1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/resize.c
5 * Support for resizing an ext4 filesystem while it is mounted.
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
9 * This could probably be made into a module, because it is not often in use.
15 #include <linux/errno.h>
16 #include <linux/slab.h>
18 #include "ext4_jbd2.h"
25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
27 struct ext4_rcu_ptr *ptr;
29 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
34 void ext4_kvfree_array_rcu(void *to_free)
36 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
40 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
47 int ext4_resize_begin(struct super_block *sb)
49 struct ext4_sb_info *sbi = EXT4_SB(sb);
52 if (!capable(CAP_SYS_RESOURCE))
56 * If the reserved GDT blocks is non-zero, the resize_inode feature
57 * should always be set.
59 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
60 !ext4_has_feature_resize_inode(sb)) {
61 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
66 * If we are not using the primary superblock/GDT copy don't resize,
67 * because the user tools have no way of handling this. Probably a
68 * bad time to do it anyways.
70 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
71 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
72 ext4_warning(sb, "won't resize using backup superblock at %llu",
73 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
78 * We are not allowed to do online-resizing on a filesystem mounted
79 * with error, because it can destroy the filesystem easily.
81 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
82 ext4_warning(sb, "There are errors in the filesystem, "
83 "so online resizing is not allowed");
87 if (ext4_has_feature_sparse_super2(sb)) {
88 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
92 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
93 &EXT4_SB(sb)->s_ext4_flags))
99 void ext4_resize_end(struct super_block *sb)
101 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
102 smp_mb__after_atomic();
105 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
106 ext4_group_t group) {
107 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
108 EXT4_DESC_PER_BLOCK_BITS(sb);
111 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
112 ext4_group_t group) {
113 group = ext4_meta_bg_first_group(sb, group);
114 return ext4_group_first_block_no(sb, group);
117 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
118 ext4_group_t group) {
119 ext4_grpblk_t overhead;
120 overhead = ext4_bg_num_gdb(sb, group);
121 if (ext4_bg_has_super(sb, group))
123 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
127 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
128 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
130 static int verify_group_input(struct super_block *sb,
131 struct ext4_new_group_data *input)
133 struct ext4_sb_info *sbi = EXT4_SB(sb);
134 struct ext4_super_block *es = sbi->s_es;
135 ext4_fsblk_t start = ext4_blocks_count(es);
136 ext4_fsblk_t end = start + input->blocks_count;
137 ext4_group_t group = input->group;
138 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
140 ext4_fsblk_t metaend;
141 struct buffer_head *bh = NULL;
142 ext4_grpblk_t free_blocks_count, offset;
145 if (group != sbi->s_groups_count) {
146 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
147 input->group, sbi->s_groups_count);
151 overhead = ext4_group_overhead_blocks(sb, group);
152 metaend = start + overhead;
153 input->free_clusters_count = free_blocks_count =
154 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
156 if (test_opt(sb, DEBUG))
157 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
158 "(%d free, %u reserved)\n",
159 ext4_bg_has_super(sb, input->group) ? "normal" :
160 "no-super", input->group, input->blocks_count,
161 free_blocks_count, input->reserved_blocks);
163 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
165 ext4_warning(sb, "Last group not full");
166 else if (input->reserved_blocks > input->blocks_count / 5)
167 ext4_warning(sb, "Reserved blocks too high (%u)",
168 input->reserved_blocks);
169 else if (free_blocks_count < 0)
170 ext4_warning(sb, "Bad blocks count %u",
171 input->blocks_count);
172 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
175 ext4_warning(sb, "Cannot read last block (%llu)",
177 } else if (outside(input->block_bitmap, start, end))
178 ext4_warning(sb, "Block bitmap not in group (block %llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (outside(input->inode_bitmap, start, end))
181 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
182 (unsigned long long)input->inode_bitmap);
183 else if (outside(input->inode_table, start, end) ||
184 outside(itend - 1, start, end))
185 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
186 (unsigned long long)input->inode_table, itend - 1);
187 else if (input->inode_bitmap == input->block_bitmap)
188 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
189 (unsigned long long)input->block_bitmap);
190 else if (inside(input->block_bitmap, input->inode_table, itend))
191 ext4_warning(sb, "Block bitmap (%llu) in inode table "
193 (unsigned long long)input->block_bitmap,
194 (unsigned long long)input->inode_table, itend - 1);
195 else if (inside(input->inode_bitmap, input->inode_table, itend))
196 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
198 (unsigned long long)input->inode_bitmap,
199 (unsigned long long)input->inode_table, itend - 1);
200 else if (inside(input->block_bitmap, start, metaend))
201 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
202 (unsigned long long)input->block_bitmap,
204 else if (inside(input->inode_bitmap, start, metaend))
205 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
206 (unsigned long long)input->inode_bitmap,
208 else if (inside(input->inode_table, start, metaend) ||
209 inside(itend - 1, start, metaend))
210 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
212 (unsigned long long)input->inode_table,
213 itend - 1, start, metaend - 1);
222 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
225 struct ext4_new_flex_group_data {
226 struct ext4_new_group_data *groups; /* new_group_data for groups
228 __u16 *bg_flags; /* block group flags of groups
230 ext4_group_t count; /* number of groups in @groups
235 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
238 * Returns NULL on failure otherwise address of the allocated structure.
240 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
242 struct ext4_new_flex_group_data *flex_gd;
244 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
248 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
250 flex_gd->count = flexbg_size;
252 flex_gd->groups = kmalloc_array(flexbg_size,
253 sizeof(struct ext4_new_group_data),
255 if (flex_gd->groups == NULL)
258 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
260 if (flex_gd->bg_flags == NULL)
266 kfree(flex_gd->groups);
273 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
275 kfree(flex_gd->bg_flags);
276 kfree(flex_gd->groups);
281 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
282 * and inode tables for a flex group.
284 * This function is used by 64bit-resize. Note that this function allocates
285 * group tables from the 1st group of groups contained by @flexgd, which may
286 * be a partial of a flex group.
