1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/resize.c
5 * Support for resizing an ext4 filesystem while it is mounted.
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
9 * This could probably be made into a module, because it is not often in use.
15 #include <linux/errno.h>
16 #include <linux/slab.h>
18 #include "ext4_jbd2.h"
25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
27 struct ext4_rcu_ptr *ptr;
29 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
34 void ext4_kvfree_array_rcu(void *to_free)
36 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
40 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
47 int ext4_resize_begin(struct super_block *sb)
49 struct ext4_sb_info *sbi = EXT4_SB(sb);
52 if (!capable(CAP_SYS_RESOURCE))
56 * If we are not using the primary superblock/GDT copy don't resize,
57 * because the user tools have no way of handling this. Probably a
58 * bad time to do it anyways.
60 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
61 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
62 ext4_warning(sb, "won't resize using backup superblock at %llu",
63 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
68 * We are not allowed to do online-resizing on a filesystem mounted
69 * with error, because it can destroy the filesystem easily.
71 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
72 ext4_warning(sb, "There are errors in the filesystem, "
73 "so online resizing is not allowed");
77 if (ext4_has_feature_bigalloc(sb)) {
78 ext4_msg(sb, KERN_ERR, "Online resizing not supported with bigalloc");
81 if (ext4_has_feature_sparse_super2(sb)) {
82 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
86 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
87 &EXT4_SB(sb)->s_ext4_flags))
93 void ext4_resize_end(struct super_block *sb)
95 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
96 smp_mb__after_atomic();
99 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
100 ext4_group_t group) {
101 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
102 EXT4_DESC_PER_BLOCK_BITS(sb);
105 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
106 ext4_group_t group) {
107 group = ext4_meta_bg_first_group(sb, group);
108 return ext4_group_first_block_no(sb, group);
111 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
112 ext4_group_t group) {
113 ext4_grpblk_t overhead;
114 overhead = ext4_bg_num_gdb(sb, group);
115 if (ext4_bg_has_super(sb, group))
117 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
121 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
122 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
124 static int verify_group_input(struct super_block *sb,
125 struct ext4_new_group_data *input)
127 struct ext4_sb_info *sbi = EXT4_SB(sb);
128 struct ext4_super_block *es = sbi->s_es;
129 ext4_fsblk_t start = ext4_blocks_count(es);
130 ext4_fsblk_t end = start + input->blocks_count;
131 ext4_group_t group = input->group;
132 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
134 ext4_fsblk_t metaend;
135 struct buffer_head *bh = NULL;
136 ext4_grpblk_t free_blocks_count, offset;
139 if (group != sbi->s_groups_count) {
140 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
141 input->group, sbi->s_groups_count);
145 overhead = ext4_group_overhead_blocks(sb, group);
146 metaend = start + overhead;
147 input->free_clusters_count = free_blocks_count =
148 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
150 if (test_opt(sb, DEBUG))
151 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
152 "(%d free, %u reserved)\n",
153 ext4_bg_has_super(sb, input->group) ? "normal" :
154 "no-super", input->group, input->blocks_count,
155 free_blocks_count, input->reserved_blocks);
157 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
159 ext4_warning(sb, "Last group not full");
160 else if (input->reserved_blocks > input->blocks_count / 5)
161 ext4_warning(sb, "Reserved blocks too high (%u)",
162 input->reserved_blocks);
163 else if (free_blocks_count < 0)
164 ext4_warning(sb, "Bad blocks count %u",
165 input->blocks_count);
166 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
169 ext4_warning(sb, "Cannot read last block (%llu)",
171 } else if (outside(input->block_bitmap, start, end))
172 ext4_warning(sb, "Block bitmap not in group (block %llu)",
173 (unsigned long long)input->block_bitmap);
