks_count, offset;
145 if (group != sbi->s_groups_count) {
146 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
147 input->group, sbi->s_groups_count);
151 overhead = ext4_group_overhead_blocks(sb, group);
152 metaend = start + overhead;
153 input->free_clusters_count = free_blocks_count =
154 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
156 if (test_opt(sb, DEBUG))
157 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
158 "(%d free, %u reserved)\n",
159 ext4_bg_has_super(sb, input->group) ? "normal" :
160 "no-super", input->group, input->blocks_count,
161 free_blocks_count, input->reserved_blocks);
163 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
165 ext4_warning(sb, "Last group not full");
166 else if (input->reserved_blocks > input->blocks_count / 5)
167 ext4_warning(sb, "Reserved blocks too high (%u)",
168 input->reserved_blocks);
169 else if (free_blocks_count < 0)
170 ext4_warning(sb, "Bad blocks count %u",
171 input->blocks_count);
172 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
175 ext4_warning(sb, "Cannot read last block (%llu)",
177 } else if (outside(input->block_bitmap, start, end))
178 ext4_warning(sb, "Block bitmap not in group (block %llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (outside(input->inode_bitmap, start, end))
181 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
182 (unsigned long long)input->inode_bitmap);
183 else if (outside(input->inode_table, start, end) ||
184 outside(itend - 1, start, end))
185 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
186 (unsigned long long)input->inode_table, itend - 1);
187 else if (input->inode_bitmap == input->block_bitmap)
188 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
189 (unsigned long long)input->block_bitmap);
190 else if (inside(input->block_bitmap, input->inode_table, itend))
191 ext4_warning(sb, "Block bitmap (%llu) in inode table "
193 (unsigned long long)input->block_bitmap,
194 (unsigned long long)input->inode_table, itend - 1);
195 else if (inside(input->inode_bitmap, input->inode_table, itend))
196 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
198 (unsigned long long)input->inode_bitmap,
199 (unsigned long long)input->inode_table, itend - 1);
200 else if (inside(input->block_bitmap, start, metaend))
201 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
202 (unsigned long long)input->block_bitmap,
204 else if (inside(input->inode_bitmap, start, metaend))
205 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
206 (unsigned long long)input->inode_bitmap,
208 else if (inside(input->inode_table, start, metaend) ||
209 inside(itend - 1, start, metaend))
210 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
212 (unsigned long long)input->inode_table,
213 itend - 1, start, metaend - 1);
222 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
225 struct ext4_new_flex_group_data {
226 struct ext4_new_group_data *groups; /* new_group_data for groups
228 __u16 *bg_flags; /* block group flags of groups
230 ext4_group_t count; /* number of groups in @groups
235 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
238 * Returns NULL on failure otherwise address of the allocated structure.
240 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
242 struct ext4_new_flex_group_data *flex_gd;
244 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
248 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
250 flex_gd->count = flexbg_size;
252 flex_gd->groups = kmalloc_array(flexbg_size,
253 sizeof(struct ext4_new_group_data),
255 if (flex_gd->groups == NULL)
258 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
260 if (flex_gd->bg_flags == NULL)
266 kfree(flex_gd->groups);
273 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
275 kfree(flex_gd->bg_flags);
276 kfree(flex_gd->groups);
281 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
282 * and inode tables for a flex group.
284 * This function is used by 64bit-resize. Note that this function allocates
285 * group tables from the 1st group of groups contained by @flexgd, which may
286 * be a partial of a flex group.
288 * @sb: super block of fs to which the groups belongs
290 * Returns 0 on a successful allocation of the metadata blocks in the
293 static int ext4_alloc_group_tables(struct super_block *sb,
294 struct ext4_new_flex_group_data *flex_gd,
297 struct ext4_new_group_data *group_data = flex_gd->groups;
298 ext4_fsblk_t start_blk;
299 ext4_fsblk_t last_blk;
300 ext4_group_t src_group;
301 ext4_group_t bb_index = 0;
302 ext4_group_t ib_index = 0;
303 ext4_group_t it_index = 0;
305 ext4_group_t last_group;
307 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
310 BUG_ON(flex_gd->count == 0 || group_data == NULL);
312 src_group = group_data[0].group;
313 last_group = src_group + flex_gd->count - 1;
315 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
316 (last_group & ~(flexbg_size - 1))));
318 group = group_data[0].group;
319 if (src_group >= group_data[0].group + flex_gd->count)
321 start_blk = ext4_group_first_block_no(sb, src_group);
322 last_blk = start_blk + group_data[src_group - group].blocks_count;
324 overhead = ext4_group_overhead_blocks(sb, src_group);
326 start_blk += overhead;
328 /* We collect contiguous blocks as much as possible. */
330 for (; src_group <= last_group; src_group++) {
331 overhead = ext4_group_overhead_blocks(sb, src_group);
333 last_blk += group_data[src_group - group].blocks_count;
338 /* Allocate block bitmaps */
339 for (; bb_index < flex_gd->count; bb_index++) {
340 if (start_blk >= last_blk)
342 group_data[bb_index].block_bitmap = start_blk++;
343 group = ext4_get_group_number(sb, start_blk - 1);
344 group -= group_data[0].group;
345 group_data[group].mdata_blocks++;
346 flex_gd->bg_flags[group] &= uninit_mask;
349 /* Allocate inode bitmaps */
350 for (; ib_index < flex_gd->count; ib_index++) {
351 if (start_blk >= last_blk)
353 group_data[ib_index].inode_bitmap = start_blk++;
354 group = ext4_get_group_number(sb, start_blk - 1);
355 group -= group_data[0].group;