288 * @sb: super block of fs to which the groups belongs
290 * Returns 0 on a successful allocation of the metadata blocks in the
293 static int ext4_alloc_group_tables(struct super_block *sb,
294 struct ext4_new_flex_group_data *flex_gd,
297 struct ext4_new_group_data *group_data = flex_gd->groups;
298 ext4_fsblk_t start_blk;
299 ext4_fsblk_t last_blk;
300 ext4_group_t src_group;
301 ext4_group_t bb_index = 0;
302 ext4_group_t ib_index = 0;
303 ext4_group_t it_index = 0;
305 ext4_group_t last_group;
307 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
310 BUG_ON(flex_gd->count == 0 || group_data == NULL);
312 src_group = group_data[0].group;
313 last_group = src_group + flex_gd->count - 1;
315 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
316 (last_group & ~(flexbg_size - 1))));
318 group = group_data[0].group;
319 if (src_group >= group_data[0].group + flex_gd->count)
321 start_blk = ext4_group_first_block_no(sb, src_group);
322 last_blk = start_blk + group_data[src_group - group].blocks_count;
324 overhead = ext4_group_overhead_blocks(sb, src_group);
326 start_blk += overhead;
328 /* We collect contiguous blocks as much as possible. */
330 for (; src_group <= last_group; src_group++) {
331 overhead = ext4_group_overhead_blocks(sb, src_group);
333 last_blk += group_data[src_group - group].blocks_count;
338 /* Allocate block bitmaps */
339 for (; bb_index < flex_gd->count; bb_index++) {
340 if (start_blk >= last_blk)
342 group_data[bb_index].block_bitmap = start_blk++;
343 group = ext4_get_group_number(sb, start_blk - 1);
344 group -= group_data[0].group;
345 group_data[group].mdata_blocks++;
346 flex_gd->bg_flags[group] &= uninit_mask;
349 /* Allocate inode bitmaps */
350 for (; ib_index < flex_gd->count; ib_index++) {
351 if (start_blk >= last_blk)
353 group_data[ib_index].inode_bitmap = start_blk++;
354 group = ext4_get_group_number(sb, start_blk - 1);
355 group -= group_data[0].group;
356 group_data[group].mdata_blocks++;
357 flex_gd->bg_flags[group] &= uninit_mask;
360 /* Allocate inode tables */
361 for (; it_index < flex_gd->count; it_index++) {
362 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
363 ext4_fsblk_t next_group_start;
365 if (start_blk + itb > last_blk)
367 group_data[it_index].inode_table = start_blk;
368 group = ext4_get_group_number(sb, start_blk);
369 next_group_start = ext4_group_first_block_no(sb, group + 1);
370 group -= group_data[0].group;
372 if (start_blk + itb > next_group_start) {
373 flex_gd->bg_flags[group + 1] &= uninit_mask;
374 overhead = start_blk + itb - next_group_start;
375 group_data[group + 1].mdata_blocks += overhead;
379 group_data[group].mdata_blocks += itb;
380 flex_gd->bg_flags[group] &= uninit_mask;
381 start_blk += EXT4_SB(sb)->s_itb_per_group;
384 /* Update free clusters count to exclude metadata blocks */
385 for (i = 0; i < flex_gd->count; i++) {
386 group_data[i].free_clusters_count -=
387 EXT4_NUM_B2C(EXT4_SB(sb),
388 group_data[i].mdata_blocks);
391 if (test_opt(sb, DEBUG)) {
393 group = group_data[0].group;
395 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
396 "%d groups, flexbg size is %d:\n", flex_gd->count,
399 for (i = 0; i < flex_gd->count; i++) {
401 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
402 ext4_bg_has_super(sb, group + i) ? "normal" :
403 "no-super", group + i,
404 group_data[i].blocks_count,
405 group_data[i].free_clusters_count,
406 group_data[i].mdata_blocks);
412 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
415 struct buffer_head *bh;
418 bh = sb_getblk(sb, blk);
420 return ERR_PTR(-ENOMEM);
421 BUFFER_TRACE(bh, "get_write_access");
422 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
427 memset(bh->b_data, 0, sb->s_blocksize);
428 set_buffer_uptodate(bh);
434 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
436 return ext4_journal_ensure_credits_fn(handle, credits,
437 EXT4_MAX_TRANS_DATA, 0, 0);
441 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
443 * Helper function for ext4_setup_new_group_blocks() which set .
446 * @handle: journal handle
447 * @flex_gd: flex group data
449 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
450 struct ext4_new_flex_group_data *flex_gd,
451 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
453 struct ext4_sb_info *sbi = EXT4_SB(sb);
454 ext4_group_t count = last_cluster - first_cluster + 1;
457 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
459 for (count2 = count; count > 0;
460 count -= count2, first_cluster += count2) {
462 struct buffer_head *bh;
466 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
467 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
468 group -= flex_gd->groups[0].group;
470 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
474 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
475 BUG_ON(flex_gd->count > 1);
479 err = ext4_resize_ensure_credits_batch(handle, 1);
483 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
487 BUFFER_TRACE(bh, "get_write_access");
488 err = ext4_journal_get_write_access(handle, sb, bh,
494 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
495 first_cluster, first_cluster - start, count2);
496 ext4_set_bits(bh->b_data, first_cluster - start, count2);
498 err = ext4_handle_dirty_metadata(handle, NULL, bh);
508 * Set up the block and inode bitmaps, and the inode table for the new groups.
509 * This doesn't need to be part of the main transaction, since we are only
510 * changing blocks outside the actual filesystem. We still do journaling to
511 * ensure the recovery is correct in case of a failure just after resize.
512 * If any part of this fails, we simply abort the resize.
514 * setup_new_flex_group_blocks handles a flex group as follow:
515 * 1. copy super block and GDT, and initialize group tables if necessary.
