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>
17 #include <linux/jiffies.h>
19 #include "ext4_jbd2.h"
26 static void ext4_rcu_ptr_callback(struct rcu_head *head)
28 struct ext4_rcu_ptr *ptr;
30 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
35 void ext4_kvfree_array_rcu(void *to_free)
37 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
41 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
48 int ext4_resize_begin(struct super_block *sb)
50 struct ext4_sb_info *sbi = EXT4_SB(sb);
53 if (!capable(CAP_SYS_RESOURCE))
57 * If the reserved GDT blocks is non-zero, the resize_inode feature
58 * should always be set.
60 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
61 !ext4_has_feature_resize_inode(sb)) {
62 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
67 * If we are not using the primary superblock/GDT copy don't resize,
68 * because the user tools have no way of handling this. Probably a
69 * bad time to do it anyways.
71 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
72 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
73 ext4_warning(sb, "won't resize using backup superblock at %llu",
74 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
79 * We are not allowed to do online-resizing on a filesystem mounted
80 * with error, because it can destroy the filesystem easily.
82 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
83 ext4_warning(sb, "There are errors in the filesystem, "
84 "so online resizing is not allowed");
88 if (ext4_has_feature_sparse_super2(sb)) {
89 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
93 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
94 &EXT4_SB(sb)->s_ext4_flags))
100 int ext4_resize_end(struct super_block *sb, bool update_backups)
102 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
103 smp_mb__after_atomic();
105 return ext4_update_overhead(sb, true);
109 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
110 ext4_group_t group) {
111 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
112 EXT4_DESC_PER_BLOCK_BITS(sb);
115 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
116 ext4_group_t group) {
117 group = ext4_meta_bg_first_group(sb, group);
118 return ext4_group_first_block_no(sb, group);
121 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
122 ext4_group_t group) {
123 ext4_grpblk_t overhead;
124 overhead = ext4_bg_num_gdb(sb, group);
125 if (ext4_bg_has_super(sb, group))
127 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
131 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
132 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
134 static int verify_group_input(struct super_block *sb,
135 struct ext4_new_group_data *input)
137 struct ext4_sb_info *sbi = EXT4_SB(sb);
138 struct ext4_super_block *es = sbi->s_es;
139 ext4_fsblk_t start = ext4_blocks_count(es);
140 ext4_fsblk_t end = start + input->blocks_count;
141 ext4_group_t group = input->group;
142 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
144 ext4_fsblk_t metaend;
145 struct buffer_head *bh = NULL;
146 ext4_grpblk_t free_blocks_count, offset;
149 if (group != sbi->s_groups_count) {
150 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
151 input->group, sbi->s_groups_count);
155 overhead = ext4_group_overhead_blocks(sb, group);
156 metaend = start + overhead;
157 input->free_clusters_count = free_blocks_count =
158 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
160 if (test_opt(sb, DEBUG))
161 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
162 "(%d free, %u reserved)\n",
163 ext4_bg_has_super(sb, input->group) ? "normal" :
164 "no-super", input->group, input->blocks_count,
165 free_blocks_count, input->reserved_blocks);
167 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
169 ext4_warning(sb, "Last group not full");
170 else if (input->reserved_blocks > input->blocks_count / 5)
171 ext4_warning(sb, "Reserved blocks too high (%u)",
172 input->reserved_blocks);
173 else if (free_blocks_count < 0)
174 ext4_warning(sb, "Bad blocks count %u",
175 input->blocks_count);
176 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
179 ext4_warning(sb, "Cannot read last block (%llu)",
181 } else if (outside(input->block_bitmap, start, end))
182 ext4_warning(sb, "Block bitmap not in group (block %llu)",
183 (unsigned long long)input->block_bitmap);
184 else if (outside(input->inode_bitmap, start, end))
185 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
186 (unsigned long long)input->inode_bitmap);
187 else if (outside(input->inode_table, start, end) ||
188 outside(itend - 1, start, end))
189 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
190 (unsigned long long)input->inode_table, itend - 1);
191 else if (input->inode_bitmap == input->block_bitmap)
192 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
193 (unsigned long long)input->block_bitmap);
194 else if (inside(input->block_bitmap, input->inode_table, itend))
195 ext4_warning(sb, "Block bitmap (%llu) in inode table "
197 (unsigned long long)input->block_bitmap,
198 (unsigned long long)input->inode_table, itend - 1);
199 else if (inside(input->inode_bitmap, input->inode_table, itend))
200 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
202 (unsigned long long)input->inode_bitmap,
203 (unsigned long long)input->inode_table, itend - 1);
204 else if (inside(input->block_bitmap, start, metaend))
205 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
206 (unsigned long long)input->block_bitmap,
208 else if (inside(input->inode_bitmap, start, metaend))
209 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
210 (unsigned long long)input->inode_bitmap,
212 else if (inside(input->inode_table, start, metaend) ||
213 inside(itend - 1, start, metaend))
214 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
216 (unsigned long long)input->inode_table,
217 itend - 1, start, metaend - 1);
226 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
229 struct ext4_new_flex_group_data {
230 struct ext4_new_group_data *groups; /* new_group_data for groups
232 __u16 *bg_flags; /* block group flags of groups
234 ext4_group_t count; /* number of groups in @groups
239 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
242 * Returns NULL on failure otherwise address of the allocated structure.
244 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
246 struct ext4_new_flex_group_data *flex_gd;
248 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
252 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
254 flex_gd->count = flexbg_size;
256 flex_gd->groups = kmalloc_array(flexbg_size,
257 sizeof(struct ext4_new_group_data),
259 if (flex_gd->groups == NULL)
262 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
264 if (flex_gd->bg_flags == NULL)
270 kfree(flex_gd->groups);
277 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
279 kfree(flex_gd->bg_flags);
280 kfree(flex_gd->groups);
285 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
286 * and inode tables for a flex group.
288 * This function is used by 64bit-resize. Note that this function allocates
289 * group tables from the 1st group of groups contained by @flexgd, which may
290 * be a partial of a flex group.
