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_blocks_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(sizeof(struct ext4_new_group_data) *
253 flexbg_size, GFP_NOFS);
254 if (flex_gd->groups == NULL)
257 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
258 if (flex_gd->bg_flags == NULL)
264 kfree(flex_gd->groups);
271 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
273 kfree(flex_gd->bg_flags);
274 kfree(flex_gd->groups);
279 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
280 * and inode tables for a flex group.
282 * This function is used by 64bit-resize. Note that this function allocates
283 * group tables from the 1st group of groups contained by @flexgd, which may
284 * be a partial of a flex group.
286 * @sb: super block of fs to which the groups belongs
288 * Returns 0 on a successful allocation of the metadata blocks in the
291 static int ext4_alloc_group_tables(struct super_block *sb,
292 struct ext4_new_flex_group_data *flex_gd,
295 struct ext4_new_group_data *group_data = flex_gd->groups;
296 ext4_fsblk_t start_blk;
297 ext4_fsblk_t last_blk;
298 ext4_group_t src_group;
299 ext4_group_t bb_index = 0;
300 ext4_group_t ib_index = 0;
301 ext4_group_t it_index = 0;
303 ext4_group_t last_group;
305 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
307 BUG_ON(flex_gd->count == 0 || group_data == NULL);
309 src_group = group_data[0].group;
310 last_group = src_group + flex_gd->count - 1;
312 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
313 (last_group & ~(flexbg_size - 1))));
315 group = group_data[0].group;
316 if (src_group >= group_data[0].group + flex_gd->count)
318 start_blk = ext4_group_first_block_no(sb, src_group);
319 last_blk = start_blk + group_data[src_group - group].blocks_count;
321 overhead = ext4_group_overhead_blocks(sb, src_group);
323 start_blk += overhead;
325 /* We collect contiguous blocks as much as possible. */
327 for (; src_group <= last_group; src_group++) {
328 overhead = ext4_group_overhead_blocks(sb, src_group);
330 last_blk += group_data[src_group - group].blocks_count;
335 /* Allocate block bitmaps */
336 for (; bb_index < flex_gd->count; bb_index++) {
337 if (start_blk >= last_blk)
339 group_data[bb_index].block_bitmap = start_blk++;
340 group = ext4_get_group_number(sb, start_blk - 1);
341 group -= group_data[0].group;
342 group_data[group].free_blocks_count--;
343 flex_gd->bg_flags[group] &= uninit_mask;
346 /* Allocate inode bitmaps */
347 for (; ib_index < flex_gd->count; ib_index++) {
348 if (start_blk >= last_blk)
350 group_data[ib_index].inode_bitmap = start_blk++;
351 group = ext4_get_group_number(sb, start_blk - 1);
352 group -= group_data[0].group;
353 group_data[group].free_blocks_count--;
354 flex_gd->bg_flags[group] &= uninit_mask;
357 /* Allocate inode tables */
358 for (; it_index < flex_gd->count; it_index++) {
359 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
360 ext4_fsblk_t next_group_start;
362 if (start_blk + itb > last_blk)
364 group_data[it_index].inode_table = start_blk;
365 group = ext4_get_group_number(sb, start_blk);
366 next_group_start = ext4_group_first_block_no(sb, group + 1);
367 group -= group_data[0].group;
369 if (start_blk + itb > next_group_start) {
370 flex_gd->bg_flags[group + 1] &= uninit_mask;
371 overhead = start_blk + itb - next_group_start;
372 group_data[group + 1].free_blocks_count -= overhead;
376 group_data[group].free_blocks_count -= itb;
377 flex_gd->bg_flags[group] &= uninit_mask;
378 start_blk += EXT4_SB(sb)->s_itb_per_group;
381 if (test_opt(sb, DEBUG)) {
383 group = group_data[0].group;
385 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
386 "%d groups, flexbg size is %d:\n", flex_gd->count,
389 for (i = 0; i < flex_gd->count; i++) {
390 printk(KERN_DEBUG "adding %s group %u: %u "
391 "blocks (%d free)\n",
392 ext4_bg_has_super(sb, group + i) ? "normal" :
393 "no-super", group + i,
394 group_data[i].blocks_count,
395 group_data[i].free_blocks_count);
401 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
404 struct buffer_head *bh;
407 bh = sb_getblk(sb, blk);
409 return ERR_PTR(-ENOMEM);
410 BUFFER_TRACE(bh, "get_write_access");
411 if ((err = ext4_journal_get_write_access(handle, bh))) {
415 memset(bh->b_data, 0, sb->s_blocksize);
416 set_buffer_uptodate(bh);
423 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
424 * If that fails, restart the transaction & regain write access for the
425 * buffer head which is used for block_bitmap modifications.
427 static int extend_or_restart_transaction(handle_t *handle, int thresh)
431 if (ext4_handle_has_enough_credits(handle, thresh))
434 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
438 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
447 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
449 * Helper function for ext4_setup_new_group_blocks() which set .