516 * In this step, we only set bits in blocks bitmaps for blocks taken by
517 * super block and GDT.
518 * 2. allocate group tables in block bitmaps, that is, set bits in block
519 * bitmap for blocks taken by group tables.
521 static int setup_new_flex_group_blocks(struct super_block *sb,
522 struct ext4_new_flex_group_data *flex_gd)
524 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
527 struct ext4_sb_info *sbi = EXT4_SB(sb);
528 struct ext4_super_block *es = sbi->s_es;
529 struct ext4_new_group_data *group_data = flex_gd->groups;
530 __u16 *bg_flags = flex_gd->bg_flags;
532 ext4_group_t group, count;
533 struct buffer_head *bh = NULL;
534 int reserved_gdb, i, j, err = 0, err2;
537 BUG_ON(!flex_gd->count || !group_data ||
538 group_data[0].group != sbi->s_groups_count);
540 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
541 meta_bg = ext4_has_feature_meta_bg(sb);
543 /* This transaction may be extended/restarted along the way */
544 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
546 return PTR_ERR(handle);
548 group = group_data[0].group;
549 for (i = 0; i < flex_gd->count; i++, group++) {
550 unsigned long gdblocks;
551 ext4_grpblk_t overhead;
553 gdblocks = ext4_bg_num_gdb(sb, group);
554 start = ext4_group_first_block_no(sb, group);
556 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
560 ext4_group_t first_group;
561 first_group = ext4_meta_bg_first_group(sb, group);
562 if (first_group != group + 1 &&
563 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
567 block = start + ext4_bg_has_super(sb, group);
568 /* Copy all of the GDT blocks into the backup in this group */
569 for (j = 0; j < gdblocks; j++, block++) {
570 struct buffer_head *gdb;
572 ext4_debug("update backup group %#04llx\n", block);
573 err = ext4_resize_ensure_credits_batch(handle, 1);
577 gdb = sb_getblk(sb, block);
578 if (unlikely(!gdb)) {
583 BUFFER_TRACE(gdb, "get_write_access");
584 err = ext4_journal_get_write_access(handle, sb, 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 = ext4_resize_ensure_credits_batch(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,
646 EXT4_NUM_B2C(sbi, overhead));
648 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
649 sb->s_blocksize * 8, bh->b_data);
650 err = ext4_handle_dirty_metadata(handle, NULL, bh);
656 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
659 /* Initialize inode bitmap of the @group */
660 block = group_data[i].inode_bitmap;
661 err = ext4_resize_ensure_credits_batch(handle, 1);
664 /* Mark unused entries in inode bitmap used */
665 bh = bclean(handle, sb, block);
671 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
672 sb->s_blocksize * 8, bh->b_data);
673 err = ext4_handle_dirty_metadata(handle, NULL, bh);
679 /* Mark group tables in block bitmap */
680 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
681 count = group_table_count[j];
682 start = (&group_data[0].block_bitmap)[j];
684 for (i = 1; i < flex_gd->count; i++) {
685 block += group_table_count[j];
686 if (block == (&group_data[i].block_bitmap)[j]) {
687 count += group_table_count[j];
690 err = set_flexbg_block_bitmap(sb, handle,
692 EXT4_B2C(sbi, start),
698 count = group_table_count[j];
699 start = (&group_data[i].block_bitmap)[j];
704 err = set_flexbg_block_bitmap(sb, handle,
706 EXT4_B2C(sbi, start),
716 err2 = ext4_journal_stop(handle);
724 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
725 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
726 * calling this for the first time. In a sparse filesystem it will be the
727 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
728 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
730 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
731 unsigned *five, unsigned *seven)
733 unsigned *min = three;
737 if (!ext4_has_feature_sparse_super(sb)) {
759 * Check that all of the backup GDT blocks are held in the primary GDT block.
760 * It is assumed that they are stored in group order. Returns the number of
761 * groups in current filesystem that have BACKUPS, or -ve error code.
763 static int verify_reserved_gdb(struct super_block *sb,
765 struct buffer_head *primary)
767 const ext4_fsblk_t blk = primary->b_blocknr;
772 __le32 *p = (__le32 *)primary->b_data;
775 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
776 if (le32_to_cpu(*p++) !=
777 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
778 ext4_warning(sb, "reserved GDT %llu"
779 " missing grp %d (%llu)",
782 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
786 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
794 * Called when we need to bring a reserved group descriptor table block into
795 * use from the resize inode. The primary copy of the new GDT block currently
796 * is an indirect block (under the double indirect block in the resize inode).
797 * The new backup GDT blocks will be stored as leaf blocks in this indirect
798 * block, in group order. Even though we know all the block numbers we need,
799 * we check to ensure that the resize inode has actually reserved these blocks.
801 * Don't need to update the block bitmaps because the blocks are still in use.
803 * We get all of the error cases out of the way, so that we are sure to not
804 * fail once we start modifying the data on disk, because JBD has no rollback.
806 static int add_new_gdb(handle_t *handle, struct inode *inode,
809 struct super_block *sb = inode->i_sb;
810 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
811 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
812 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
813 struct buffer_head **o_group_desc, **n_group_desc = NULL;
814 struct buffer_head *dind = NULL;
815 struct buffer_head *gdb_bh = NULL;
817 struct ext4_iloc iloc = { .bh = NULL };
821 if (test_opt(sb, DEBUG))
823 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
826 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
828 return PTR_ERR(gdb_bh);
830 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
836 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
837 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
844 data = (__le32 *)dind->b_data;
845 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
846 ext4_warning(sb, "new group %u GDT block %llu not reserved",
852 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
853 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
858 BUFFER_TRACE(gdb_bh, "get_write_access");
859 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
863 BUFFER_TRACE(dind, "get_write_access");
864 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
866 ext4_std_error(sb, err);
870 /* ext4_reserve_inode_write() gets a reference on the iloc */
871 err = ext4_reserve_inode_write(handle, inode, &iloc);
875 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
879 ext4_warning(sb, "not enough memory for %lu groups",
885 * Finally, we have all of the possible failures behind us...