356 group_data[group].mdata_blocks++;
357 flex_gd->bg_flags[group] &= uninit_mask;
360 /* Allocate inode tables */
361 for (; it_index < flex_gd->count; it_index++) {
362 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
363 ext4_fsblk_t next_group_start;
365 if (start_blk + itb > last_blk)
367 group_data[it_index].inode_table = start_blk;
368 group = ext4_get_group_number(sb, start_blk);
369 next_group_start = ext4_group_first_block_no(sb, group + 1);
370 group -= group_data[0].group;
372 if (start_blk + itb > next_group_start) {
373 flex_gd->bg_flags[group + 1] &= uninit_mask;
374 overhead = start_blk + itb - next_group_start;
375 group_data[group + 1].mdata_blocks += overhead;
379 group_data[group].mdata_blocks += itb;
380 flex_gd->bg_flags[group] &= uninit_mask;
381 start_blk += EXT4_SB(sb)->s_itb_per_group;
384 /* Update free clusters count to exclude metadata blocks */
385 for (i = 0; i < flex_gd->count; i++) {
386 group_data[i].free_clusters_count -=
387 EXT4_NUM_B2C(EXT4_SB(sb),
388 group_data[i].mdata_blocks);
391 if (test_opt(sb, DEBUG)) {
393 group = group_data[0].group;
395 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
396 "%d groups, flexbg size is %d:\n", flex_gd->count,
399 for (i = 0; i < flex_gd->count; i++) {
401 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
402 ext4_bg_has_super(sb, group + i) ? "normal" :
403 "no-super", group + i,
404 group_data[i].blocks_count,
405 group_data[i].free_clusters_count,
406 group_data[i].mdata_blocks);
412 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
415 struct buffer_head *bh;
418 bh = sb_getblk(sb, blk);
420 return ERR_PTR(-ENOMEM);
421 BUFFER_TRACE(bh, "get_write_access");
422 if ((err = ext4_journal_get_write_access(handle, bh))) {
426 memset(bh->b_data, 0, sb->s_blocksize);
427 set_buffer_uptodate(bh);
433 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
435 return ext4_journal_ensure_credits_fn(handle, credits,
436 EXT4_MAX_TRANS_DATA, 0, 0);
440 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
442 * Helper function for ext4_setup_new_group_blocks() which set .
445 * @handle: journal handle
446 * @flex_gd: flex group data
448 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
449 struct ext4_new_flex_group_data *flex_gd,
450 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
452 struct ext4_sb_info *sbi = EXT4_SB(sb);
453 ext4_group_t count = last_cluster - first_cluster + 1;
456 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
458 for (count2 = count; count > 0;
459 count -= count2, first_cluster += count2) {
461 struct buffer_head *bh;
465 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
466 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
467 group -= flex_gd->groups[0].group;
469 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
473 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
474 BUG_ON(flex_gd->count > 1);
478 err = ext4_resize_ensure_credits_batch(handle, 1);
482 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
486 BUFFER_TRACE(bh, "get_write_access");
487 err = ext4_journal_get_write_access(handle, bh);
492 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
493 first_cluster, first_cluster - start, count2);
494 ext4_set_bits(bh->b_data, first_cluster - start, count2);
496 err = ext4_handle_dirty_metadata(handle, NULL, bh);
506 * Set up the block and inode bitmaps, and the inode table for the new groups.
507 * This doesn't need to be part of the main transaction, since we are only
508 * changing blocks outside the actual filesystem. We still do journaling to
509 * ensure the recovery is correct in case of a failure just after resize.
510 * If any part of this fails, we simply abort the resize.
512 * setup_new_flex_group_blocks handles a flex group as follow:
513 * 1. copy super block and GDT, and initialize group tables if necessary.
514 * In this step, we only set bits in blocks bitmaps for blocks taken by
515 * super block and GDT.
516 * 2. allocate group tables in block bitmaps, that is, set bits in block
517 * bitmap for blocks taken by group tables.
519 static int setup_new_flex_group_blocks(struct super_block *sb,
520 struct ext4_new_flex_group_data *flex_gd)
522 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
525 struct ext4_sb_info *sbi = EXT4_SB(sb);
526 struct ext4_super_block *es = sbi->s_es;
527 struct ext4_new_group_data *group_data = flex_gd->groups;
528 __u16 *bg_flags = flex_gd->bg_flags;
530 ext4_group_t group, count;
531 struct buffer_head *bh = NULL;
532 int reserved_gdb, i, j, err = 0, err2;
535 BUG_ON(!flex_gd->count || !group_data ||
536 group_data[0].group != sbi->s_groups_count);
538 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
539 meta_bg = ext4_has_feature_meta_bg(sb);
541 /* This transaction may be extended/restarted along the way */
542 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
544 return PTR_ERR(handle);
546 group = group_data[0].group;
547 for (i = 0; i < flex_gd->count; i++, group++) {
548 unsigned long gdblocks;
549 ext4_grpblk_t overhead;
551 gdblocks = ext4_bg_num_gdb(sb, group);
552 start = ext4_group_first_block_no(sb, group);
554 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
558 ext4_group_t first_group;
559 first_group = ext4_meta_bg_first_group(sb, group);
560 if (first_group != group + 1 &&
561 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
565 block = start + ext4_bg_has_super(sb, group);
566 /* Copy all of the GDT blocks into the backup in this group */
567 for (j = 0; j < gdblocks; j++, block++) {
568 struct buffer_head *gdb;
570 ext4_debug("update backup group %#04llx\n", block);
571 err = ext4_resize_ensure_credits_batch(handle, 1);
575 gdb = sb_getblk(sb, block);
576 if (unlikely(!gdb)) {
581 BUFFER_TRACE(gdb, "get_write_access");
582 err = ext4_journal_get_write_access(handle, gdb);
587 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
588 s_group_desc, j)->b_data, gdb->b_size);
589 set_buffer_uptodate(gdb);
591 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
599 /* Zero out all of the reserved backup group descriptor