452 * @handle: journal handle
453 * @flex_gd: flex group data
455 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
456 struct ext4_new_flex_group_data *flex_gd,
457 ext4_fsblk_t block, ext4_group_t count)
461 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
462 for (count2 = count; count > 0; count -= count2, block += count2) {
464 struct buffer_head *bh;
468 group = ext4_get_group_number(sb, block);
469 start = ext4_group_first_block_no(sb, group);
470 group -= flex_gd->groups[0].group;
472 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
476 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
477 BUG_ON(flex_gd->count > 1);
481 err = extend_or_restart_transaction(handle, 1);
485 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
489 BUFFER_TRACE(bh, "get_write_access");
490 err = ext4_journal_get_write_access(handle, bh);
495 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
496 block - start, count2);
497 ext4_set_bits(bh->b_data, block - start, count2);
499 err = ext4_handle_dirty_metadata(handle, NULL, bh);
509 * Set up the block and inode bitmaps, and the inode table for the new groups.
510 * This doesn't need to be part of the main transaction, since we are only
511 * changing blocks outside the actual filesystem. We still do journaling to
512 * ensure the recovery is correct in case of a failure just after resize.
513 * If any part of this fails, we simply abort the resize.
515 * setup_new_flex_group_blocks handles a flex group as follow:
516 * 1. copy super block and GDT, and initialize group tables if necessary.
517 * In this step, we only set bits in blocks bitmaps for blocks taken by
518 * super block and GDT.
519 * 2. allocate group tables in block bitmaps, that is, set bits in block
520 * bitmap for blocks taken by group tables.
522 static int setup_new_flex_group_blocks(struct super_block *sb,
523 struct ext4_new_flex_group_data *flex_gd)
525 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
528 struct ext4_sb_info *sbi = EXT4_SB(sb);
529 struct ext4_super_block *es = sbi->s_es;
530 struct ext4_new_group_data *group_data = flex_gd->groups;
531 __u16 *bg_flags = flex_gd->bg_flags;
533 ext4_group_t group, count;
534 struct buffer_head *bh = NULL;
535 int reserved_gdb, i, j, err = 0, err2;
538 BUG_ON(!flex_gd->count || !group_data ||
539 group_data[0].group != sbi->s_groups_count);
541 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
542 meta_bg = ext4_has_feature_meta_bg(sb);
544 /* This transaction may be extended/restarted along the way */
545 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
547 return PTR_ERR(handle);
549 group = group_data[0].group;
550 for (i = 0; i < flex_gd->count; i++, group++) {
551 unsigned long gdblocks;
552 ext4_grpblk_t overhead;
554 gdblocks = ext4_bg_num_gdb(sb, group);
555 start = ext4_group_first_block_no(sb, group);
557 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
561 ext4_group_t first_group;
562 first_group = ext4_meta_bg_first_group(sb, group);
563 if (first_group != group + 1 &&
564 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
568 block = start + ext4_bg_has_super(sb, group);
569 /* Copy all of the GDT blocks into the backup in this group */
570 for (j = 0; j < gdblocks; j++, block++) {
571 struct buffer_head *gdb;
573 ext4_debug("update backup group %#04llx\n", block);
574 err = extend_or_restart_transaction(handle, 1);
578 gdb = sb_getblk(sb, block);
579 if (unlikely(!gdb)) {
584 BUFFER_TRACE(gdb, "get_write_access");
585 err = ext4_journal_get_write_access(handle, gdb);
590 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
591 s_group_desc, j)->b_data, gdb->b_size);
592 set_buffer_uptodate(gdb);
594 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
602 /* Zero out all of the reserved backup group descriptor
605 if (ext4_bg_has_super(sb, group)) {
606 err = sb_issue_zeroout(sb, gdblocks + start + 1,
607 reserved_gdb, GFP_NOFS);
613 /* Initialize group tables of the grop @group */
614 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
617 /* Zero out all of the inode table blocks */
618 block = group_data[i].inode_table;
619 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
620 block, sbi->s_itb_per_group);
621 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
627 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
630 /* Initialize block bitmap of the @group */
631 block = group_data[i].block_bitmap;
632 err = extend_or_restart_transaction(handle, 1);
636 bh = bclean(handle, sb, block);
641 overhead = ext4_group_overhead_blocks(sb, group);
643 ext4_debug("mark backup superblock %#04llx (+0)\n",
645 ext4_set_bits(bh->b_data, 0, overhead);
647 ext4_mark_bitmap_end(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 = extend_or_restart_transaction(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,
690 flex_gd, start, count);
693 count = group_table_count[j];
694 start = (&group_data[i].block_bitmap)[j];
699 err = set_flexbg_block_bitmap(sb, handle,
700 flex_gd, start, count);
707 err2 = ext4_journal_stop(handle);
715 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
716 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
717 * calling this for the first time. In a sparse filesystem it will be the
718 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
719 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
721 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
722 unsigned *five, unsigned *seven)
724 unsigned *min = three;
728 if (!ext4_has_feature_sparse_super(sb)) {
750 * Check that all of the backup GDT blocks are held in the primary GDT block.
751 * It is assumed that they are stored in group order. Returns the number of
752 * groups in current filesystem that have BACKUPS, or -ve error code.