887 * Remove new GDT block from inode double-indirect block and clear out
888 * the new GDT block for use (which also "frees" the backup GDT blocks
889 * from the reserved inode). We don't need to change the bitmaps for
890 * these blocks, because they are marked as in-use from being in the
891 * reserved inode, and will become GDT blocks (primary and backup).
893 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
894 err = ext4_handle_dirty_metadata(handle, NULL, dind);
896 ext4_std_error(sb, err);
899 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
900 (9 - EXT4_SB(sb)->s_cluster_bits);
901 ext4_mark_iloc_dirty(handle, inode, &iloc);
902 memset(gdb_bh->b_data, 0, sb->s_blocksize);
903 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
905 ext4_std_error(sb, err);
912 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
913 memcpy(n_group_desc, o_group_desc,
914 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
916 n_group_desc[gdb_num] = gdb_bh;
917 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
918 EXT4_SB(sb)->s_gdb_count++;
919 ext4_kvfree_array_rcu(o_group_desc);
921 lock_buffer(EXT4_SB(sb)->s_sbh);
922 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
923 ext4_superblock_csum_set(sb);
924 unlock_buffer(EXT4_SB(sb)->s_sbh);
925 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
927 ext4_std_error(sb, err);
930 kvfree(n_group_desc);
935 ext4_debug("leaving with error %d\n", err);
940 * add_new_gdb_meta_bg is the sister of add_new_gdb.
942 static int add_new_gdb_meta_bg(struct super_block *sb,
943 handle_t *handle, ext4_group_t group) {
944 ext4_fsblk_t gdblock;
945 struct buffer_head *gdb_bh;
946 struct buffer_head **o_group_desc, **n_group_desc;
947 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
950 gdblock = ext4_meta_bg_first_block_no(sb, group) +
951 ext4_bg_has_super(sb, group);
952 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
954 return PTR_ERR(gdb_bh);
955 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
960 ext4_warning(sb, "not enough memory for %lu groups",
966 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
967 memcpy(n_group_desc, o_group_desc,
968 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
970 n_group_desc[gdb_num] = gdb_bh;
972 BUFFER_TRACE(gdb_bh, "get_write_access");
973 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
975 kvfree(n_group_desc);
980 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
981 EXT4_SB(sb)->s_gdb_count++;
982 ext4_kvfree_array_rcu(o_group_desc);
987 * Called when we are adding a new group which has a backup copy of each of
988 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
989 * We need to add these reserved backup GDT blocks to the resize inode, so
990 * that they are kept for future resizing and not allocated to files.
992 * Each reserved backup GDT block will go into a different indirect block.
993 * The indirect blocks are actually the primary reserved GDT blocks,
994 * so we know in advance what their block numbers are. We only get the
995 * double-indirect block to verify it is pointing to the primary reserved
996 * GDT blocks so we don't overwrite a data block by accident. The reserved
997 * backup GDT blocks are stored in their reserved primary GDT block.
999 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1002 struct super_block *sb = inode->i_sb;
1003 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1004 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1005 struct buffer_head **primary;
1006 struct buffer_head *dind;
1007 struct ext4_iloc iloc;
1014 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1018 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1019 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1021 err = PTR_ERR(dind);
1026 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1027 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1028 EXT4_ADDR_PER_BLOCK(sb));
1029 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1031 /* Get each reserved primary GDT block and verify it holds backups */
1032 for (res = 0; res < reserved_gdb; res++, blk++) {
1033 if (le32_to_cpu(*data) != blk) {
1034 ext4_warning(sb, "reserved block %llu"
1035 " not at offset %ld",
1037 (long)(data - (__le32 *)dind->b_data));
1041 primary[res] = ext4_sb_bread(sb, blk, 0);
1042 if (IS_ERR(primary[res])) {
1043 err = PTR_ERR(primary[res]);
1044 primary[res] = NULL;
1047 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1048 if (gdbackups < 0) {
1049 brelse(primary[res]);
1054 data = (__le32 *)dind->b_data;
1057 for (i = 0; i < reserved_gdb; i++) {
1058 BUFFER_TRACE(primary[i], "get_write_access");
1059 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1064 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1068 * Finally we can add each of the reserved backup GDT blocks from
1069 * the new group to its reserved primary GDT block.
1071 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1072 for (i = 0; i < reserved_gdb; i++) {
1074 data = (__le32 *)primary[i]->b_data;
1075 /* printk("reserving backup %lu[%u] = %lu\n",
1076 primary[i]->b_blocknr, gdbackups,
1077 blk + primary[i]->b_blocknr); */
1078 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1079 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1084 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1085 ext4_mark_iloc_dirty(handle, inode, &iloc);
1089 brelse(primary[res]);
1099 * Update the backup copies of the ext4 metadata. These don't need to be part
1100 * of the main resize transaction, because e2fsck will re-write them if there
1101 * is a problem (basically only OOM will cause a problem). However, we
1102 * _should_ update the backups if possible, in case the primary gets trashed
1103 * for some reason and we need to run e2fsck from a backup superblock. The
1104 * important part is that the new block and inode counts are in the backup
1105 * superblocks, and the location of the new group metadata in the GDT backups.
1107 * We do not need take the s_resize_lock for this, because these
1108 * blocks are not otherwise touched by the filesystem code when it is
1109 * mounted. We don't need to worry about last changing from
1110 * sbi->s_groups_count, because the worst that can happen is that we
1111 * do not copy the full number of backups at this time. The resize
1112 * which changed s_groups_count will backup again.