292 * @sb: super block of fs to which the groups belongs
294 * Returns 0 on a successful allocation of the metadata blocks in the
297 static int ext4_alloc_group_tables(struct super_block *sb,
298 struct ext4_new_flex_group_data *flex_gd,
301 struct ext4_new_group_data *group_data = flex_gd->groups;
302 ext4_fsblk_t start_blk;
303 ext4_fsblk_t last_blk;
304 ext4_group_t src_group;
305 ext4_group_t bb_index = 0;
306 ext4_group_t ib_index = 0;
307 ext4_group_t it_index = 0;
309 ext4_group_t last_group;
311 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
314 BUG_ON(flex_gd->count == 0 || group_data == NULL);
316 src_group = group_data[0].group;
317 last_group = src_group + flex_gd->count - 1;
319 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
320 (last_group & ~(flexbg_size - 1))));
322 group = group_data[0].group;
323 if (src_group >= group_data[0].group + flex_gd->count)
325 start_blk = ext4_group_first_block_no(sb, src_group);
326 last_blk = start_blk + group_data[src_group - group].blocks_count;
328 overhead = ext4_group_overhead_blocks(sb, src_group);
330 start_blk += overhead;
332 /* We collect contiguous blocks as much as possible. */
334 for (; src_group <= last_group; src_group++) {
335 overhead = ext4_group_overhead_blocks(sb, src_group);
337 last_blk += group_data[src_group - group].blocks_count;
342 /* Allocate block bitmaps */
343 for (; bb_index < flex_gd->count; bb_index++) {
344 if (start_blk >= last_blk)
346 group_data[bb_index].block_bitmap = start_blk++;
347 group = ext4_get_group_number(sb, start_blk - 1);
348 group -= group_data[0].group;
349 group_data[group].mdata_blocks++;
350 flex_gd->bg_flags[group] &= uninit_mask;
353 /* Allocate inode bitmaps */
354 for (; ib_index < flex_gd->count; ib_index++) {
355 if (start_blk >= last_blk)
357 group_data[ib_index].inode_bitmap = start_blk++;
358 group = ext4_get_group_number(sb, start_blk - 1);
359 group -= group_data[0].group;
360 group_data[group].mdata_blocks++;
361 flex_gd->bg_flags[group] &= uninit_mask;
364 /* Allocate inode tables */
365 for (; it_index < flex_gd->count; it_index++) {
366 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
367 ext4_fsblk_t next_group_start;
369 if (start_blk + itb > last_blk)
371 group_data[it_index].inode_table = start_blk;
372 group = ext4_get_group_number(sb, start_blk);
373 next_group_start = ext4_group_first_block_no(sb, group + 1);
374 group -= group_data[0].group;
376 if (start_blk + itb > next_group_start) {
377 flex_gd->bg_flags[group + 1] &= uninit_mask;
378 overhead = start_blk + itb - next_group_start;
379 group_data[group + 1].mdata_blocks += overhead;
383 group_data[group].mdata_blocks += itb;
384 flex_gd->bg_flags[group] &= uninit_mask;
385 start_blk += EXT4_SB(sb)->s_itb_per_group;
388 /* Update free clusters count to exclude metadata blocks */
389 for (i = 0; i < flex_gd->count; i++) {
390 group_data[i].free_clusters_count -=
391 EXT4_NUM_B2C(EXT4_SB(sb),
392 group_data[i].mdata_blocks);
395 if (test_opt(sb, DEBUG)) {
397 group = group_data[0].group;
399 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
400 "%d groups, flexbg size is %d:\n", flex_gd->count,
403 for (i = 0; i < flex_gd->count; i++) {
405 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
406 ext4_bg_has_super(sb, group + i) ? "normal" :
407 "no-super", group + i,
408 group_data[i].blocks_count,
409 group_data[i].free_clusters_count,
410 group_data[i].mdata_blocks);
416 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
419 struct buffer_head *bh;
422 bh = sb_getblk(sb, blk);
424 return ERR_PTR(-ENOMEM);
425 BUFFER_TRACE(bh, "get_write_access");
426 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
431 memset(bh->b_data, 0, sb->s_blocksize);
432 set_buffer_uptodate(bh);
438 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
440 return ext4_journal_ensure_credits_fn(handle, credits,
441 EXT4_MAX_TRANS_DATA, 0, 0);
445 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
447 * Helper function for ext4_setup_new_group_blocks() which set .
450 * @handle: journal handle
451 * @flex_gd: flex group data
453 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
454 struct ext4_new_flex_group_data *flex_gd,
455 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
457 struct ext4_sb_info *sbi = EXT4_SB(sb);
458 ext4_group_t count = last_cluster - first_cluster + 1;
461 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
463 for (count2 = count; count > 0;
464 count -= count2, first_cluster += count2) {
466 struct buffer_head *bh;
470 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
471 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
472 group -= flex_gd->groups[0].group;
474 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
478 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
479 BUG_ON(flex_gd->count > 1);
483 err = ext4_resize_ensure_credits_batch(handle, 1);
487 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
491 BUFFER_TRACE(bh, "get_write_access");
492 err = ext4_journal_get_write_access(handle, sb, bh,
498 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
499 first_cluster, first_cluster - start, count2);
500 mb_set_bits(bh->b_data, first_cluster - start, count2);
502 err = ext4_handle_dirty_metadata(handle, NULL, bh);
512 * Set up the block and inode bitmaps, and the inode table for the new groups.
513 * This doesn't need to be part of the main transaction, since we are only
514 * changing blocks outside the actual filesystem. We still do journaling to
515 * ensure the recovery is correct in case of a failure just after resize.
516 * If any part of this fails, we simply abort the resize.
518 * setup_new_flex_group_blocks handles a flex group as follow:
519 * 1. copy super block and GDT, and initialize group tables if necessary.
520 * In this step, we only set bits in blocks bitmaps for blocks taken by
521 * super block and GDT.
522 * 2. allocate group tables in block bitmaps, that is, set bits in block
523 * bitmap for blocks taken by group tables.