602 if (ext4_bg_has_super(sb, group)) {
603 err = sb_issue_zeroout(sb, gdblocks + start + 1,
604 reserved_gdb, GFP_NOFS);
610 /* Initialize group tables of the grop @group */
611 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
614 /* Zero out all of the inode table blocks */
615 block = group_data[i].inode_table;
616 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
617 block, sbi->s_itb_per_group);
618 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
624 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
627 /* Initialize block bitmap of the @group */
628 block = group_data[i].block_bitmap;
629 err = ext4_resize_ensure_credits_batch(handle, 1);
633 bh = bclean(handle, sb, block);
638 overhead = ext4_group_overhead_blocks(sb, group);
640 ext4_debug("mark backup superblock %#04llx (+0)\n",
642 ext4_set_bits(bh->b_data, 0,
643 EXT4_NUM_B2C(sbi, overhead));
645 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
646 sb->s_blocksize * 8, bh->b_data);
647 err = ext4_handle_dirty_metadata(handle, NULL, bh);
653 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
656 /* Initialize inode bitmap of the @group */
657 block = group_data[i].inode_bitmap;
658 err = ext4_resize_ensure_credits_batch(handle, 1);
661 /* Mark unused entries in inode bitmap used */
662 bh = bclean(handle, sb, block);
668 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
669 sb->s_blocksize * 8, bh->b_data);
670 err = ext4_handle_dirty_metadata(handle, NULL, bh);
676 /* Mark group tables in block bitmap */
677 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
678 count = group_table_count[j];
679 start = (&group_data[0].block_bitmap)[j];
681 for (i = 1; i < flex_gd->count; i++) {
682 block += group_table_count[j];
683 if (block == (&group_data[i].block_bitmap)[j]) {
684 count += group_table_count[j];
687 err = set_flexbg_block_bitmap(sb, handle,
689 EXT4_B2C(sbi, start),
695 count = group_table_count[j];
696 start = (&group_data[i].block_bitmap)[j];
701 err = set_flexbg_block_bitmap(sb, handle,
703 EXT4_B2C(sbi, start),
713 err2 = ext4_journal_stop(handle);
721 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
722 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
723 * calling this for the first time. In a sparse filesystem it will be the
724 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
725 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
727 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
728 unsigned *five, unsigned *seven)
730 unsigned *min = three;
734 if (!ext4_has_feature_sparse_super(sb)) {
756 * Check that all of the backup GDT blocks are held in the primary GDT block.
757 * It is assumed that they are stored in group order. Returns the number of
758 * groups in current filesystem that have BACKUPS, or -ve error code.
760 static int verify_reserved_gdb(struct super_block *sb,
762 struct buffer_head *primary)
764 const ext4_fsblk_t blk = primary->b_blocknr;
769 __le32 *p = (__le32 *)primary->b_data;
772 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
773 if (le32_to_cpu(*p++) !=
774 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
775 ext4_warning(sb, "reserved GDT %llu"
776 " missing grp %d (%llu)",
779 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
783 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
791 * Called when we need to bring a reserved group descriptor table block into
792 * use from the resize inode. The primary copy of the new GDT block currently
793 * is an indirect block (under the double indirect block in the resize inode).
794 * The new backup GDT blocks will be stored as leaf blocks in this indirect
795 * block, in group order. Even though we know all the block numbers we need,
796 * we check to ensure that the resize inode has actually reserved these blocks.
798 * Don't need to update the block bitmaps because the blocks are still in use.
800 * We get all of the error cases out of the way, so that we are sure to not
801 * fail once we start modifying the data on disk, because JBD has no rollback.
803 static int add_new_gdb(handle_t *handle, struct inode *inode,
806 struct super_block *sb = inode->i_sb;
807 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
808 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
809 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
810 struct buffer_head **o_group_desc, **n_group_desc = NULL;
811 struct buffer_head *dind = NULL;
812 struct buffer_head *gdb_bh = NULL;
814 struct ext4_iloc iloc = { .bh = NULL };
818 if (test_opt(sb, DEBUG))
820 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
823 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
825 return PTR_ERR(gdb_bh);
827 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
833 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
834 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
841 data = (__le32 *)dind->b_data;
842 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
843 ext4_warning(sb, "new group %u GDT block %llu not reserved",
849 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
850 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
854 BUFFER_TRACE(gdb_bh, "get_write_access");
855 err = ext4_journal_get_write_access(handle, gdb_bh);
859 BUFFER_TRACE(dind, "get_write_access");
860 err = ext4_journal_get_write_access(handle, dind);
862 ext4_std_error(sb, err);
866 /* ext4_reserve_inode_write() gets a reference on the iloc */
867 err = ext4_reserve_inode_write(handle, inode, &iloc);
871 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
875 ext4_warning(sb, "not enough memory for %lu groups",
881 * Finally, we have all of the possible failures behind us...
883 * Remove new GDT block from inode double-indirect block and clear out
884 * the new GDT block for use (which also "frees" the backup GDT blocks
885 * from the reserved inode). We don't need to change the bitmaps for
886 * these blocks, because they are marked as in-use from being in the
887 * reserved inode, and will become GDT blocks (primary and backup).