754 static int verify_reserved_gdb(struct super_block *sb,
756 struct buffer_head *primary)
758 const ext4_fsblk_t blk = primary->b_blocknr;
763 __le32 *p = (__le32 *)primary->b_data;
766 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
767 if (le32_to_cpu(*p++) !=
768 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
769 ext4_warning(sb, "reserved GDT %llu"
770 " missing grp %d (%llu)",
773 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
777 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
785 * Called when we need to bring a reserved group descriptor table block into
786 * use from the resize inode. The primary copy of the new GDT block currently
787 * is an indirect block (under the double indirect block in the resize inode).
788 * The new backup GDT blocks will be stored as leaf blocks in this indirect
789 * block, in group order. Even though we know all the block numbers we need,
790 * we check to ensure that the resize inode has actually reserved these blocks.
792 * Don't need to update the block bitmaps because the blocks are still in use.
794 * We get all of the error cases out of the way, so that we are sure to not
795 * fail once we start modifying the data on disk, because JBD has no rollback.
797 static int add_new_gdb(handle_t *handle, struct inode *inode,
800 struct super_block *sb = inode->i_sb;
801 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
802 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
803 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
804 struct buffer_head **o_group_desc, **n_group_desc = NULL;
805 struct buffer_head *dind = NULL;
806 struct buffer_head *gdb_bh = NULL;
808 struct ext4_iloc iloc = { .bh = NULL };
812 if (test_opt(sb, DEBUG))
814 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
817 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
819 return PTR_ERR(gdb_bh);
821 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
827 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
828 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
835 data = (__le32 *)dind->b_data;
836 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
837 ext4_warning(sb, "new group %u GDT block %llu not reserved",
843 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
844 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
848 BUFFER_TRACE(gdb_bh, "get_write_access");
849 err = ext4_journal_get_write_access(handle, gdb_bh);
853 BUFFER_TRACE(dind, "get_write_access");
854 err = ext4_journal_get_write_access(handle, dind);
856 ext4_std_error(sb, err);
860 /* ext4_reserve_inode_write() gets a reference on the iloc */
861 err = ext4_reserve_inode_write(handle, inode, &iloc);
865 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
866 sizeof(struct buffer_head *),
870 ext4_warning(sb, "not enough memory for %lu groups",
876 * Finally, we have all of the possible failures behind us...
878 * Remove new GDT block from inode double-indirect block and clear out
879 * the new GDT block for use (which also "frees" the backup GDT blocks
880 * from the reserved inode). We don't need to change the bitmaps for
881 * these blocks, because they are marked as in-use from being in the
882 * reserved inode, and will become GDT blocks (primary and backup).
884 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
885 err = ext4_handle_dirty_metadata(handle, NULL, dind);
887 ext4_std_error(sb, err);
890 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
891 ext4_mark_iloc_dirty(handle, inode, &iloc);
892 memset(gdb_bh->b_data, 0, sb->s_blocksize);
893 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
895 ext4_std_error(sb, err);
902 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
903 memcpy(n_group_desc, o_group_desc,
904 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
906 n_group_desc[gdb_num] = gdb_bh;
907 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
908 EXT4_SB(sb)->s_gdb_count++;
909 ext4_kvfree_array_rcu(o_group_desc);
911 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
912 err = ext4_handle_dirty_super(handle, sb);
914 ext4_std_error(sb, err);
917 kvfree(n_group_desc);
922 ext4_debug("leaving with error %d\n", err);
927 * add_new_gdb_meta_bg is the sister of add_new_gdb.
929 static int add_new_gdb_meta_bg(struct super_block *sb,
930 handle_t *handle, ext4_group_t group) {
931 ext4_fsblk_t gdblock;
932 struct buffer_head *gdb_bh;
933 struct buffer_head **o_group_desc, **n_group_desc;
934 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
937 gdblock = ext4_meta_bg_first_block_no(sb, group) +
938 ext4_bg_has_super(sb, group);
939 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
941 return PTR_ERR(gdb_bh);
942 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
943 sizeof(struct buffer_head *),
948 ext4_warning(sb, "not enough memory for %lu groups",
954 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
955 memcpy(n_group_desc, o_group_desc,
956 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
958 n_group_desc[gdb_num] = gdb_bh;
960 BUFFER_TRACE(gdb_bh, "get_write_access");
961 err = ext4_journal_get_write_access(handle, gdb_bh);
963 kvfree(n_group_desc);
968 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
969 EXT4_SB(sb)->s_gdb_count++;
970 ext4_kvfree_array_rcu(o_group_desc);
975 * Called when we are adding a new group which has a backup copy of each of
976 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
977 * We need to add these reserved backup GDT blocks to the resize inode, so
978 * that they are kept for future resizing and not allocated to files.
980 * Each reserved backup GDT block will go into a different indirect block.
981 * The indirect blocks are actually the primary reserved GDT blocks,
982 * so we know in advance what their block numbers are. We only get the
983 * double-indirect block to verify it is pointing to the primary reserved
984 * GDT blocks so we don't overwrite a data block by accident. The reserved
985 * backup GDT blocks are stored in their reserved primary GDT block.