525 static int setup_new_flex_group_blocks(struct super_block *sb,
526 struct ext4_new_flex_group_data *flex_gd)
528 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
531 struct ext4_sb_info *sbi = EXT4_SB(sb);
532 struct ext4_super_block *es = sbi->s_es;
533 struct ext4_new_group_data *group_data = flex_gd->groups;
534 __u16 *bg_flags = flex_gd->bg_flags;
536 ext4_group_t group, count;
537 struct buffer_head *bh = NULL;
538 int reserved_gdb, i, j, err = 0, err2;
541 BUG_ON(!flex_gd->count || !group_data ||
542 group_data[0].group != sbi->s_groups_count);
544 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
545 meta_bg = ext4_has_feature_meta_bg(sb);
547 /* This transaction may be extended/restarted along the way */
548 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
550 return PTR_ERR(handle);
552 group = group_data[0].group;
553 for (i = 0; i < flex_gd->count; i++, group++) {
554 unsigned long gdblocks;
555 ext4_grpblk_t overhead;
557 gdblocks = ext4_bg_num_gdb(sb, group);
558 start = ext4_group_first_block_no(sb, group);
560 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
566 block = start + ext4_bg_has_super(sb, group);
567 /* Copy all of the GDT blocks into the backup in this group */
568 for (j = 0; j < gdblocks; j++, block++) {
569 struct buffer_head *gdb;
571 ext4_debug("update backup group %#04llx\n", block);
572 err = ext4_resize_ensure_credits_batch(handle, 1);
576 gdb = sb_getblk(sb, block);
577 if (unlikely(!gdb)) {
582 BUFFER_TRACE(gdb, "get_write_access");
583 err = ext4_journal_get_write_access(handle, sb, gdb,
589 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
590 s_group_desc, j)->b_data, gdb->b_size);
591 set_buffer_uptodate(gdb);
593 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
601 /* Zero out all of the reserved backup group descriptor
604 if (ext4_bg_has_super(sb, group)) {
605 err = sb_issue_zeroout(sb, gdblocks + start + 1,
606 reserved_gdb, GFP_NOFS);
612 /* Initialize group tables of the grop @group */
613 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
616 /* Zero out all of the inode table blocks */
617 block = group_data[i].inode_table;
618 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
619 block, sbi->s_itb_per_group);
620 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
626 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
629 /* Initialize block bitmap of the @group */
630 block = group_data[i].block_bitmap;
631 err = ext4_resize_ensure_credits_batch(handle, 1);
635 bh = bclean(handle, sb, block);
640 overhead = ext4_group_overhead_blocks(sb, group);
642 ext4_debug("mark backup superblock %#04llx (+0)\n",
644 mb_set_bits(bh->b_data, 0,
645 EXT4_NUM_B2C(sbi, overhead));
647 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
648 sb->s_blocksize * 8, bh->b_data);
649 err = ext4_handle_dirty_metadata(handle, NULL, bh);
655 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
658 /* Initialize inode bitmap of the @group */
659 block = group_data[i].inode_bitmap;
660 err = ext4_resize_ensure_credits_batch(handle, 1);
663 /* Mark unused entries in inode bitmap used */
664 bh = bclean(handle, sb, block);
670 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
671 sb->s_blocksize * 8, bh->b_data);
672 err = ext4_handle_dirty_metadata(handle, NULL, bh);
678 /* Mark group tables in block bitmap */
679 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
680 count = group_table_count[j];
681 start = (&group_data[0].block_bitmap)[j];
683 for (i = 1; i < flex_gd->count; i++) {
684 block += group_table_count[j];
685 if (block == (&group_data[i].block_bitmap)[j]) {
686 count += group_table_count[j];
689 err = set_flexbg_block_bitmap(sb, handle,
691 EXT4_B2C(sbi, start),
697 count = group_table_count[j];
698 start = (&group_data[i].block_bitmap)[j];
703 err = set_flexbg_block_bitmap(sb, handle,
705 EXT4_B2C(sbi, start),
715 err2 = ext4_journal_stop(handle);
723 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
724 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
725 * calling this for the first time. In a sparse filesystem it will be the
726 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
727 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
729 unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
730 unsigned int *five, unsigned int *seven)
732 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
733 unsigned int *min = three;
737 if (ext4_has_feature_sparse_super2(sb)) {
741 ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
747 if (!ext4_has_feature_sparse_super(sb)) {
769 * Check that all of the backup GDT blocks are held in the primary GDT block.
770 * It is assumed that they are stored in group order. Returns the number of
771 * groups in current filesystem that have BACKUPS, or -ve error code.
773 static int verify_reserved_gdb(struct super_block *sb,
775 struct buffer_head *primary)
777 const ext4_fsblk_t blk = primary->b_blocknr;
782 __le32 *p = (__le32 *)primary->b_data;
785 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
786 if (le32_to_cpu(*p++) !=
787 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
788 ext4_warning(sb, "reserved GDT %llu"
789 " missing grp %d (%llu)",
792 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
796 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
804 * Called when we need to bring a reserved group descriptor table block into
805 * use from the resize inode. The primary copy of the new GDT block currently
806 * is an indirect block (under the double indirect block in the resize inode).
807 * The new backup GDT blocks will be stored as leaf blocks in this indirect
808 * block, in group order. Even though we know all the block numbers we need,
809 * we check to ensure that the resize inode has actually reserved these blocks.
811 * Don't need to update the block bitmaps because the blocks are still in use.
813 * We get all of the error cases out of the way, so that we are sure to not
814 * fail once we start modifying the data on disk, because JBD has no rollback.
816 static int add_new_gdb(handle_t *handle, struct inode *inode,
819 struct super_block *sb = inode->i_sb;
820 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
821 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
822 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
823 struct buffer_head **o_group_desc, **n_group_desc = NULL;
824 struct buffer_head *dind = NULL;
825 struct buffer_head *gdb_bh = NULL;
827 struct ext4_iloc iloc = { .bh = NULL };
831 if (test_opt(sb, DEBUG))
833 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
836 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
838 return PTR_ERR(gdb_bh);
840 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
846 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
847 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
854 data = (__le32 *)dind->b_data;
855 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
856 ext4_warning(sb, "new group %u GDT block %llu not reserved",
862 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
863 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
868 BUFFER_TRACE(gdb_bh, "get_write_access");
869 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
873 BUFFER_TRACE(dind, "get_write_access");
874 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
876 ext4_std_error(sb, err);
880 /* ext4_reserve_inode_write() gets a reference on the iloc */
881 err = ext4_reserve_inode_write(handle, inode, &iloc);
885 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
889 ext4_warning(sb, "not enough memory for %lu groups",
895 * Finally, we have all of the possible failures behind us...