889 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
890 err = ext4_handle_dirty_metadata(handle, NULL, dind);
892 ext4_std_error(sb, err);
895 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
896 (9 - EXT4_SB(sb)->s_cluster_bits);
897 ext4_mark_iloc_dirty(handle, inode, &iloc);
898 memset(gdb_bh->b_data, 0, sb->s_blocksize);
899 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
901 ext4_std_error(sb, err);
908 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
909 memcpy(n_group_desc, o_group_desc,
910 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
912 n_group_desc[gdb_num] = gdb_bh;
913 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
914 EXT4_SB(sb)->s_gdb_count++;
915 ext4_kvfree_array_rcu(o_group_desc);
917 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
918 err = ext4_handle_dirty_super(handle, sb);
920 ext4_std_error(sb, err);
923 kvfree(n_group_desc);
928 ext4_debug("leaving with error %d\n", err);
933 * add_new_gdb_meta_bg is the sister of add_new_gdb.
935 static int add_new_gdb_meta_bg(struct super_block *sb,
936 handle_t *handle, ext4_group_t group) {
937 ext4_fsblk_t gdblock;
938 struct buffer_head *gdb_bh;
939 struct buffer_head **o_group_desc, **n_group_desc;
940 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
943 gdblock = ext4_meta_bg_first_block_no(sb, group) +
944 ext4_bg_has_super(sb, group);
945 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
947 return PTR_ERR(gdb_bh);
948 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
953 ext4_warning(sb, "not enough memory for %lu groups",
959 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
960 memcpy(n_group_desc, o_group_desc,
961 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
963 n_group_desc[gdb_num] = gdb_bh;
965 BUFFER_TRACE(gdb_bh, "get_write_access");
966 err = ext4_journal_get_write_access(handle, gdb_bh);
968 kvfree(n_group_desc);
973 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
974 EXT4_SB(sb)->s_gdb_count++;
975 ext4_kvfree_array_rcu(o_group_desc);
980 * Called when we are adding a new group which has a backup copy of each of
981 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
982 * We need to add these reserved backup GDT blocks to the resize inode, so
983 * that they are kept for future resizing and not allocated to files.
985 * Each reserved backup GDT block will go into a different indirect block.
986 * The indirect blocks are actually the primary reserved GDT blocks,
987 * so we know in advance what their block numbers are. We only get the
988 * double-indirect block to verify it is pointing to the primary reserved
989 * GDT blocks so we don't overwrite a data block by accident. The reserved
990 * backup GDT blocks are stored in their reserved primary GDT block.
992 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
995 struct super_block *sb = inode->i_sb;
996 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
997 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
998 struct buffer_head **primary;
999 struct buffer_head *dind;
1000 struct ext4_iloc iloc;
1007 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1011 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1012 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1014 err = PTR_ERR(dind);
1019 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1020 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1021 EXT4_ADDR_PER_BLOCK(sb));
1022 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1024 /* Get each reserved primary GDT block and verify it holds backups */
1025 for (res = 0; res < reserved_gdb; res++, blk++) {
1026 if (le32_to_cpu(*data) != blk) {
1027 ext4_warning(sb, "reserved block %llu"
1028 " not at offset %ld",
1030 (long)(data - (__le32 *)dind->b_data));
1034 primary[res] = ext4_sb_bread(sb, blk, 0);
1035 if (IS_ERR(primary[res])) {
1036 err = PTR_ERR(primary[res]);
1037 primary[res] = NULL;
1040 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1041 if (gdbackups < 0) {
1042 brelse(primary[res]);
1047 data = (__le32 *)dind->b_data;
1050 for (i = 0; i < reserved_gdb; i++) {
1051 BUFFER_TRACE(primary[i], "get_write_access");
1052 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1056 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1060 * Finally we can add each of the reserved backup GDT blocks from
1061 * the new group to its reserved primary GDT block.
1063 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1064 for (i = 0; i < reserved_gdb; i++) {
1066 data = (__le32 *)primary[i]->b_data;
1067 /* printk("reserving backup %lu[%u] = %lu\n",
1068 primary[i]->b_blocknr, gdbackups,
1069 blk + primary[i]->b_blocknr); */
1070 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1071 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1076 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1077 ext4_mark_iloc_dirty(handle, inode, &iloc);
1081 brelse(primary[res]);
1091 * Update the backup copies of the ext4 metadata. These don't need to be part
1092 * of the main resize transaction, because e2fsck will re-write them if there
1093 * is a problem (basically only OOM will cause a problem). However, we
1094 * _should_ update the backups if possible, in case the primary gets trashed
1095 * for some reason and we need to run e2fsck from a backup superblock. The
1096 * important part is that the new block and inode counts are in the backup
1097 * superblocks, and the location of the new group metadata in the GDT backups.
1099 * We do not need take the s_resize_lock for this, because these
1100 * blocks are not otherwise touched by the filesystem code when it is
1101 * mounted. We don't need to worry about last changing from
1102 * sbi->s_groups_count, because the worst that can happen is that we
1103 * do not copy the full number of backups at this time. The resize
1104 * which changed s_groups_count will backup again.