1114 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1115 int size, int meta_bg)
1117 struct ext4_sb_info *sbi = EXT4_SB(sb);
1119 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1123 ext4_group_t group = 0;
1124 int rest = sb->s_blocksize - size;
1128 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1129 if (IS_ERR(handle)) {
1131 err = PTR_ERR(handle);
1136 group = ext4_list_backups(sb, &three, &five, &seven);
1137 last = sbi->s_groups_count;
1139 group = ext4_get_group_number(sb, blk_off) + 1;
1140 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1143 while (group < sbi->s_groups_count) {
1144 struct buffer_head *bh;
1145 ext4_fsblk_t backup_block;
1147 /* Out of journal space, and can't get more - abort - so sad */
1148 err = ext4_resize_ensure_credits_batch(handle, 1);
1153 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1155 backup_block = (ext4_group_first_block_no(sb, group) +
1156 ext4_bg_has_super(sb, group));
1158 bh = sb_getblk(sb, backup_block);
1159 if (unlikely(!bh)) {
1163 ext4_debug("update metadata backup %llu(+%llu)\n",
1164 backup_block, backup_block -
1165 ext4_group_first_block_no(sb, group));
1166 BUFFER_TRACE(bh, "get_write_access");
1167 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1171 memcpy(bh->b_data, data, size);
1173 memset(bh->b_data + size, 0, rest);
1174 set_buffer_uptodate(bh);
1176 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1178 ext4_std_error(sb, err);
1182 group = ext4_list_backups(sb, &three, &five, &seven);
1183 else if (group == last)
1188 if ((err2 = ext4_journal_stop(handle)) && !err)
1192 * Ugh! Need to have e2fsck write the backup copies. It is too
1193 * late to revert the resize, we shouldn't fail just because of
1194 * the backup copies (they are only needed in case of corruption).
1196 * However, if we got here we have a journal problem too, so we
1197 * can't really start a transaction to mark the superblock.
1198 * Chicken out and just set the flag on the hope it will be written
1199 * to disk, and if not - we will simply wait until next fsck.
1203 ext4_warning(sb, "can't update backup for group %u (err %d), "
1204 "forcing fsck on next reboot", group, err);
1205 sbi->s_mount_state &= ~EXT4_VALID_FS;
1206 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1207 mark_buffer_dirty(sbi->s_sbh);
1212 * ext4_add_new_descs() adds @count group descriptor of groups
1213 * starting at @group
1215 * @handle: journal handle
1217 * @group: the group no. of the first group desc to be added
1218 * @resize_inode: the resize inode
1219 * @count: number of group descriptors to be added
1221 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1222 ext4_group_t group, struct inode *resize_inode,
1225 struct ext4_sb_info *sbi = EXT4_SB(sb);
1226 struct ext4_super_block *es = sbi->s_es;
1227 struct buffer_head *gdb_bh;
1228 int i, gdb_off, gdb_num, err = 0;
1231 meta_bg = ext4_has_feature_meta_bg(sb);
1232 for (i = 0; i < count; i++, group++) {
1233 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1234 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1236 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1237 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1240 * We will only either add reserved group blocks to a backup group
1241 * or remove reserved blocks for the first group in a new group block.
1242 * Doing both would be mean more complex code, and sane people don't
1243 * use non-sparse filesystems anymore. This is already checked above.
1246 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1248 BUFFER_TRACE(gdb_bh, "get_write_access");
1249 err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1252 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1253 err = reserve_backup_gdb(handle, resize_inode, group);
1254 } else if (meta_bg != 0) {
1255 err = add_new_gdb_meta_bg(sb, handle, group);
1257 err = add_new_gdb(handle, resize_inode, group);
1265 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1267 struct buffer_head *bh = sb_getblk(sb, block);
1270 if (!bh_uptodate_or_lock(bh)) {
1271 if (ext4_read_bh(bh, 0, NULL) < 0) {
1280 static int ext4_set_bitmap_checksums(struct super_block *sb,
1282 struct ext4_group_desc *gdp,
1283 struct ext4_new_group_data *group_data)
1285 struct buffer_head *bh;
1287 if (!ext4_has_metadata_csum(sb))
1290 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1293 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1294 EXT4_INODES_PER_GROUP(sb) / 8);
1297 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1300 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1307 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1309 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1310 struct ext4_new_flex_group_data *flex_gd)
1312 struct ext4_new_group_data *group_data = flex_gd->groups;
1313 struct ext4_group_desc *gdp;
1314 struct ext4_sb_info *sbi = EXT4_SB(sb);
1315 struct buffer_head *gdb_bh;
1317 __u16 *bg_flags = flex_gd->bg_flags;
1318 int i, gdb_off, gdb_num, err = 0;
1321 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1322 group = group_data->group;
1324 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1325 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1328 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1330 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1331 /* Update group descriptor block for new group */
1332 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1333 gdb_off * EXT4_DESC_SIZE(sb));
1335 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1336 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1337 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1338 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1340 ext4_std_error(sb, err);
1344 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1345 ext4_free_group_clusters_set(sb, gdp,
1346 group_data->free_clusters_count);
1347 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1348 if (ext4_has_group_desc_csum(sb))
1349 ext4_itable_unused_set(sb, gdp,
1350 EXT4_INODES_PER_GROUP(sb));
1351 gdp->bg_flags = cpu_to_le16(*bg_flags);
1352 ext4_group_desc_csum_set(sb, group, gdp);
1354 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1355 if (unlikely(err)) {
1356 ext4_std_error(sb, err);
1361 * We can allocate memory for mb_alloc based on the new group
1364 err = ext4_mb_add_groupinfo(sb, group, gdp);
1372 * ext4_update_super() updates the super block so that the newly added
1373 * groups can be seen by the filesystem.
1376 * @flex_gd: new added groups
1378 static void ext4_update_super(struct super_block *sb,
1379 struct ext4_new_flex_group_data *flex_gd)
1381 ext4_fsblk_t blocks_count = 0;
1382 ext4_fsblk_t free_blocks = 0;
1383 ext4_fsblk_t reserved_blocks = 0;
1384 struct ext4_new_group_data *group_data = flex_gd->groups;
1385 struct ext4_sb_info *sbi = EXT4_SB(sb);
1386 struct ext4_super_block *es = sbi->s_es;
1389 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1391 * Make the new blocks and inodes valid next. We do this before
1392 * increasing the group count so that once the group is enabled,
1393 * all of its blocks and inodes are already valid.