897 * Remove new GDT block from inode double-indirect block and clear out
898 * the new GDT block for use (which also "frees" the backup GDT blocks
899 * from the reserved inode). We don't need to change the bitmaps for
900 * these blocks, because they are marked as in-use from being in the
901 * reserved inode, and will become GDT blocks (primary and backup).
903 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
904 err = ext4_handle_dirty_metadata(handle, NULL, dind);
906 ext4_std_error(sb, err);
909 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
910 (9 - EXT4_SB(sb)->s_cluster_bits);
911 ext4_mark_iloc_dirty(handle, inode, &iloc);
912 memset(gdb_bh->b_data, 0, sb->s_blocksize);
913 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
915 ext4_std_error(sb, err);
922 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
923 memcpy(n_group_desc, o_group_desc,
924 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
926 n_group_desc[gdb_num] = gdb_bh;
927 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
928 EXT4_SB(sb)->s_gdb_count++;
929 ext4_kvfree_array_rcu(o_group_desc);
931 lock_buffer(EXT4_SB(sb)->s_sbh);
932 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
933 ext4_superblock_csum_set(sb);
934 unlock_buffer(EXT4_SB(sb)->s_sbh);
935 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
937 ext4_std_error(sb, err);
940 kvfree(n_group_desc);
945 ext4_debug("leaving with error %d\n", err);
950 * add_new_gdb_meta_bg is the sister of add_new_gdb.
952 static int add_new_gdb_meta_bg(struct super_block *sb,
953 handle_t *handle, ext4_group_t group) {
954 ext4_fsblk_t gdblock;
955 struct buffer_head *gdb_bh;
956 struct buffer_head **o_group_desc, **n_group_desc;
957 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
960 gdblock = ext4_meta_bg_first_block_no(sb, group) +
961 ext4_bg_has_super(sb, group);
962 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
964 return PTR_ERR(gdb_bh);
965 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
970 ext4_warning(sb, "not enough memory for %lu groups",
976 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
977 memcpy(n_group_desc, o_group_desc,
978 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
980 n_group_desc[gdb_num] = gdb_bh;
982 BUFFER_TRACE(gdb_bh, "get_write_access");
983 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
985 kvfree(n_group_desc);
990 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
991 EXT4_SB(sb)->s_gdb_count++;
992 ext4_kvfree_array_rcu(o_group_desc);
997 * Called when we are adding a new group which has a backup copy of each of
998 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
999 * We need to add these reserved backup GDT blocks to the resize inode, so
1000 * that they are kept for future resizing and not allocated to files.
1002 * Each reserved backup GDT block will go into a different indirect block.
1003 * The indirect blocks are actually the primary reserved GDT blocks,
1004 * so we know in advance what their block numbers are. We only get the
1005 * double-indirect block to verify it is pointing to the primary reserved
1006 * GDT blocks so we don't overwrite a data block by accident. The reserved
1007 * backup GDT blocks are stored in their reserved primary GDT block.
1009 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1012 struct super_block *sb = inode->i_sb;
1013 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1014 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1015 struct buffer_head **primary;
1016 struct buffer_head *dind;
1017 struct ext4_iloc iloc;
1024 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1028 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1029 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1031 err = PTR_ERR(dind);
1036 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1037 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1038 EXT4_ADDR_PER_BLOCK(sb));
1039 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1041 /* Get each reserved primary GDT block and verify it holds backups */
1042 for (res = 0; res < reserved_gdb; res++, blk++) {
1043 if (le32_to_cpu(*data) != blk) {
1044 ext4_warning(sb, "reserved block %llu"
1045 " not at offset %ld",
1047 (long)(data - (__le32 *)dind->b_data));
1051 primary[res] = ext4_sb_bread(sb, blk, 0);
1052 if (IS_ERR(primary[res])) {
1053 err = PTR_ERR(primary[res]);
1054 primary[res] = NULL;
1057 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1058 if (gdbackups < 0) {
1059 brelse(primary[res]);
1064 data = (__le32 *)dind->b_data;
1067 for (i = 0; i < reserved_gdb; i++) {
1068 BUFFER_TRACE(primary[i], "get_write_access");
1069 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1074 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1078 * Finally we can add each of the reserved backup GDT blocks from
1079 * the new group to its reserved primary GDT block.
1081 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1082 for (i = 0; i < reserved_gdb; i++) {
1084 data = (__le32 *)primary[i]->b_data;
1085 /* printk("reserving backup %lu[%u] = %lu\n",
1086 primary[i]->b_blocknr, gdbackups,
1087 blk + primary[i]->b_blocknr); */
1088 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1089 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1094 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1095 ext4_mark_iloc_dirty(handle, inode, &iloc);
1099 brelse(primary[res]);
1108 static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1111 struct ext4_super_block *es = (struct ext4_super_block *) data;
1113 es->s_block_group_nr = cpu_to_le16(group);
1114 if (ext4_has_metadata_csum(sb))
1115 es->s_checksum = ext4_superblock_csum(sb, es);
1119 * Update the backup copies of the ext4 metadata. These don't need to be part
1120 * of the main resize transaction, because e2fsck will re-write them if there
1121 * is a problem (basically only OOM will cause a problem). However, we
1122 * _should_ update the backups if possible, in case the primary gets trashed
1123 * for some reason and we need to run e2fsck from a backup superblock. The
1124 * important part is that the new block and inode counts are in the backup
1125 * superblocks, and the location of the new group metadata in the GDT backups.
1127 * We do not need take the s_resize_lock for this, because these
1128 * blocks are not otherwise touched by the filesystem code when it is
1129 * mounted. We don't need to worry about last changing from
1130 * sbi->s_groups_count, because the worst that can happen is that we
1131 * do not copy the full number of backups at this time. The resize
1132 * which changed s_groups_count will backup again.