1106 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1107 int size, int meta_bg)
1109 struct ext4_sb_info *sbi = EXT4_SB(sb);
1111 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1115 ext4_group_t group = 0;
1116 int rest = sb->s_blocksize - size;
1120 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1121 if (IS_ERR(handle)) {
1123 err = PTR_ERR(handle);
1128 group = ext4_list_backups(sb, &three, &five, &seven);
1129 last = sbi->s_groups_count;
1131 group = ext4_get_group_number(sb, blk_off) + 1;
1132 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1135 while (group < sbi->s_groups_count) {
1136 struct buffer_head *bh;
1137 ext4_fsblk_t backup_block;
1139 /* Out of journal space, and can't get more - abort - so sad */
1140 err = ext4_resize_ensure_credits_batch(handle, 1);
1145 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1147 backup_block = (ext4_group_first_block_no(sb, group) +
1148 ext4_bg_has_super(sb, group));
1150 bh = sb_getblk(sb, backup_block);
1151 if (unlikely(!bh)) {
1155 ext4_debug("update metadata backup %llu(+%llu)\n",
1156 backup_block, backup_block -
1157 ext4_group_first_block_no(sb, group));
1158 BUFFER_TRACE(bh, "get_write_access");
1159 if ((err = ext4_journal_get_write_access(handle, bh))) {
1164 memcpy(bh->b_data, data, size);
1166 memset(bh->b_data + size, 0, rest);
1167 set_buffer_uptodate(bh);
1169 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1171 ext4_std_error(sb, err);
1175 group = ext4_list_backups(sb, &three, &five, &seven);
1176 else if (group == last)
1181 if ((err2 = ext4_journal_stop(handle)) && !err)
1185 * Ugh! Need to have e2fsck write the backup copies. It is too
1186 * late to revert the resize, we shouldn't fail just because of
1187 * the backup copies (they are only needed in case of corruption).
1189 * However, if we got here we have a journal problem too, so we
1190 * can't really start a transaction to mark the superblock.
1191 * Chicken out and just set the flag on the hope it will be written
1192 * to disk, and if not - we will simply wait until next fsck.
1196 ext4_warning(sb, "can't update backup for group %u (err %d), "
1197 "forcing fsck on next reboot", group, err);
1198 sbi->s_mount_state &= ~EXT4_VALID_FS;
1199 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1200 mark_buffer_dirty(sbi->s_sbh);
1205 * ext4_add_new_descs() adds @count group descriptor of groups
1206 * starting at @group
1208 * @handle: journal handle
1210 * @group: the group no. of the first group desc to be added
1211 * @resize_inode: the resize inode
1212 * @count: number of group descriptors to be added
1214 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1215 ext4_group_t group, struct inode *resize_inode,
1218 struct ext4_sb_info *sbi = EXT4_SB(sb);
1219 struct ext4_super_block *es = sbi->s_es;
1220 struct buffer_head *gdb_bh;
1221 int i, gdb_off, gdb_num, err = 0;
1224 meta_bg = ext4_has_feature_meta_bg(sb);
1225 for (i = 0; i < count; i++, group++) {
1226 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1227 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1229 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1230 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1233 * We will only either add reserved group blocks to a backup group
1234 * or remove reserved blocks for the first group in a new group block.
1235 * Doing both would be mean more complex code, and sane people don't
1236 * use non-sparse filesystems anymore. This is already checked above.
1239 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1241 BUFFER_TRACE(gdb_bh, "get_write_access");
1242 err = ext4_journal_get_write_access(handle, gdb_bh);
1244 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1245 err = reserve_backup_gdb(handle, resize_inode, group);
1246 } else if (meta_bg != 0) {
1247 err = add_new_gdb_meta_bg(sb, handle, group);
1249 err = add_new_gdb(handle, resize_inode, group);
1257 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1259 struct buffer_head *bh = sb_getblk(sb, block);
1262 if (!bh_uptodate_or_lock(bh)) {
1263 if (ext4_read_bh(bh, 0, NULL) < 0) {
1272 static int ext4_set_bitmap_checksums(struct super_block *sb,
1274 struct ext4_group_desc *gdp,
1275 struct ext4_new_group_data *group_data)
1277 struct buffer_head *bh;
1279 if (!ext4_has_metadata_csum(sb))
1282 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1285 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1286 EXT4_INODES_PER_GROUP(sb) / 8);
1289 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1292 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1299 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1301 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1302 struct ext4_new_flex_group_data *flex_gd)
1304 struct ext4_new_group_data *group_data = flex_gd->groups;
1305 struct ext4_group_desc *gdp;
1306 struct ext4_sb_info *sbi = EXT4_SB(sb);
1307 struct buffer_head *gdb_bh;
1309 __u16 *bg_flags = flex_gd->bg_flags;
1310 int i, gdb_off, gdb_num, err = 0;
1313 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1314 group = group_data->group;
1316 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1317 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1320 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1322 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1323 /* Update group descriptor block for new group */
1324 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1325 gdb_off * EXT4_DESC_SIZE(sb));
1327 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1328 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1329 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1330 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1332 ext4_std_error(sb, err);
1336 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1337 ext4_free_group_clusters_set(sb, gdp,
1338 group_data->free_clusters_count);
1339 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1340 if (ext4_has_group_desc_csum(sb))
1341 ext4_itable_unused_set(sb, gdp,
1342 EXT4_INODES_PER_GROUP(sb));
1343 gdp->bg_flags = cpu_to_le16(*bg_flags);
1344 ext4_group_desc_csum_set(sb, group, gdp);
1346 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1347 if (unlikely(err)) {
1348 ext4_std_error(sb, err);
1353 * We can allocate memory for mb_alloc based on the new group
1356 err = ext4_mb_add_groupinfo(sb, group, gdp);
1364 * ext4_update_super() updates the super block so that the newly added
1365 * groups can be seen by the filesystem.
1368 * @flex_gd: new added groups
1370 static void ext4_update_super(struct super_block *sb,
1371 struct ext4_new_flex_group_data *flex_gd)
1373 ext4_fsblk_t blocks_count = 0;
1374 ext4_fsblk_t free_blocks = 0;
1375 ext4_fsblk_t reserved_blocks = 0;
1376 struct ext4_new_group_data *group_data = flex_gd->groups;
1377 struct ext4_sb_info *sbi = EXT4_SB(sb);
1378 struct ext4_super_block *es = sbi->s_es;
1381 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1383 * Make the new blocks and inodes valid next. We do this before
1384 * increasing the group count so that once the group is enabled,
1385 * all of its blocks and inodes are already valid.