987 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
990 struct super_block *sb = inode->i_sb;
991 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
992 struct buffer_head **primary;
993 struct buffer_head *dind;
994 struct ext4_iloc iloc;
1001 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
1005 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1006 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1008 err = PTR_ERR(dind);
1013 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1014 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1015 EXT4_ADDR_PER_BLOCK(sb));
1016 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1018 /* Get each reserved primary GDT block and verify it holds backups */
1019 for (res = 0; res < reserved_gdb; res++, blk++) {
1020 if (le32_to_cpu(*data) != blk) {
1021 ext4_warning(sb, "reserved block %llu"
1022 " not at offset %ld",
1024 (long)(data - (__le32 *)dind->b_data));
1028 primary[res] = ext4_sb_bread(sb, blk, 0);
1029 if (IS_ERR(primary[res])) {
1030 err = PTR_ERR(primary[res]);
1031 primary[res] = NULL;
1034 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1035 if (gdbackups < 0) {
1036 brelse(primary[res]);
1041 data = (__le32 *)dind->b_data;
1044 for (i = 0; i < reserved_gdb; i++) {
1045 BUFFER_TRACE(primary[i], "get_write_access");
1046 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1050 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1054 * Finally we can add each of the reserved backup GDT blocks from
1055 * the new group to its reserved primary GDT block.
1057 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1058 for (i = 0; i < reserved_gdb; i++) {
1060 data = (__le32 *)primary[i]->b_data;
1061 /* printk("reserving backup %lu[%u] = %lu\n",
1062 primary[i]->b_blocknr, gdbackups,
1063 blk + primary[i]->b_blocknr); */
1064 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1065 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1069 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1070 ext4_mark_iloc_dirty(handle, inode, &iloc);
1074 brelse(primary[res]);
1084 * Update the backup copies of the ext4 metadata. These don't need to be part
1085 * of the main resize transaction, because e2fsck will re-write them if there
1086 * is a problem (basically only OOM will cause a problem). However, we
1087 * _should_ update the backups if possible, in case the primary gets trashed
1088 * for some reason and we need to run e2fsck from a backup superblock. The
1089 * important part is that the new block and inode counts are in the backup
1090 * superblocks, and the location of the new group metadata in the GDT backups.
1092 * We do not need take the s_resize_lock for this, because these
1093 * blocks are not otherwise touched by the filesystem code when it is
1094 * mounted. We don't need to worry about last changing from
1095 * sbi->s_groups_count, because the worst that can happen is that we
1096 * do not copy the full number of backups at this time. The resize
1097 * which changed s_groups_count will backup again.
1099 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1100 int size, int meta_bg)
1102 struct ext4_sb_info *sbi = EXT4_SB(sb);
1104 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1108 ext4_group_t group = 0;
1109 int rest = sb->s_blocksize - size;
1113 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1114 if (IS_ERR(handle)) {
1116 err = PTR_ERR(handle);
1121 group = ext4_list_backups(sb, &three, &five, &seven);
1122 last = sbi->s_groups_count;
1124 group = ext4_get_group_number(sb, blk_off) + 1;
1125 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1128 while (group < sbi->s_groups_count) {
1129 struct buffer_head *bh;
1130 ext4_fsblk_t backup_block;
1132 /* Out of journal space, and can't get more - abort - so sad */
1133 if (ext4_handle_valid(handle) &&
1134 handle->h_buffer_credits == 0 &&
1135 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1136 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1140 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1142 backup_block = (ext4_group_first_block_no(sb, group) +
1143 ext4_bg_has_super(sb, group));
1145 bh = sb_getblk(sb, backup_block);
1146 if (unlikely(!bh)) {
1150 ext4_debug("update metadata backup %llu(+%llu)\n",
1151 backup_block, backup_block -
1152 ext4_group_first_block_no(sb, group));
1153 BUFFER_TRACE(bh, "get_write_access");
1154 if ((err = ext4_journal_get_write_access(handle, bh))) {
1159 memcpy(bh->b_data, data, size);
1161 memset(bh->b_data + size, 0, rest);
1162 set_buffer_uptodate(bh);
1164 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1166 ext4_std_error(sb, err);
1170 group = ext4_list_backups(sb, &three, &five, &seven);
1171 else if (group == last)
1176 if ((err2 = ext4_journal_stop(handle)) && !err)
1180 * Ugh! Need to have e2fsck write the backup copies. It is too
1181 * late to revert the resize, we shouldn't fail just because of
1182 * the backup copies (they are only needed in case of corruption).
1184 * However, if we got here we have a journal problem too, so we
1185 * can't really start a transaction to mark the superblock.
1186 * Chicken out and just set the flag on the hope it will be written
1187 * to disk, and if not - we will simply wait until next fsck.
1191 ext4_warning(sb, "can't update backup for group %u (err %d), "
1192 "forcing fsck on next reboot", group, err);
1193 sbi->s_mount_state &= ~EXT4_VALID_FS;
1194 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1195 mark_buffer_dirty(sbi->s_sbh);
1200 * ext4_add_new_descs() adds @count group descriptor of groups
1201 * starting at @group
1203 * @handle: journal handle
1205 * @group: the group no. of the first group desc to be added
1206 * @resize_inode: the resize inode
1207 * @count: number of group descriptors to be added
1209 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1210 ext4_group_t group, struct inode *resize_inode,
1213 struct ext4_sb_info *sbi = EXT4_SB(sb);
1214 struct ext4_super_block *es = sbi->s_es;
1215 struct buffer_head *gdb_bh;
1216 int i, gdb_off, gdb_num, err = 0;
1219 meta_bg = ext4_has_feature_meta_bg(sb);
1220 for (i = 0; i < count; i++, group++) {
1221 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1222 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1224 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1225 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1228 * We will only either add reserved group blocks to a backup group
1229 * or remove reserved blocks for the first group in a new group block.