1134 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1135 int size, int meta_bg)
1137 struct ext4_sb_info *sbi = EXT4_SB(sb);
1139 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1143 ext4_group_t group = 0;
1144 int rest = sb->s_blocksize - size;
1148 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1149 if (IS_ERR(handle)) {
1151 err = PTR_ERR(handle);
1156 group = ext4_list_backups(sb, &three, &five, &seven);
1157 last = sbi->s_groups_count;
1159 group = ext4_get_group_number(sb, blk_off) + 1;
1160 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1163 while (group < sbi->s_groups_count) {
1164 struct buffer_head *bh;
1165 ext4_fsblk_t backup_block;
1166 int has_super = ext4_bg_has_super(sb, group);
1167 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1169 /* Out of journal space, and can't get more - abort - so sad */
1170 err = ext4_resize_ensure_credits_batch(handle, 1);
1175 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1177 backup_block = first_block + has_super;
1179 bh = sb_getblk(sb, backup_block);
1180 if (unlikely(!bh)) {
1184 ext4_debug("update metadata backup %llu(+%llu)\n",
1185 backup_block, backup_block -
1186 ext4_group_first_block_no(sb, group));
1187 BUFFER_TRACE(bh, "get_write_access");
1188 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1194 memcpy(bh->b_data, data, size);
1196 memset(bh->b_data + size, 0, rest);
1197 if (has_super && (backup_block == first_block))
1198 ext4_set_block_group_nr(sb, bh->b_data, group);
1199 set_buffer_uptodate(bh);
1201 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1203 ext4_std_error(sb, err);
1207 group = ext4_list_backups(sb, &three, &five, &seven);
1208 else if (group == last)
1213 if ((err2 = ext4_journal_stop(handle)) && !err)
1217 * Ugh! Need to have e2fsck write the backup copies. It is too
1218 * late to revert the resize, we shouldn't fail just because of
1219 * the backup copies (they are only needed in case of corruption).
1221 * However, if we got here we have a journal problem too, so we
1222 * can't really start a transaction to mark the superblock.
1223 * Chicken out and just set the flag on the hope it will be written
1224 * to disk, and if not - we will simply wait until next fsck.
1228 ext4_warning(sb, "can't update backup for group %u (err %d), "
1229 "forcing fsck on next reboot", group, err);
1230 sbi->s_mount_state &= ~EXT4_VALID_FS;
1231 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1232 mark_buffer_dirty(sbi->s_sbh);
1237 * ext4_add_new_descs() adds @count group descriptor of groups
1238 * starting at @group
1240 * @handle: journal handle
1242 * @group: the group no. of the first group desc to be added
1243 * @resize_inode: the resize inode
1244 * @count: number of group descriptors to be added
1246 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1247 ext4_group_t group, struct inode *resize_inode,
1250 struct ext4_sb_info *sbi = EXT4_SB(sb);
1251 struct ext4_super_block *es = sbi->s_es;
1252 struct buffer_head *gdb_bh;
1253 int i, gdb_off, gdb_num, err = 0;
1256 meta_bg = ext4_has_feature_meta_bg(sb);
1257 for (i = 0; i < count; i++, group++) {
1258 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1259 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1261 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1262 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1265 * We will only either add reserved group blocks to a backup group
1266 * or remove reserved blocks for the first group in a new group block.
1267 * Doing both would be mean more complex code, and sane people don't
1268 * use non-sparse filesystems anymore. This is already checked above.
1271 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1273 BUFFER_TRACE(gdb_bh, "get_write_access");
1274 err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1277 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1278 err = reserve_backup_gdb(handle, resize_inode, group);
1279 } else if (meta_bg != 0) {
1280 err = add_new_gdb_meta_bg(sb, handle, group);
1282 err = add_new_gdb(handle, resize_inode, group);
1290 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1292 struct buffer_head *bh = sb_getblk(sb, block);
1295 if (!bh_uptodate_or_lock(bh)) {
1296 if (ext4_read_bh(bh, 0, NULL) < 0) {
1305 static int ext4_set_bitmap_checksums(struct super_block *sb,
1306 struct ext4_group_desc *gdp,
1307 struct ext4_new_group_data *group_data)
1309 struct buffer_head *bh;
1311 if (!ext4_has_metadata_csum(sb))
1314 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1317 ext4_inode_bitmap_csum_set(sb, gdp, bh,
1318 EXT4_INODES_PER_GROUP(sb) / 8);
1321 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1324 ext4_block_bitmap_csum_set(sb, gdp, bh);
1331 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1333 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1334 struct ext4_new_flex_group_data *flex_gd)
1336 struct ext4_new_group_data *group_data = flex_gd->groups;
1337 struct ext4_group_desc *gdp;
1338 struct ext4_sb_info *sbi = EXT4_SB(sb);
1339 struct buffer_head *gdb_bh;
1341 __u16 *bg_flags = flex_gd->bg_flags;
1342 int i, gdb_off, gdb_num, err = 0;
1345 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1346 group = group_data->group;
1348 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1349 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1352 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1354 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1355 /* Update group descriptor block for new group */
1356 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1357 gdb_off * EXT4_DESC_SIZE(sb));
1359 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1360 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1361 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1362 err = ext4_set_bitmap_checksums(sb, gdp, group_data);
1364 ext4_std_error(sb, err);
1368 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1369 ext4_free_group_clusters_set(sb, gdp,
1370 group_data->free_clusters_count);
1371 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1372 if (ext4_has_group_desc_csum(sb))
1373 ext4_itable_unused_set(sb, gdp,
1374 EXT4_INODES_PER_GROUP(sb));
1375 gdp->bg_flags = cpu_to_le16(*bg_flags);
1376 ext4_group_desc_csum_set(sb, group, gdp);
1378 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1379 if (unlikely(err)) {
1380 ext4_std_error(sb, err);
1385 * We can allocate memory for mb_alloc based on the new group
1388 err = ext4_mb_add_groupinfo(sb, group, gdp);
1395 static void ext4_add_overhead(struct super_block *sb,
1396 const ext4_fsblk_t overhead)
1398 struct ext4_sb_info *sbi = EXT4_SB(sb);
1399 struct ext4_super_block *es = sbi->s_es;
1401 sbi->s_overhead += overhead;
1402 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1407 * ext4_update_super() updates the super block so that the newly added
1408 * groups can be seen by the filesystem.