1395 * We always allocate group-by-group, then block-by-block or
1396 * inode-by-inode within a group, so enabling these
1397 * blocks/inodes before the group is live won't actually let us
1398 * allocate the new space yet.
1400 for (i = 0; i < flex_gd->count; i++) {
1401 blocks_count += group_data[i].blocks_count;
1402 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1405 reserved_blocks = ext4_r_blocks_count(es) * 100;
1406 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1407 reserved_blocks *= blocks_count;
1408 do_div(reserved_blocks, 100);
1410 lock_buffer(sbi->s_sbh);
1411 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1412 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1413 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1415 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1418 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1420 * We need to protect s_groups_count against other CPUs seeing
1421 * inconsistent state in the superblock.
1423 * The precise rules we use are:
1425 * * Writers must perform a smp_wmb() after updating all
1426 * dependent data and before modifying the groups count
1428 * * Readers must perform an smp_rmb() after reading the groups
1429 * count and before reading any dependent data.
1431 * NB. These rules can be relaxed when checking the group count
1432 * while freeing data, as we can only allocate from a block
1433 * group after serialising against the group count, and we can
1434 * only then free after serialising in turn against that
1439 /* Update the global fs size fields */
1440 sbi->s_groups_count += flex_gd->count;
1441 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1442 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1444 /* Update the reserved block counts only once the new group is
1446 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1449 /* Update the free space counts */
1450 percpu_counter_add(&sbi->s_freeclusters_counter,
1451 EXT4_NUM_B2C(sbi, free_blocks));
1452 percpu_counter_add(&sbi->s_freeinodes_counter,
1453 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1455 ext4_debug("free blocks count %llu",
1456 percpu_counter_read(&sbi->s_freeclusters_counter));
1457 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1458 ext4_group_t flex_group;
1459 struct flex_groups *fg;
1461 flex_group = ext4_flex_group(sbi, group_data[0].group);
1462 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1463 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1464 &fg->free_clusters);
1465 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1470 * Update the fs overhead information
1472 ext4_calculate_overhead(sb);
1473 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1475 ext4_superblock_csum_set(sb);
1476 unlock_buffer(sbi->s_sbh);
1477 if (test_opt(sb, DEBUG))
1478 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1479 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1480 blocks_count, free_blocks, reserved_blocks);
1483 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1484 * _before_ we start modifying the filesystem, because we cannot abort the
1485 * transaction and not have it write the data to disk.
1487 static int ext4_flex_group_add(struct super_block *sb,
1488 struct inode *resize_inode,
1489 struct ext4_new_flex_group_data *flex_gd)
1491 struct ext4_sb_info *sbi = EXT4_SB(sb);
1492 struct ext4_super_block *es = sbi->s_es;
1493 ext4_fsblk_t o_blocks_count;
1497 unsigned reserved_gdb;
1498 int err = 0, err2 = 0, credit;
1500 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1502 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1503 o_blocks_count = ext4_blocks_count(es);
1504 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1507 err = setup_new_flex_group_blocks(sb, flex_gd);
1511 * We will always be modifying at least the superblock and GDT
1512 * blocks. If we are adding a group past the last current GDT block,
1513 * we will also modify the inode and the dindirect block. If we
1514 * are adding a group with superblock/GDT backups we will also
1515 * modify each of the reserved GDT dindirect blocks.
1517 credit = 3; /* sb, resize inode, resize inode dindirect */
1519 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1520 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1521 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1522 if (IS_ERR(handle)) {
1523 err = PTR_ERR(handle);
1527 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1528 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1533 group = flex_gd->groups[0].group;
1534 BUG_ON(group != sbi->s_groups_count);
1535 err = ext4_add_new_descs(handle, sb, group,
1536 resize_inode, flex_gd->count);
1540 err = ext4_setup_new_descs(handle, sb, flex_gd);
1544 ext4_update_super(sb, flex_gd);
1546 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1549 err2 = ext4_journal_stop(handle);
1554 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1555 int gdb_num_end = ((group + flex_gd->count - 1) /
1556 EXT4_DESC_PER_BLOCK(sb));
1557 int meta_bg = ext4_has_feature_meta_bg(sb);
1558 sector_t old_gdb = 0;
1560 update_backups(sb, ext4_group_first_block_no(sb, 0),
1561 (char *)es, sizeof(struct ext4_super_block), 0);
1562 for (; gdb_num <= gdb_num_end; gdb_num++) {
1563 struct buffer_head *gdb_bh;
1565 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1567 if (old_gdb == gdb_bh->b_blocknr)
1569 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1570 gdb_bh->b_size, meta_bg);
1571 old_gdb = gdb_bh->b_blocknr;
1578 static int ext4_setup_next_flex_gd(struct super_block *sb,
1579 struct ext4_new_flex_group_data *flex_gd,
1580 ext4_fsblk_t n_blocks_count,
1581 unsigned long flexbg_size)
1583 struct ext4_sb_info *sbi = EXT4_SB(sb);
1584 struct ext4_super_block *es = sbi->s_es;
1585 struct ext4_new_group_data *group_data = flex_gd->groups;
1586 ext4_fsblk_t o_blocks_count;
1587 ext4_group_t n_group;
1589 ext4_group_t last_group;
1591 ext4_grpblk_t clusters_per_group;
1594 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1596 o_blocks_count = ext4_blocks_count(es);
1598 if (o_blocks_count == n_blocks_count)
1601 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1603 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1605 last_group = group | (flexbg_size - 1);
1606 if (last_group > n_group)
1607 last_group = n_group;
1609 flex_gd->count = last_group - group + 1;
1611 for (i = 0; i < flex_gd->count; i++) {
1614 group_data[i].group = group + i;
1615 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1616 overhead = ext4_group_overhead_blocks(sb, group + i);
1617 group_data[i].mdata_blocks = overhead;
1618 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1619 if (ext4_has_group_desc_csum(sb)) {
1620 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1621 EXT4_BG_INODE_UNINIT;
1622 if (!test_opt(sb, INIT_INODE_TABLE))
1623 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1625 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1628 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1629 /* We need to initialize block bitmap of last group. */
1630 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1632 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1633 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1634 group_data[i - 1].free_clusters_count -= clusters_per_group -
1641 /* Add group descriptor data to an existing or new group descriptor block.