1230 * Doing both would be mean more complex code, and sane people don't
1231 * use non-sparse filesystems anymore. This is already checked above.
1234 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1236 BUFFER_TRACE(gdb_bh, "get_write_access");
1237 err = ext4_journal_get_write_access(handle, gdb_bh);
1239 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1240 err = reserve_backup_gdb(handle, resize_inode, group);
1241 } else if (meta_bg != 0) {
1242 err = add_new_gdb_meta_bg(sb, handle, group);
1244 err = add_new_gdb(handle, resize_inode, group);
1252 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1254 struct buffer_head *bh = sb_getblk(sb, block);
1257 if (!bh_uptodate_or_lock(bh)) {
1258 if (bh_submit_read(bh) < 0) {
1267 static int ext4_set_bitmap_checksums(struct super_block *sb,
1269 struct ext4_group_desc *gdp,
1270 struct ext4_new_group_data *group_data)
1272 struct buffer_head *bh;
1274 if (!ext4_has_metadata_csum(sb))
1277 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1280 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1281 EXT4_INODES_PER_GROUP(sb) / 8);
1284 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1287 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1294 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1296 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1297 struct ext4_new_flex_group_data *flex_gd)
1299 struct ext4_new_group_data *group_data = flex_gd->groups;
1300 struct ext4_group_desc *gdp;
1301 struct ext4_sb_info *sbi = EXT4_SB(sb);
1302 struct buffer_head *gdb_bh;
1304 __u16 *bg_flags = flex_gd->bg_flags;
1305 int i, gdb_off, gdb_num, err = 0;
1308 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1309 group = group_data->group;
1311 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1312 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1315 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1317 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1318 /* Update group descriptor block for new group */
1319 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1320 gdb_off * EXT4_DESC_SIZE(sb));
1322 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1323 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1324 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1325 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1327 ext4_std_error(sb, err);
1331 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1332 ext4_free_group_clusters_set(sb, gdp,
1333 EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1334 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1335 if (ext4_has_group_desc_csum(sb))
1336 ext4_itable_unused_set(sb, gdp,
1337 EXT4_INODES_PER_GROUP(sb));
1338 gdp->bg_flags = cpu_to_le16(*bg_flags);
1339 ext4_group_desc_csum_set(sb, group, gdp);
1341 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1342 if (unlikely(err)) {
1343 ext4_std_error(sb, err);
1348 * We can allocate memory for mb_alloc based on the new group
1351 err = ext4_mb_add_groupinfo(sb, group, gdp);
1359 * ext4_update_super() updates the super block so that the newly added
1360 * groups can be seen by the filesystem.
1363 * @flex_gd: new added groups
1365 static void ext4_update_super(struct super_block *sb,
1366 struct ext4_new_flex_group_data *flex_gd)
1368 ext4_fsblk_t blocks_count = 0;
1369 ext4_fsblk_t free_blocks = 0;
1370 ext4_fsblk_t reserved_blocks = 0;
1371 struct ext4_new_group_data *group_data = flex_gd->groups;
1372 struct ext4_sb_info *sbi = EXT4_SB(sb);
1373 struct ext4_super_block *es = sbi->s_es;
1376 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1378 * Make the new blocks and inodes valid next. We do this before
1379 * increasing the group count so that once the group is enabled,
1380 * all of its blocks and inodes are already valid.
1382 * We always allocate group-by-group, then block-by-block or
1383 * inode-by-inode within a group, so enabling these
1384 * blocks/inodes before the group is live won't actually let us
1385 * allocate the new space yet.
1387 for (i = 0; i < flex_gd->count; i++) {
1388 blocks_count += group_data[i].blocks_count;
1389 free_blocks += group_data[i].free_blocks_count;
1392 reserved_blocks = ext4_r_blocks_count(es) * 100;
1393 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1394 reserved_blocks *= blocks_count;
1395 do_div(reserved_blocks, 100);
1397 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1398 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1399 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1401 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1404 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1406 * We need to protect s_groups_count against other CPUs seeing
1407 * inconsistent state in the superblock.
1409 * The precise rules we use are:
1411 * * Writers must perform a smp_wmb() after updating all
1412 * dependent data and before modifying the groups count
1414 * * Readers must perform an smp_rmb() after reading the groups
1415 * count and before reading any dependent data.