1387 * We always allocate group-by-group, then block-by-block or
1388 * inode-by-inode within a group, so enabling these
1389 * blocks/inodes before the group is live won't actually let us
1390 * allocate the new space yet.
1392 for (i = 0; i < flex_gd->count; i++) {
1393 blocks_count += group_data[i].blocks_count;
1394 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1397 reserved_blocks = ext4_r_blocks_count(es) * 100;
1398 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1399 reserved_blocks *= blocks_count;
1400 do_div(reserved_blocks, 100);
1402 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1403 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1404 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1406 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1409 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1411 * We need to protect s_groups_count against other CPUs seeing
1412 * inconsistent state in the superblock.
1414 * The precise rules we use are:
1416 * * Writers must perform a smp_wmb() after updating all
1417 * dependent data and before modifying the groups count
1419 * * Readers must perform an smp_rmb() after reading the groups
1420 * count and before reading any dependent data.
1422 * NB. These rules can be relaxed when checking the group count
1423 * while freeing data, as we can only allocate from a block
1424 * group after serialising against the group count, and we can
1425 * only then free after serialising in turn against that
1430 /* Update the global fs size fields */
1431 sbi->s_groups_count += flex_gd->count;
1432 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1433 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1435 /* Update the reserved block counts only once the new group is
1437 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1440 /* Update the free space counts */
1441 percpu_counter_add(&sbi->s_freeclusters_counter,
1442 EXT4_NUM_B2C(sbi, free_blocks));
1443 percpu_counter_add(&sbi->s_freeinodes_counter,
1444 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1446 ext4_debug("free blocks count %llu",
1447 percpu_counter_read(&sbi->s_freeclusters_counter));
1448 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1449 ext4_group_t flex_group;
1450 struct flex_groups *fg;
1452 flex_group = ext4_flex_group(sbi, group_data[0].group);
1453 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1454 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1455 &fg->free_clusters);
1456 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1461 * Update the fs overhead information
1463 ext4_calculate_overhead(sb);
1464 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1466 if (test_opt(sb, DEBUG))
1467 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1468 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1469 blocks_count, free_blocks, reserved_blocks);
1472 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1473 * _before_ we start modifying the filesystem, because we cannot abort the
1474 * transaction and not have it write the data to disk.
1476 static int ext4_flex_group_add(struct super_block *sb,
1477 struct inode *resize_inode,
1478 struct ext4_new_flex_group_data *flex_gd)
1480 struct ext4_sb_info *sbi = EXT4_SB(sb);
1481 struct ext4_super_block *es = sbi->s_es;
1482 ext4_fsblk_t o_blocks_count;
1486 unsigned reserved_gdb;
1487 int err = 0, err2 = 0, credit;
1489 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1491 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1492 o_blocks_count = ext4_blocks_count(es);
1493 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1496 err = setup_new_flex_group_blocks(sb, flex_gd);
1500 * We will always be modifying at least the superblock and GDT
1501 * blocks. If we are adding a group past the last current GDT block,
1502 * we will also modify the inode and the dindirect block. If we
1503 * are adding a group with superblock/GDT backups we will also
1504 * modify each of the reserved GDT dindirect blocks.
1506 credit = 3; /* sb, resize inode, resize inode dindirect */
1508 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1509 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1510 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1511 if (IS_ERR(handle)) {
1512 err = PTR_ERR(handle);
1516 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1517 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1521 group = flex_gd->groups[0].group;
1522 BUG_ON(group != sbi->s_groups_count);
1523 err = ext4_add_new_descs(handle, sb, group,
1524 resize_inode, flex_gd->count);
1528 err = ext4_setup_new_descs(handle, sb, flex_gd);
1532 ext4_update_super(sb, flex_gd);
1534 err = ext4_handle_dirty_super(handle, sb);
1537 err2 = ext4_journal_stop(handle);
1542 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1543 int gdb_num_end = ((group + flex_gd->count - 1) /
1544 EXT4_DESC_PER_BLOCK(sb));
1545 int meta_bg = ext4_has_feature_meta_bg(sb);
1546 sector_t old_gdb = 0;
1548 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1549 sizeof(struct ext4_super_block), 0);
1550 for (; gdb_num <= gdb_num_end; gdb_num++) {
1551 struct buffer_head *gdb_bh;
1553 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1555 if (old_gdb == gdb_bh->b_blocknr)
1557 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1558 gdb_bh->b_size, meta_bg);
1559 old_gdb = gdb_bh->b_blocknr;
1566 static int ext4_setup_next_flex_gd(struct super_block *sb,
1567 struct ext4_new_flex_group_data *flex_gd,
1568 ext4_fsblk_t n_blocks_count,
1569 unsigned long flexbg_size)
1571 struct ext4_sb_info *sbi = EXT4_SB(sb);
1572 struct ext4_super_block *es = sbi->s_es;
1573 struct ext4_new_group_data *group_data = flex_gd->groups;
1574 ext4_fsblk_t o_blocks_count;
1575 ext4_group_t n_group;
1577 ext4_group_t last_group;
1579 ext4_grpblk_t clusters_per_group;
1582 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1584 o_blocks_count = ext4_blocks_count(es);
1586 if (o_blocks_count == n_blocks_count)
1589 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1591 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1593 last_group = group | (flexbg_size - 1);
1594 if (last_group > n_group)
1595 last_group = n_group;
1597 flex_gd->count = last_group - group + 1;
1599 for (i = 0; i < flex_gd->count; i++) {
1602 group_data[i].group = group + i;
1603 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1604 overhead = ext4_group_overhead_blocks(sb, group + i);
1605 group_data[i].mdata_blocks = overhead;
1606 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1607 if (ext4_has_group_desc_csum(sb)) {
1608 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1609 EXT4_BG_INODE_UNINIT;
1610 if (!test_opt(sb, INIT_INODE_TABLE))
1611 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1613 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1616 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1617 /* We need to initialize block bitmap of last group. */
1618 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1620 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1621 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1622 group_data[i - 1].free_clusters_count -= clusters_per_group -
1629 /* Add group descriptor data to an existing or new group descriptor block.