1411 * @flex_gd: new added groups
1413 static void ext4_update_super(struct super_block *sb,
1414 struct ext4_new_flex_group_data *flex_gd)
1416 ext4_fsblk_t blocks_count = 0;
1417 ext4_fsblk_t free_blocks = 0;
1418 ext4_fsblk_t reserved_blocks = 0;
1419 struct ext4_new_group_data *group_data = flex_gd->groups;
1420 struct ext4_sb_info *sbi = EXT4_SB(sb);
1421 struct ext4_super_block *es = sbi->s_es;
1424 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1426 * Make the new blocks and inodes valid next. We do this before
1427 * increasing the group count so that once the group is enabled,
1428 * all of its blocks and inodes are already valid.
1430 * We always allocate group-by-group, then block-by-block or
1431 * inode-by-inode within a group, so enabling these
1432 * blocks/inodes before the group is live won't actually let us
1433 * allocate the new space yet.
1435 for (i = 0; i < flex_gd->count; i++) {
1436 blocks_count += group_data[i].blocks_count;
1437 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1440 reserved_blocks = ext4_r_blocks_count(es) * 100;
1441 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1442 reserved_blocks *= blocks_count;
1443 do_div(reserved_blocks, 100);
1445 lock_buffer(sbi->s_sbh);
1446 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1447 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1448 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1450 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1453 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1455 * We need to protect s_groups_count against other CPUs seeing
1456 * inconsistent state in the superblock.
1458 * The precise rules we use are:
1460 * * Writers must perform a smp_wmb() after updating all
1461 * dependent data and before modifying the groups count
1463 * * Readers must perform an smp_rmb() after reading the groups
1464 * count and before reading any dependent data.
1466 * NB. These rules can be relaxed when checking the group count
1467 * while freeing data, as we can only allocate from a block
1468 * group after serialising against the group count, and we can
1469 * only then free after serialising in turn against that
1474 /* Update the global fs size fields */
1475 sbi->s_groups_count += flex_gd->count;
1476 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1477 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1479 /* Update the reserved block counts only once the new group is
1481 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1484 /* Update the free space counts */
1485 percpu_counter_add(&sbi->s_freeclusters_counter,
1486 EXT4_NUM_B2C(sbi, free_blocks));
1487 percpu_counter_add(&sbi->s_freeinodes_counter,
1488 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1490 ext4_debug("free blocks count %llu",
1491 percpu_counter_read(&sbi->s_freeclusters_counter));
1492 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1493 ext4_group_t flex_group;
1494 struct flex_groups *fg;
1496 flex_group = ext4_flex_group(sbi, group_data[0].group);
1497 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1498 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1499 &fg->free_clusters);
1500 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1505 * Update the fs overhead information.
1507 * For bigalloc, if the superblock already has a properly calculated
1508 * overhead, update it with a value based on numbers already computed
1509 * above for the newly allocated capacity.
1511 if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1512 ext4_add_overhead(sb,
1513 EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1515 ext4_calculate_overhead(sb);
1516 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1518 ext4_superblock_csum_set(sb);
1519 unlock_buffer(sbi->s_sbh);
1520 if (test_opt(sb, DEBUG))
1521 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1522 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1523 blocks_count, free_blocks, reserved_blocks);
1526 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1527 * _before_ we start modifying the filesystem, because we cannot abort the
1528 * transaction and not have it write the data to disk.
1530 static int ext4_flex_group_add(struct super_block *sb,
1531 struct inode *resize_inode,
1532 struct ext4_new_flex_group_data *flex_gd)
1534 struct ext4_sb_info *sbi = EXT4_SB(sb);
1535 struct ext4_super_block *es = sbi->s_es;
1536 ext4_fsblk_t o_blocks_count;
1540 unsigned reserved_gdb;
1541 int err = 0, err2 = 0, credit;
1543 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1545 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1546 o_blocks_count = ext4_blocks_count(es);
1547 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1550 err = setup_new_flex_group_blocks(sb, flex_gd);
1554 * We will always be modifying at least the superblock and GDT
1555 * blocks. If we are adding a group past the last current GDT block,
1556 * we will also modify the inode and the dindirect block. If we
1557 * are adding a group with superblock/GDT backups we will also
1558 * modify each of the reserved GDT dindirect blocks.
1560 credit = 3; /* sb, resize inode, resize inode dindirect */
1562 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1563 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1564 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1565 if (IS_ERR(handle)) {
1566 err = PTR_ERR(handle);
1570 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1571 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1576 group = flex_gd->groups[0].group;
1577 BUG_ON(group != sbi->s_groups_count);
1578 err = ext4_add_new_descs(handle, sb, group,
1579 resize_inode, flex_gd->count);
1583 err = ext4_setup_new_descs(handle, sb, flex_gd);
1587 ext4_update_super(sb, flex_gd);
1589 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1592 err2 = ext4_journal_stop(handle);
1597 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1598 int gdb_num_end = ((group + flex_gd->count - 1) /
1599 EXT4_DESC_PER_BLOCK(sb));
1600 int meta_bg = ext4_has_feature_meta_bg(sb);
1601 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1602 ext4_group_first_block_no(sb, 0);
1603 sector_t old_gdb = 0;
1605 update_backups(sb, ext4_group_first_block_no(sb, 0),
1606 (char *)es, sizeof(struct ext4_super_block), 0);
1607 for (; gdb_num <= gdb_num_end; gdb_num++) {
1608 struct buffer_head *gdb_bh;
1610 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1612 if (old_gdb == gdb_bh->b_blocknr)
1614 update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1615 gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1616 old_gdb = gdb_bh->b_blocknr;
1623 static int ext4_setup_next_flex_gd(struct super_block *sb,
1624 struct ext4_new_flex_group_data *flex_gd,
1625 ext4_fsblk_t n_blocks_count,
1626 unsigned long flexbg_size)
1628 struct ext4_sb_info *sbi = EXT4_SB(sb);
1629 struct ext4_super_block *es = sbi->s_es;
1630 struct ext4_new_group_data *group_data = flex_gd->groups;
1631 ext4_fsblk_t o_blocks_count;
1632 ext4_group_t n_group;
1634 ext4_group_t last_group;
1636 ext4_grpblk_t clusters_per_group;
1639 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1641 o_blocks_count = ext4_blocks_count(es);
1643 if (o_blocks_count == n_blocks_count)
1646 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1648 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1650 last_group = group | (flexbg_size - 1);
1651 if (last_group > n_group)
1652 last_group = n_group;
1654 flex_gd->count = last_group - group + 1;
1656 for (i = 0; i < flex_gd->count; i++) {
1659 group_data[i].group = group + i;
1660 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1661 overhead = ext4_group_overhead_blocks(sb, group + i);
1662 group_data[i].mdata_blocks = overhead;
1663 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1664 if (ext4_has_group_desc_csum(sb)) {
1665 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1666 EXT4_BG_INODE_UNINIT;
1667 if (!test_opt(sb, INIT_INODE_TABLE))
1668 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1670 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1673 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1674 /* We need to initialize block bitmap of last group. */
1675 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1677 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1678 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1679 group_data[i - 1].free_clusters_count -= clusters_per_group -
1686 /* Add group descriptor data to an existing or new group descriptor block.