1417 * NB. These rules can be relaxed when checking the group count
1418 * while freeing data, as we can only allocate from a block
1419 * group after serialising against the group count, and we can
1420 * only then free after serialising in turn against that
1425 /* Update the global fs size fields */
1426 sbi->s_groups_count += flex_gd->count;
1427 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1428 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1430 /* Update the reserved block counts only once the new group is
1432 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1435 /* Update the free space counts */
1436 percpu_counter_add(&sbi->s_freeclusters_counter,
1437 EXT4_NUM_B2C(sbi, free_blocks));
1438 percpu_counter_add(&sbi->s_freeinodes_counter,
1439 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1441 ext4_debug("free blocks count %llu",
1442 percpu_counter_read(&sbi->s_freeclusters_counter));
1443 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1444 ext4_group_t flex_group;
1445 struct flex_groups *fg;
1447 flex_group = ext4_flex_group(sbi, group_data[0].group);
1448 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1449 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1450 &fg->free_clusters);
1451 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1456 * Update the fs overhead information
1458 ext4_calculate_overhead(sb);
1459 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1461 if (test_opt(sb, DEBUG))
1462 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1463 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1464 blocks_count, free_blocks, reserved_blocks);
1467 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1468 * _before_ we start modifying the filesystem, because we cannot abort the
1469 * transaction and not have it write the data to disk.
1471 static int ext4_flex_group_add(struct super_block *sb,
1472 struct inode *resize_inode,
1473 struct ext4_new_flex_group_data *flex_gd)
1475 struct ext4_sb_info *sbi = EXT4_SB(sb);
1476 struct ext4_super_block *es = sbi->s_es;
1477 ext4_fsblk_t o_blocks_count;
1481 unsigned reserved_gdb;
1482 int err = 0, err2 = 0, credit;
1484 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1486 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1487 o_blocks_count = ext4_blocks_count(es);
1488 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1491 err = setup_new_flex_group_blocks(sb, flex_gd);
1495 * We will always be modifying at least the superblock and GDT
1496 * blocks. If we are adding a group past the last current GDT block,
1497 * we will also modify the inode and the dindirect block. If we
1498 * are adding a group with superblock/GDT backups we will also
1499 * modify each of the reserved GDT dindirect blocks.
1501 credit = 3; /* sb, resize inode, resize inode dindirect */
1503 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1504 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1505 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1506 if (IS_ERR(handle)) {
1507 err = PTR_ERR(handle);
1511 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1512 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1516 group = flex_gd->groups[0].group;
1517 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1518 err = ext4_add_new_descs(handle, sb, group,
1519 resize_inode, flex_gd->count);
1523 err = ext4_setup_new_descs(handle, sb, flex_gd);
1527 ext4_update_super(sb, flex_gd);
1529 err = ext4_handle_dirty_super(handle, sb);
1532 err2 = ext4_journal_stop(handle);
1537 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1538 int gdb_num_end = ((group + flex_gd->count - 1) /
1539 EXT4_DESC_PER_BLOCK(sb));
1540 int meta_bg = ext4_has_feature_meta_bg(sb);
1541 sector_t old_gdb = 0;
1543 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1544 sizeof(struct ext4_super_block), 0);
1545 for (; gdb_num <= gdb_num_end; gdb_num++) {
1546 struct buffer_head *gdb_bh;
1548 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1550 if (old_gdb == gdb_bh->b_blocknr)
1552 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1553 gdb_bh->b_size, meta_bg);
1554 old_gdb = gdb_bh->b_blocknr;
1561 static int ext4_setup_next_flex_gd(struct super_block *sb,
1562 struct ext4_new_flex_group_data *flex_gd,
1563 ext4_fsblk_t n_blocks_count,
1564 unsigned long flexbg_size)
1566 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1567 struct ext4_new_group_data *group_data = flex_gd->groups;
1568 ext4_fsblk_t o_blocks_count;
1569 ext4_group_t n_group;
1571 ext4_group_t last_group;
1573 ext4_grpblk_t blocks_per_group;
1576 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1578 o_blocks_count = ext4_blocks_count(es);
1580 if (o_blocks_count == n_blocks_count)
1583 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1585 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1587 last_group = group | (flexbg_size - 1);
1588 if (last_group > n_group)
1589 last_group = n_group;
1591 flex_gd->count = last_group - group + 1;
1593 for (i = 0; i < flex_gd->count; i++) {
1596 group_data[i].group = group + i;
1597 group_data[i].blocks_count = blocks_per_group;
1598 overhead = ext4_group_overhead_blocks(sb, group + i);
1599 group_data[i].free_blocks_count = blocks_per_group - overhead;
1600 if (ext4_has_group_desc_csum(sb)) {
1601 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1602 EXT4_BG_INODE_UNINIT;
1603 if (!test_opt(sb, INIT_INODE_TABLE))
1604 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1606 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1609 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1610 /* We need to initialize block bitmap of last group. */
1611 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1613 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1614 group_data[i - 1].blocks_count = last + 1;
1615 group_data[i - 1].free_blocks_count -= blocks_per_group-
1622 /* Add group descriptor data to an existing or new group descriptor block.
1623 * Ensure we handle all possible error conditions _before_ we start modifying
1624 * the filesystem, because we cannot abort the transaction and not have it
1625 * write the data to disk.
1627 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1628 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1630 * We only need to hold the superblock lock while we are actually adding
1631 * in the new group's counts to the superblock. Prior to that we have
1632 * not really "added" the group at all. We re-check that we are still
1633 * adding in the last group in case things have changed since verifying.