1630 * Ensure we handle all possible error conditions _before_ we start modifying
1631 * the filesystem, because we cannot abort the transaction and not have it
1632 * write the data to disk.
1634 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1635 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1637 * We only need to hold the superblock lock while we are actually adding
1638 * in the new group's counts to the superblock. Prior to that we have
1639 * not really "added" the group at all. We re-check that we are still
1640 * adding in the last group in case things have changed since verifying.
1642 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1644 struct ext4_new_flex_group_data flex_gd;
1645 struct ext4_sb_info *sbi = EXT4_SB(sb);
1646 struct ext4_super_block *es = sbi->s_es;
1647 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1648 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1649 struct inode *inode = NULL;
1654 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1656 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1657 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1661 if (ext4_blocks_count(es) + input->blocks_count <
1662 ext4_blocks_count(es)) {
1663 ext4_warning(sb, "blocks_count overflow");
1667 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1668 le32_to_cpu(es->s_inodes_count)) {
1669 ext4_warning(sb, "inodes_count overflow");
1673 if (reserved_gdb || gdb_off == 0) {
1674 if (!ext4_has_feature_resize_inode(sb) ||
1675 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1677 "No reserved GDT blocks, can't resize");
1680 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1681 if (IS_ERR(inode)) {
1682 ext4_warning(sb, "Error opening resize inode");
1683 return PTR_ERR(inode);
1688 err = verify_group_input(sb, input);
1692 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1696 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1701 flex_gd.groups = input;
1702 flex_gd.bg_flags = &bg_flags;
1703 err = ext4_flex_group_add(sb, inode, &flex_gd);
1707 } /* ext4_group_add */
1710 * extend a group without checking assuming that checking has been done.
1712 static int ext4_group_extend_no_check(struct super_block *sb,
1713 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1715 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1719 /* We will update the superblock, one block bitmap, and
1720 * one group descriptor via ext4_group_add_blocks().
1722 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1723 if (IS_ERR(handle)) {
1724 err = PTR_ERR(handle);
1725 ext4_warning(sb, "error %d on journal start", err);
1729 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1730 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1732 ext4_warning(sb, "error %d on journal write access", err);
1736 ext4_blocks_count_set(es, o_blocks_count + add);
1737 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1738 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1739 o_blocks_count + add);
1740 /* We add the blocks to the bitmap and set the group need init bit */
1741 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1744 ext4_handle_dirty_super(handle, sb);
1745 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1746 o_blocks_count + add);
1748 err2 = ext4_journal_stop(handle);
1753 if (test_opt(sb, DEBUG))
1754 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1755 "blocks\n", ext4_blocks_count(es));
1756 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1757 (char *)es, sizeof(struct ext4_super_block), 0);
1763 * Extend the filesystem to the new number of blocks specified. This entry
1764 * point is only used to extend the current filesystem to the end of the last
1765 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1766 * for emergencies (because it has no dependencies on reserved blocks).
1768 * If we _really_ wanted, we could use default values to call ext4_group_add()
1769 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1770 * GDT blocks are reserved to grow to the desired size.
1772 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1773 ext4_fsblk_t n_blocks_count)
1775 ext4_fsblk_t o_blocks_count;
1778 struct buffer_head *bh;
1782 o_blocks_count = ext4_blocks_count(es);
1784 if (test_opt(sb, DEBUG))
1785 ext4_msg(sb, KERN_DEBUG,
1786 "extending last group from %llu to %llu blocks",
1787 o_blocks_count, n_blocks_count);
1789 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1792 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1793 ext4_msg(sb, KERN_ERR,
1794 "filesystem too large to resize to %llu blocks safely",
1799 if (n_blocks_count < o_blocks_count) {
1800 ext4_warning(sb, "can't shrink FS - resize aborted");
1804 /* Handle the remaining blocks in the last group only. */
1805 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1808 ext4_warning(sb, "need to use ext2online to resize further");
1812 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1814 if (o_blocks_count + add < o_blocks_count) {
1815 ext4_warning(sb, "blocks_count overflow");
1819 if (o_blocks_count + add > n_blocks_count)
1820 add = n_blocks_count - o_blocks_count;
1822 if (o_blocks_count + add < n_blocks_count)
1823 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1824 o_blocks_count + add, add);
1826 /* See if the device is actually as big as what was requested */
1827 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1829 ext4_warning(sb, "can't read last block, resize aborted");
1834 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1836 } /* ext4_group_extend */
1839 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1841 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1845 * Release the resize inode and drop the resize_inode feature if there
1846 * are no more reserved gdt blocks, and then convert the file system
1849 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1852 struct ext4_sb_info *sbi = EXT4_SB(sb);
1853 struct ext4_super_block *es = sbi->s_es;
1854 struct ext4_inode_info *ei = EXT4_I(inode);
1856 int i, ret, err = 0;
1859 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1861 if (es->s_reserved_gdt_blocks) {
1862 ext4_error(sb, "Unexpected non-zero "
1863 "s_reserved_gdt_blocks");
1867 /* Do a quick sanity check of the resize inode */