1642 * Ensure we handle all possible error conditions _before_ we start modifying
1643 * the filesystem, because we cannot abort the transaction and not have it
1644 * write the data to disk.
1646 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1647 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1649 * We only need to hold the superblock lock while we are actually adding
1650 * in the new group's counts to the superblock. Prior to that we have
1651 * not really "added" the group at all. We re-check that we are still
1652 * adding in the last group in case things have changed since verifying.
1654 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1656 struct ext4_new_flex_group_data flex_gd;
1657 struct ext4_sb_info *sbi = EXT4_SB(sb);
1658 struct ext4_super_block *es = sbi->s_es;
1659 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1660 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1661 struct inode *inode = NULL;
1666 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1668 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1669 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1673 if (ext4_blocks_count(es) + input->blocks_count <
1674 ext4_blocks_count(es)) {
1675 ext4_warning(sb, "blocks_count overflow");
1679 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1680 le32_to_cpu(es->s_inodes_count)) {
1681 ext4_warning(sb, "inodes_count overflow");
1685 if (reserved_gdb || gdb_off == 0) {
1686 if (!ext4_has_feature_resize_inode(sb) ||
1687 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1689 "No reserved GDT blocks, can't resize");
1692 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1693 if (IS_ERR(inode)) {
1694 ext4_warning(sb, "Error opening resize inode");
1695 return PTR_ERR(inode);
1700 err = verify_group_input(sb, input);
1704 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1708 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1713 flex_gd.groups = input;
1714 flex_gd.bg_flags = &bg_flags;
1715 err = ext4_flex_group_add(sb, inode, &flex_gd);
1719 } /* ext4_group_add */
1722 * extend a group without checking assuming that checking has been done.
1724 static int ext4_group_extend_no_check(struct super_block *sb,
1725 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1727 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1731 /* We will update the superblock, one block bitmap, and
1732 * one group descriptor via ext4_group_add_blocks().
1734 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1735 if (IS_ERR(handle)) {
1736 err = PTR_ERR(handle);
1737 ext4_warning(sb, "error %d on journal start", err);
1741 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1742 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1745 ext4_warning(sb, "error %d on journal write access", err);
1749 lock_buffer(EXT4_SB(sb)->s_sbh);
1750 ext4_blocks_count_set(es, o_blocks_count + add);
1751 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1752 ext4_superblock_csum_set(sb);
1753 unlock_buffer(EXT4_SB(sb)->s_sbh);
1754 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1755 o_blocks_count + add);
1756 /* We add the blocks to the bitmap and set the group need init bit */
1757 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1760 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1761 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1762 o_blocks_count + add);
1764 err2 = ext4_journal_stop(handle);
1769 if (test_opt(sb, DEBUG))
1770 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1771 "blocks\n", ext4_blocks_count(es));
1772 update_backups(sb, ext4_group_first_block_no(sb, 0),
1773 (char *)es, sizeof(struct ext4_super_block), 0);
1779 * Extend the filesystem to the new number of blocks specified. This entry
1780 * point is only used to extend the current filesystem to the end of the last
1781 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1782 * for emergencies (because it has no dependencies on reserved blocks).
1784 * If we _really_ wanted, we could use default values to call ext4_group_add()
1785 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1786 * GDT blocks are reserved to grow to the desired size.
1788 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1789 ext4_fsblk_t n_blocks_count)
1791 ext4_fsblk_t o_blocks_count;
1794 struct buffer_head *bh;
1798 o_blocks_count = ext4_blocks_count(es);
1800 if (test_opt(sb, DEBUG))
1801 ext4_msg(sb, KERN_DEBUG,
1802 "extending last group from %llu to %llu blocks",
1803 o_blocks_count, n_blocks_count);
1805 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1808 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1809 ext4_msg(sb, KERN_ERR,
1810 "filesystem too large to resize to %llu blocks safely",
1815 if (n_blocks_count < o_blocks_count) {
1816 ext4_warning(sb, "can't shrink FS - resize aborted");
1820 /* Handle the remaining blocks in the last group only. */
1821 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1824 ext4_warning(sb, "need to use ext2online to resize further");
1828 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1830 if (o_blocks_count + add < o_blocks_count) {
1831 ext4_warning(sb, "blocks_count overflow");
1835 if (o_blocks_count + add > n_blocks_count)
1836 add = n_blocks_count - o_blocks_count;
1838 if (o_blocks_count + add < n_blocks_count)
1839 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1840 o_blocks_count + add, add);
1842 /* See if the device is actually as big as what was requested */
1843 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1845 ext4_warning(sb, "can't read last block, resize aborted");
1850 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1852 } /* ext4_group_extend */
1855 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1857 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1861 * Release the resize inode and drop the resize_inode feature if there
1862 * are no more reserved gdt blocks, and then convert the file system
1865 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1868 struct ext4_sb_info *sbi = EXT4_SB(sb);
1869 struct ext4_super_block *es = sbi->s_es;
1870 struct ext4_inode_info *ei = EXT4_I(inode);
1872 int i, ret, err = 0;
1875 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1877 if (es->s_reserved_gdt_blocks) {
1878 ext4_error(sb, "Unexpected non-zero "
1879 "s_reserved_gdt_blocks");
1883 /* Do a quick sanity check of the resize inode */