1635 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1637 struct ext4_new_flex_group_data flex_gd;
1638 struct ext4_sb_info *sbi = EXT4_SB(sb);
1639 struct ext4_super_block *es = sbi->s_es;
1640 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1641 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1642 struct inode *inode = NULL;
1647 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1649 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1650 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1654 if (ext4_blocks_count(es) + input->blocks_count <
1655 ext4_blocks_count(es)) {
1656 ext4_warning(sb, "blocks_count overflow");
1660 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1661 le32_to_cpu(es->s_inodes_count)) {
1662 ext4_warning(sb, "inodes_count overflow");
1666 if (reserved_gdb || gdb_off == 0) {
1667 if (!ext4_has_feature_resize_inode(sb) ||
1668 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1670 "No reserved GDT blocks, can't resize");
1673 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1674 if (IS_ERR(inode)) {
1675 ext4_warning(sb, "Error opening resize inode");
1676 return PTR_ERR(inode);
1681 err = verify_group_input(sb, input);
1685 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1689 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1694 flex_gd.groups = input;
1695 flex_gd.bg_flags = &bg_flags;
1696 err = ext4_flex_group_add(sb, inode, &flex_gd);
1700 } /* ext4_group_add */
1703 * extend a group without checking assuming that checking has been done.
1705 static int ext4_group_extend_no_check(struct super_block *sb,
1706 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1708 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1712 /* We will update the superblock, one block bitmap, and
1713 * one group descriptor via ext4_group_add_blocks().
1715 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1716 if (IS_ERR(handle)) {
1717 err = PTR_ERR(handle);
1718 ext4_warning(sb, "error %d on journal start", err);
1722 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1723 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1725 ext4_warning(sb, "error %d on journal write access", err);
1729 ext4_blocks_count_set(es, o_blocks_count + add);
1730 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1731 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1732 o_blocks_count + add);
1733 /* We add the blocks to the bitmap and set the group need init bit */
1734 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1737 ext4_handle_dirty_super(handle, sb);
1738 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1739 o_blocks_count + add);
1741 err2 = ext4_journal_stop(handle);
1746 if (test_opt(sb, DEBUG))
1747 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1748 "blocks\n", ext4_blocks_count(es));
1749 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1750 (char *)es, sizeof(struct ext4_super_block), 0);
1756 * Extend the filesystem to the new number of blocks specified. This entry
1757 * point is only used to extend the current filesystem to the end of the last
1758 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1759 * for emergencies (because it has no dependencies on reserved blocks).
1761 * If we _really_ wanted, we could use default values to call ext4_group_add()
1762 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1763 * GDT blocks are reserved to grow to the desired size.
1765 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1766 ext4_fsblk_t n_blocks_count)
1768 ext4_fsblk_t o_blocks_count;
1771 struct buffer_head *bh;
1775 o_blocks_count = ext4_blocks_count(es);
1777 if (test_opt(sb, DEBUG))
1778 ext4_msg(sb, KERN_DEBUG,
1779 "extending last group from %llu to %llu blocks",
1780 o_blocks_count, n_blocks_count);
1782 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1785 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1786 ext4_msg(sb, KERN_ERR,
1787 "filesystem too large to resize to %llu blocks safely",
1789 if (sizeof(sector_t) < 8)
1790 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1794 if (n_blocks_count < o_blocks_count) {
1795 ext4_warning(sb, "can't shrink FS - resize aborted");
1799 /* Handle the remaining blocks in the last group only. */
1800 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1803 ext4_warning(sb, "need to use ext2online to resize further");
1807 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1809 if (o_blocks_count + add < o_blocks_count) {
1810 ext4_warning(sb, "blocks_count overflow");
1814 if (o_blocks_count + add > n_blocks_count)
1815 add = n_blocks_count - o_blocks_count;
1817 if (o_blocks_count + add < n_blocks_count)
1818 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1819 o_blocks_count + add, add);
1821 /* See if the device is actually as big as what was requested */
1822 bh = sb_bread(sb, o_blocks_count + add - 1);
1824 ext4_warning(sb, "can't read last block, resize aborted");
1829 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1831 } /* ext4_group_extend */
1834 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1836 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1840 * Release the resize inode and drop the resize_inode feature if there
1841 * are no more reserved gdt blocks, and then convert the file system
1844 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1847 struct ext4_sb_info *sbi = EXT4_SB(sb);
1848 struct ext4_super_block *es = sbi->s_es;
1849 struct ext4_inode_info *ei = EXT4_I(inode);
1851 int i, ret, err = 0;
1854 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1856 if (es->s_reserved_gdt_blocks) {
1857 ext4_error(sb, "Unexpected non-zero "
1858 "s_reserved_gdt_blocks");
1862 /* Do a quick sanity check of the resize inode */
1863 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1864 goto invalid_resize_inode;
1865 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1866 if (i == EXT4_DIND_BLOCK) {
1870 goto invalid_resize_inode;
1873 goto invalid_resize_inode;