1868 if (inode->i_blocks != 1 << (inode->i_blkbits -
1869 (9 - sbi->s_cluster_bits)))
1870 goto invalid_resize_inode;
1871 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1872 if (i == EXT4_DIND_BLOCK) {
1876 goto invalid_resize_inode;
1879 goto invalid_resize_inode;
1881 credits += 3; /* block bitmap, bg descriptor, resize inode */
1884 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1886 return PTR_ERR(handle);
1888 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1889 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1893 ext4_clear_feature_resize_inode(sb);
1894 ext4_set_feature_meta_bg(sb);
1895 sbi->s_es->s_first_meta_bg =
1896 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1898 err = ext4_handle_dirty_super(handle, sb);
1900 ext4_std_error(sb, err);
1905 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1906 ext4_free_blocks(handle, inode, NULL, nr, 1,
1907 EXT4_FREE_BLOCKS_METADATA |
1908 EXT4_FREE_BLOCKS_FORGET);
1909 ei->i_data[EXT4_DIND_BLOCK] = 0;
1910 inode->i_blocks = 0;
1912 err = ext4_mark_inode_dirty(handle, inode);
1914 ext4_std_error(sb, err);
1918 ret = ext4_journal_stop(handle);
1923 invalid_resize_inode:
1924 ext4_error(sb, "corrupted/inconsistent resize inode");
1929 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1931 * @sb: super block of the fs to be resized
1932 * @n_blocks_count: the number of blocks resides in the resized fs
1934 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1936 struct ext4_new_flex_group_data *flex_gd = NULL;
1937 struct ext4_sb_info *sbi = EXT4_SB(sb);
1938 struct ext4_super_block *es = sbi->s_es;
1939 struct buffer_head *bh;
1940 struct inode *resize_inode = NULL;
1941 ext4_grpblk_t add, offset;
1942 unsigned long n_desc_blocks;
1943 unsigned long o_desc_blocks;
1944 ext4_group_t o_group;
1945 ext4_group_t n_group;
1946 ext4_fsblk_t o_blocks_count;
1947 ext4_fsblk_t n_blocks_count_retry = 0;
1948 unsigned long last_update_time = 0;
1949 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1952 /* See if the device is actually as big as what was requested */
1953 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1955 ext4_warning(sb, "can't read last block, resize aborted");
1961 * For bigalloc, trim the requested size to the nearest cluster
1962 * boundary to avoid creating an unusable filesystem. We do this
1963 * silently, instead of returning an error, to avoid breaking
1964 * callers that blindly resize the filesystem to the full size of
1965 * the underlying block device.
1967 if (ext4_has_feature_bigalloc(sb))
1968 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1971 o_blocks_count = ext4_blocks_count(es);
1973 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1974 "to %llu blocks", o_blocks_count, n_blocks_count);
1976 if (n_blocks_count < o_blocks_count) {
1977 /* On-line shrinking not supported */
1978 ext4_warning(sb, "can't shrink FS - resize aborted");
1982 if (n_blocks_count == o_blocks_count)
1983 /* Nothing need to do */
1986 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1987 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1988 ext4_warning(sb, "resize would cause inodes_count overflow");
1991 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1993 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1994 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1996 meta_bg = ext4_has_feature_meta_bg(sb);
1998 if (ext4_has_feature_resize_inode(sb)) {
2000 ext4_error(sb, "resize_inode and meta_bg enabled "
2004 if (n_desc_blocks > o_desc_blocks +
2005 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2006 n_blocks_count_retry = n_blocks_count;
2007 n_desc_blocks = o_desc_blocks +
2008 le16_to_cpu(es->s_reserved_gdt_blocks);
2009 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2010 n_blocks_count = (ext4_fsblk_t)n_group *
2011 EXT4_BLOCKS_PER_GROUP(sb) +
2012 le32_to_cpu(es->s_first_data_block);
2013 n_group--; /* set to last group number */
2017 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2019 if (IS_ERR(resize_inode)) {
2020 ext4_warning(sb, "Error opening resize inode");
2021 return PTR_ERR(resize_inode);
2025 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2026 err = ext4_convert_meta_bg(sb, resize_inode);
2031 resize_inode = NULL;
2033 if (n_blocks_count_retry) {
2034 n_blocks_count = n_blocks_count_retry;
2035 n_blocks_count_retry = 0;
2041 * Make sure the last group has enough space so that it's
2042 * guaranteed to have enough space for all metadata blocks
2043 * that it might need to hold. (We might not need to store
2044 * the inode table blocks in the last block group, but there
2045 * will be cases where this might be needed.)
2047 if ((ext4_group_first_block_no(sb, n_group) +
2048 ext4_group_overhead_blocks(sb, n_group) + 2 +
2049 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2050 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2052 n_blocks_count_retry = 0;
2055 resize_inode = NULL;
2060 /* extend the last group */
2061 if (n_group == o_group)
2062 add = n_blocks_count - o_blocks_count;
2064 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2066 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2071 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2074 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2078 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2082 flex_gd = alloc_flex_gd(flexbg_size);
2083 if (flex_gd == NULL) {
2088 /* Add flex groups. Note that a regular group is a
2089 * flex group with 1 group.
2091 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2093 if (jiffies - last_update_time > HZ * 10) {
2094 if (last_update_time)
2095 ext4_msg(sb, KERN_INFO,
2096 "resized to %llu blocks",
2097 ext4_blocks_count(es));
2098 last_update_time = jiffies;
2100 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2102 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2107 if (!err && n_blocks_count_retry) {
2108 n_blocks_count = n_blocks_count_retry;
2109 n_blocks_count_retry = 0;
2110 free_flex_gd(flex_gd);
2114 resize_inode = NULL;
2121 free_flex_gd(flex_gd);
2122 if (resize_inode != NULL)
2125 ext4_warning(sb, "error (%d) occurred during "
2126 "file system resize", err);
2127 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2128 ext4_blocks_count(es));