1687 * Ensure we handle all possible error conditions _before_ we start modifying
1688 * the filesystem, because we cannot abort the transaction and not have it
1689 * write the data to disk.
1691 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1692 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1694 * We only need to hold the superblock lock while we are actually adding
1695 * in the new group's counts to the superblock. Prior to that we have
1696 * not really "added" the group at all. We re-check that we are still
1697 * adding in the last group in case things have changed since verifying.
1699 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1701 struct ext4_new_flex_group_data flex_gd;
1702 struct ext4_sb_info *sbi = EXT4_SB(sb);
1703 struct ext4_super_block *es = sbi->s_es;
1704 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1705 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1706 struct inode *inode = NULL;
1711 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1713 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1714 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1718 if (ext4_blocks_count(es) + input->blocks_count <
1719 ext4_blocks_count(es)) {
1720 ext4_warning(sb, "blocks_count overflow");
1724 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1725 le32_to_cpu(es->s_inodes_count)) {
1726 ext4_warning(sb, "inodes_count overflow");
1730 if (reserved_gdb || gdb_off == 0) {
1731 if (!ext4_has_feature_resize_inode(sb) ||
1732 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1734 "No reserved GDT blocks, can't resize");
1737 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1738 if (IS_ERR(inode)) {
1739 ext4_warning(sb, "Error opening resize inode");
1740 return PTR_ERR(inode);
1745 err = verify_group_input(sb, input);
1749 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1753 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1758 flex_gd.groups = input;
1759 flex_gd.bg_flags = &bg_flags;
1760 err = ext4_flex_group_add(sb, inode, &flex_gd);
1764 } /* ext4_group_add */
1767 * extend a group without checking assuming that checking has been done.
1769 static int ext4_group_extend_no_check(struct super_block *sb,
1770 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1772 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1776 /* We will update the superblock, one block bitmap, and
1777 * one group descriptor via ext4_group_add_blocks().
1779 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1780 if (IS_ERR(handle)) {
1781 err = PTR_ERR(handle);
1782 ext4_warning(sb, "error %d on journal start", err);
1786 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1787 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1790 ext4_warning(sb, "error %d on journal write access", err);
1794 lock_buffer(EXT4_SB(sb)->s_sbh);
1795 ext4_blocks_count_set(es, o_blocks_count + add);
1796 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1797 ext4_superblock_csum_set(sb);
1798 unlock_buffer(EXT4_SB(sb)->s_sbh);
1799 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1800 o_blocks_count + add);
1801 /* We add the blocks to the bitmap and set the group need init bit */
1802 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1805 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1806 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1807 o_blocks_count + add);
1809 err2 = ext4_journal_stop(handle);
1814 if (test_opt(sb, DEBUG))
1815 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1816 "blocks\n", ext4_blocks_count(es));
1817 update_backups(sb, ext4_group_first_block_no(sb, 0),
1818 (char *)es, sizeof(struct ext4_super_block), 0);
1824 * Extend the filesystem to the new number of blocks specified. This entry
1825 * point is only used to extend the current filesystem to the end of the last
1826 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1827 * for emergencies (because it has no dependencies on reserved blocks).
1829 * If we _really_ wanted, we could use default values to call ext4_group_add()
1830 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1831 * GDT blocks are reserved to grow to the desired size.
1833 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1834 ext4_fsblk_t n_blocks_count)
1836 ext4_fsblk_t o_blocks_count;
1839 struct buffer_head *bh;
1842 o_blocks_count = ext4_blocks_count(es);
1844 if (test_opt(sb, DEBUG))
1845 ext4_msg(sb, KERN_DEBUG,
1846 "extending last group from %llu to %llu blocks",
1847 o_blocks_count, n_blocks_count);
1849 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1852 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1853 ext4_msg(sb, KERN_ERR,
1854 "filesystem too large to resize to %llu blocks safely",
1859 if (n_blocks_count < o_blocks_count) {
1860 ext4_warning(sb, "can't shrink FS - resize aborted");
1864 /* Handle the remaining blocks in the last group only. */
1865 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1868 ext4_warning(sb, "need to use ext2online to resize further");
1872 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1874 if (o_blocks_count + add < o_blocks_count) {
1875 ext4_warning(sb, "blocks_count overflow");
1879 if (o_blocks_count + add > n_blocks_count)
1880 add = n_blocks_count - o_blocks_count;
1882 if (o_blocks_count + add < n_blocks_count)
1883 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1884 o_blocks_count + add, add);
1886 /* See if the device is actually as big as what was requested */
1887 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1889 ext4_warning(sb, "can't read last block, resize aborted");
1894 return ext4_group_extend_no_check(sb, o_blocks_count, add);
1895 } /* ext4_group_extend */
1898 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1900 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1904 * Release the resize inode and drop the resize_inode feature if there
1905 * are no more reserved gdt blocks, and then convert the file system
1908 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1911 struct ext4_sb_info *sbi = EXT4_SB(sb);
1912 struct ext4_super_block *es = sbi->s_es;
1913 struct ext4_inode_info *ei = EXT4_I(inode);
1915 int i, ret, err = 0;
1918 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1920 if (es->s_reserved_gdt_blocks) {
1921 ext4_error(sb, "Unexpected non-zero "
1922 "s_reserved_gdt_blocks");
1926 /* Do a quick sanity check of the resize inode */