1884 if (inode->i_blocks != 1 << (inode->i_blkbits -
1885 (9 - sbi->s_cluster_bits)))
1886 goto invalid_resize_inode;
1887 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1888 if (i == EXT4_DIND_BLOCK) {
1892 goto invalid_resize_inode;
1895 goto invalid_resize_inode;
1897 credits += 3; /* block bitmap, bg descriptor, resize inode */
1900 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1902 return PTR_ERR(handle);
1904 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1905 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1910 lock_buffer(sbi->s_sbh);
1911 ext4_clear_feature_resize_inode(sb);
1912 ext4_set_feature_meta_bg(sb);
1913 sbi->s_es->s_first_meta_bg =
1914 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1915 ext4_superblock_csum_set(sb);
1916 unlock_buffer(sbi->s_sbh);
1918 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1920 ext4_std_error(sb, err);
1925 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1926 ext4_free_blocks(handle, inode, NULL, nr, 1,
1927 EXT4_FREE_BLOCKS_METADATA |
1928 EXT4_FREE_BLOCKS_FORGET);
1929 ei->i_data[EXT4_DIND_BLOCK] = 0;
1930 inode->i_blocks = 0;
1932 err = ext4_mark_inode_dirty(handle, inode);
1934 ext4_std_error(sb, err);
1938 ret = ext4_journal_stop(handle);
1943 invalid_resize_inode:
1944 ext4_error(sb, "corrupted/inconsistent resize inode");
1949 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1951 * @sb: super block of the fs to be resized
1952 * @n_blocks_count: the number of blocks resides in the resized fs
1954 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1956 struct ext4_new_flex_group_data *flex_gd = NULL;
1957 struct ext4_sb_info *sbi = EXT4_SB(sb);
1958 struct ext4_super_block *es = sbi->s_es;
1959 struct buffer_head *bh;
1960 struct inode *resize_inode = NULL;
1961 ext4_grpblk_t add, offset;
1962 unsigned long n_desc_blocks;
1963 unsigned long o_desc_blocks;
1964 ext4_group_t o_group;
1965 ext4_group_t n_group;
1966 ext4_fsblk_t o_blocks_count;
1967 ext4_fsblk_t n_blocks_count_retry = 0;
1968 unsigned long last_update_time = 0;
1969 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1972 /* See if the device is actually as big as what was requested */
1973 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1975 ext4_warning(sb, "can't read last block, resize aborted");
1981 * For bigalloc, trim the requested size to the nearest cluster
1982 * boundary to avoid creating an unusable filesystem. We do this
1983 * silently, instead of returning an error, to avoid breaking
1984 * callers that blindly resize the filesystem to the full size of
1985 * the underlying block device.
1987 if (ext4_has_feature_bigalloc(sb))
1988 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1991 o_blocks_count = ext4_blocks_count(es);
1993 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1994 "to %llu blocks", o_blocks_count, n_blocks_count);
1996 if (n_blocks_count < o_blocks_count) {
1997 /* On-line shrinking not supported */
1998 ext4_warning(sb, "can't shrink FS - resize aborted");
2002 if (n_blocks_count == o_blocks_count)
2003 /* Nothing need to do */
2006 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2007 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2008 ext4_warning(sb, "resize would cause inodes_count overflow");
2011 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2013 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2014 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2016 meta_bg = ext4_has_feature_meta_bg(sb);
2018 if (ext4_has_feature_resize_inode(sb)) {
2020 ext4_error(sb, "resize_inode and meta_bg enabled "
2024 if (n_desc_blocks > o_desc_blocks +
2025 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2026 n_blocks_count_retry = n_blocks_count;
2027 n_desc_blocks = o_desc_blocks +
2028 le16_to_cpu(es->s_reserved_gdt_blocks);
2029 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2030 n_blocks_count = (ext4_fsblk_t)n_group *
2031 EXT4_BLOCKS_PER_GROUP(sb) +
2032 le32_to_cpu(es->s_first_data_block);
2033 n_group--; /* set to last group number */
2037 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2039 if (IS_ERR(resize_inode)) {
2040 ext4_warning(sb, "Error opening resize inode");
2041 return PTR_ERR(resize_inode);
2045 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2046 err = ext4_convert_meta_bg(sb, resize_inode);
2051 resize_inode = NULL;
2053 if (n_blocks_count_retry) {
2054 n_blocks_count = n_blocks_count_retry;
2055 n_blocks_count_retry = 0;
2061 * Make sure the last group has enough space so that it's
2062 * guaranteed to have enough space for all metadata blocks
2063 * that it might need to hold. (We might not need to store
2064 * the inode table blocks in the last block group, but there
2065 * will be cases where this might be needed.)
2067 if ((ext4_group_first_block_no(sb, n_group) +
2068 ext4_group_overhead_blocks(sb, n_group) + 2 +
2069 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2070 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2072 n_blocks_count_retry = 0;
2075 resize_inode = NULL;
2080 /* extend the last group */
2081 if (n_group == o_group)
2082 add = n_blocks_count - o_blocks_count;
2084 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2086 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2091 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2094 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2098 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2102 flex_gd = alloc_flex_gd(flexbg_size);
2103 if (flex_gd == NULL) {
2108 /* Add flex groups. Note that a regular group is a
2109 * flex group with 1 group.
2111 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2113 if (jiffies - last_update_time > HZ * 10) {
2114 if (last_update_time)
2115 ext4_msg(sb, KERN_INFO,
2116 "resized to %llu blocks",
2117 ext4_blocks_count(es));
2118 last_update_time = jiffies;
2120 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2122 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2127 if (!err && n_blocks_count_retry) {
2128 n_blocks_count = n_blocks_count_retry;
2129 n_blocks_count_retry = 0;
2130 free_flex_gd(flex_gd);
2134 resize_inode = NULL;
2141 free_flex_gd(flex_gd);
2142 if (resize_inode != NULL)
2145 ext4_warning(sb, "error (%d) occurred during "
2146 "file system resize", err);
2147 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2148 ext4_blocks_count(es));