1875 credits += 3; /* block bitmap, bg descriptor, resize inode */
1878 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1880 return PTR_ERR(handle);
1882 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1883 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1887 ext4_clear_feature_resize_inode(sb);
1888 ext4_set_feature_meta_bg(sb);
1889 sbi->s_es->s_first_meta_bg =
1890 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1892 err = ext4_handle_dirty_super(handle, sb);
1894 ext4_std_error(sb, err);
1899 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1900 ext4_free_blocks(handle, inode, NULL, nr, 1,
1901 EXT4_FREE_BLOCKS_METADATA |
1902 EXT4_FREE_BLOCKS_FORGET);
1903 ei->i_data[EXT4_DIND_BLOCK] = 0;
1904 inode->i_blocks = 0;
1906 err = ext4_mark_inode_dirty(handle, inode);
1908 ext4_std_error(sb, err);
1912 ret = ext4_journal_stop(handle);
1917 invalid_resize_inode:
1918 ext4_error(sb, "corrupted/inconsistent resize inode");
1923 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1925 * @sb: super block of the fs to be resized
1926 * @n_blocks_count: the number of blocks resides in the resized fs
1928 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1930 struct ext4_new_flex_group_data *flex_gd = NULL;
1931 struct ext4_sb_info *sbi = EXT4_SB(sb);
1932 struct ext4_super_block *es = sbi->s_es;
1933 struct buffer_head *bh;
1934 struct inode *resize_inode = NULL;
1935 ext4_grpblk_t add, offset;
1936 unsigned long n_desc_blocks;
1937 unsigned long o_desc_blocks;
1938 ext4_group_t o_group;
1939 ext4_group_t n_group;
1940 ext4_fsblk_t o_blocks_count;
1941 ext4_fsblk_t n_blocks_count_retry = 0;
1942 unsigned long last_update_time = 0;
1943 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1946 /* See if the device is actually as big as what was requested */
1947 bh = sb_bread(sb, n_blocks_count - 1);
1949 ext4_warning(sb, "can't read last block, resize aborted");
1955 * For bigalloc, trim the requested size to the nearest cluster
1956 * boundary to avoid creating an unusable filesystem. We do this
1957 * silently, instead of returning an error, to avoid breaking
1958 * callers that blindly resize the filesystem to the full size of
1959 * the underlying block device.
1961 if (ext4_has_feature_bigalloc(sb))
1962 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1965 o_blocks_count = ext4_blocks_count(es);
1967 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1968 "to %llu blocks", o_blocks_count, n_blocks_count);
1970 if (n_blocks_count < o_blocks_count) {
1971 /* On-line shrinking not supported */
1972 ext4_warning(sb, "can't shrink FS - resize aborted");
1976 if (n_blocks_count == o_blocks_count)
1977 /* Nothing need to do */
1980 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1981 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1982 ext4_warning(sb, "resize would cause inodes_count overflow");
1985 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1987 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1988 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1990 meta_bg = ext4_has_feature_meta_bg(sb);
1992 if (ext4_has_feature_resize_inode(sb)) {
1994 ext4_error(sb, "resize_inode and meta_bg enabled "
1998 if (n_desc_blocks > o_desc_blocks +
1999 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2000 n_blocks_count_retry = n_blocks_count;
2001 n_desc_blocks = o_desc_blocks +
2002 le16_to_cpu(es->s_reserved_gdt_blocks);
2003 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2004 n_blocks_count = (ext4_fsblk_t)n_group *
2005 EXT4_BLOCKS_PER_GROUP(sb) +
2006 le32_to_cpu(es->s_first_data_block);
2007 n_group--; /* set to last group number */
2011 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2013 if (IS_ERR(resize_inode)) {
2014 ext4_warning(sb, "Error opening resize inode");
2015 return PTR_ERR(resize_inode);
2019 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2020 err = ext4_convert_meta_bg(sb, resize_inode);
2025 resize_inode = NULL;
2027 if (n_blocks_count_retry) {
2028 n_blocks_count = n_blocks_count_retry;
2029 n_blocks_count_retry = 0;
2035 * Make sure the last group has enough space so that it's
2036 * guaranteed to have enough space for all metadata blocks
2037 * that it might need to hold. (We might not need to store
2038 * the inode table blocks in the last block group, but there
2039 * will be cases where this might be needed.)
2041 if ((ext4_group_first_block_no(sb, n_group) +
2042 ext4_group_overhead_blocks(sb, n_group) + 2 +
2043 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2044 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2046 n_blocks_count_retry = 0;
2049 resize_inode = NULL;
2054 /* extend the last group */
2055 if (n_group == o_group)
2056 add = n_blocks_count - o_blocks_count;
2058 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
2060 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2065 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2068 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2072 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2076 flex_gd = alloc_flex_gd(flexbg_size);
2077 if (flex_gd == NULL) {
2082 /* Add flex groups. Note that a regular group is a
2083 * flex group with 1 group.
2085 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2087 if (jiffies - last_update_time > HZ * 10) {
2088 if (last_update_time)
2089 ext4_msg(sb, KERN_INFO,
2090 "resized to %llu blocks",
2091 ext4_blocks_count(es));
2092 last_update_time = jiffies;
2094 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2096 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2101 if (!err && n_blocks_count_retry) {
2102 n_blocks_count = n_blocks_count_retry;
2103 n_blocks_count_retry = 0;
2104 free_flex_gd(flex_gd);
2108 resize_inode = NULL;
2115 free_flex_gd(flex_gd);
2116 if (resize_inode != NULL)
2119 ext4_warning(sb, "error (%d) occurred during "
2120 "file system resize", err);
2121 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2122 ext4_blocks_count(es));