1927 if (inode->i_blocks != 1 << (inode->i_blkbits -
1928 (9 - sbi->s_cluster_bits)))
1929 goto invalid_resize_inode;
1930 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1931 if (i == EXT4_DIND_BLOCK) {
1935 goto invalid_resize_inode;
1938 goto invalid_resize_inode;
1940 credits += 3; /* block bitmap, bg descriptor, resize inode */
1943 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1945 return PTR_ERR(handle);
1947 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1948 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1953 lock_buffer(sbi->s_sbh);
1954 ext4_clear_feature_resize_inode(sb);
1955 ext4_set_feature_meta_bg(sb);
1956 sbi->s_es->s_first_meta_bg =
1957 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1958 ext4_superblock_csum_set(sb);
1959 unlock_buffer(sbi->s_sbh);
1961 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1963 ext4_std_error(sb, err);
1968 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1969 ext4_free_blocks(handle, inode, NULL, nr, 1,
1970 EXT4_FREE_BLOCKS_METADATA |
1971 EXT4_FREE_BLOCKS_FORGET);
1972 ei->i_data[EXT4_DIND_BLOCK] = 0;
1973 inode->i_blocks = 0;
1975 err = ext4_mark_inode_dirty(handle, inode);
1977 ext4_std_error(sb, err);
1981 ret = ext4_journal_stop(handle);
1982 return err ? err : ret;
1984 invalid_resize_inode:
1985 ext4_error(sb, "corrupted/inconsistent resize inode");
1990 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1992 * @sb: super block of the fs to be resized
1993 * @n_blocks_count: the number of blocks resides in the resized fs
1995 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1997 struct ext4_new_flex_group_data *flex_gd = NULL;
1998 struct ext4_sb_info *sbi = EXT4_SB(sb);
1999 struct ext4_super_block *es = sbi->s_es;
2000 struct buffer_head *bh;
2001 struct inode *resize_inode = NULL;
2002 ext4_grpblk_t add, offset;
2003 unsigned long n_desc_blocks;
2004 unsigned long o_desc_blocks;
2005 ext4_group_t o_group;
2006 ext4_group_t n_group;
2007 ext4_fsblk_t o_blocks_count;
2008 ext4_fsblk_t n_blocks_count_retry = 0;
2009 unsigned long last_update_time = 0;
2010 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
2013 /* See if the device is actually as big as what was requested */
2014 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2016 ext4_warning(sb, "can't read last block, resize aborted");
2022 * For bigalloc, trim the requested size to the nearest cluster
2023 * boundary to avoid creating an unusable filesystem. We do this
2024 * silently, instead of returning an error, to avoid breaking
2025 * callers that blindly resize the filesystem to the full size of
2026 * the underlying block device.
2028 if (ext4_has_feature_bigalloc(sb))
2029 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2032 o_blocks_count = ext4_blocks_count(es);
2034 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2035 "to %llu blocks", o_blocks_count, n_blocks_count);
2037 if (n_blocks_count < o_blocks_count) {
2038 /* On-line shrinking not supported */
2039 ext4_warning(sb, "can't shrink FS - resize aborted");
2043 if (n_blocks_count == o_blocks_count)
2044 /* Nothing need to do */
2047 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2048 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2049 ext4_warning(sb, "resize would cause inodes_count overflow");
2052 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2054 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2055 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2057 meta_bg = ext4_has_feature_meta_bg(sb);
2059 if (ext4_has_feature_resize_inode(sb)) {
2061 ext4_error(sb, "resize_inode and meta_bg enabled "
2065 if (n_desc_blocks > o_desc_blocks +
2066 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2067 n_blocks_count_retry = n_blocks_count;
2068 n_desc_blocks = o_desc_blocks +
2069 le16_to_cpu(es->s_reserved_gdt_blocks);
2070 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2071 n_blocks_count = (ext4_fsblk_t)n_group *
2072 EXT4_BLOCKS_PER_GROUP(sb) +
2073 le32_to_cpu(es->s_first_data_block);
2074 n_group--; /* set to last group number */
2078 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2080 if (IS_ERR(resize_inode)) {
2081 ext4_warning(sb, "Error opening resize inode");
2082 return PTR_ERR(resize_inode);
2086 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2087 err = ext4_convert_meta_bg(sb, resize_inode);
2092 resize_inode = NULL;
2094 if (n_blocks_count_retry) {
2095 n_blocks_count = n_blocks_count_retry;
2096 n_blocks_count_retry = 0;
2102 * Make sure the last group has enough space so that it's
2103 * guaranteed to have enough space for all metadata blocks
2104 * that it might need to hold. (We might not need to store
2105 * the inode table blocks in the last block group, but there
2106 * will be cases where this might be needed.)
2108 if ((ext4_group_first_block_no(sb, n_group) +
2109 ext4_group_overhead_blocks(sb, n_group) + 2 +
2110 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2111 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2113 n_blocks_count_retry = 0;
2116 resize_inode = NULL;
2121 /* extend the last group */
2122 if (n_group == o_group)
2123 add = n_blocks_count - o_blocks_count;
2125 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2127 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2132 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2135 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2139 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2143 flex_gd = alloc_flex_gd(flexbg_size);
2144 if (flex_gd == NULL) {
2149 /* Add flex groups. Note that a regular group is a
2150 * flex group with 1 group.
2152 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2154 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2155 if (last_update_time)
2156 ext4_msg(sb, KERN_INFO,
2157 "resized to %llu blocks",
2158 ext4_blocks_count(es));
2159 last_update_time = jiffies;
2161 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2163 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2168 if (!err && n_blocks_count_retry) {
2169 n_blocks_count = n_blocks_count_retry;
2170 n_blocks_count_retry = 0;
2171 free_flex_gd(flex_gd);
2175 resize_inode = NULL;
2182 free_flex_gd(flex_gd);
2183 if (resize_inode != NULL)
2186 ext4_warning(sb, "error (%d) occurred during "
2187 "file system resize", err);
2188 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2189 ext4_blocks_count(es));