174 else if (outside(input->inode_bitmap, start, end))
175 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
176 (unsigned long long)input->inode_bitmap);
177 else if (outside(input->inode_table, start, end) ||
178 outside(itend - 1, start, end))
179 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
180 (unsigned long long)input->inode_table, itend - 1);
181 else if (input->inode_bitmap == input->block_bitmap)
182 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
183 (unsigned long long)input->block_bitmap);
184 else if (inside(input->block_bitmap, input->inode_table, itend))
185 ext4_warning(sb, "Block bitmap (%llu) in inode table "
187 (unsigned long long)input->block_bitmap,
188 (unsigned long long)input->inode_table, itend - 1);
189 else if (inside(input->inode_bitmap, input->inode_table, itend))
190 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
192 (unsigned long long)input->inode_bitmap,
193 (unsigned long long)input->inode_table, itend - 1);
194 else if (inside(input->block_bitmap, start, metaend))
195 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
196 (unsigned long long)input->block_bitmap,
198 else if (inside(input->inode_bitmap, start, metaend))
199 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
200 (unsigned long long)input->inode_bitmap,
202 else if (inside(input->inode_table, start, metaend) ||
203 inside(itend - 1, start, metaend))
204 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
206 (unsigned long long)input->inode_table,
207 itend - 1, start, metaend - 1);
216 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
219 struct ext4_new_flex_group_data {
220 struct ext4_new_group_data *groups; /* new_group_data for groups
222 __u16 *bg_flags; /* block group flags of groups
224 ext4_group_t count; /* number of groups in @groups
229 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
232 * Returns NULL on failure otherwise address of the allocated structure.
234 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
236 struct ext4_new_flex_group_data *flex_gd;
238 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
242 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
244 flex_gd->count = flexbg_size;
246 flex_gd->groups = kmalloc_array(flexbg_size,
247 sizeof(struct ext4_new_group_data),
249 if (flex_gd->groups == NULL)
252 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
254 if (flex_gd->bg_flags == NULL)
260 kfree(flex_gd->groups);
267 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
269 kfree(flex_gd->bg_flags);
270 kfree(flex_gd->groups);
275 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
276 * and inode tables for a flex group.
278 * This function is used by 64bit-resize. Note that this function allocates
279 * group tables from the 1st group of groups contained by @flexgd, which may
280 * be a partial of a flex group.
282 * @sb: super block of fs to which the groups belongs
284 * Returns 0 on a successful allocation of the metadata blocks in the
287 static int ext4_alloc_group_tables(struct super_block *sb,
288 struct ext4_new_flex_group_data *flex_gd,
291 struct ext4_new_group_data *group_data = flex_gd->groups;
292 ext4_fsblk_t start_blk;
293 ext4_fsblk_t last_blk;
294 ext4_group_t src_group;
295 ext4_group_t bb_index = 0;
296 ext4_group_t ib_index = 0;
297 ext4_group_t it_index = 0;
299 ext4_group_t last_group;
301 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
304 BUG_ON(flex_gd->count == 0 || group_data == NULL);
306 src_group = group_data[0].group;
307 last_group = src_group + flex_gd->count - 1;
309 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
310 (last_group & ~(flexbg_size - 1))));
312 group = group_data[0].group;
313 if (src_group >= group_data[0].group + flex_gd->count)
315 start_blk = ext4_group_first_block_no(sb, src_group);
316 last_blk = start_blk + group_data[src_group - group].blocks_count;
318 overhead = ext4_group_overhead_blocks(sb, src_group);
320 start_blk += overhead;
322 /* We collect contiguous blocks as much as possible. */
324 for (; src_group <= last_group; src_group++) {
325 overhead = ext4_group_overhead_blocks(sb, src_group);
327 last_blk += group_data[src_group - group].blocks_count;
332 /* Allocate block bitmaps */
333 for (; bb_index < flex_gd->count; bb_index++) {
334 if (start_blk >= last_blk)
336 group_data[bb_index].block_bitmap = start_blk++;
337 group = ext4_get_group_number(sb, start_blk - 1);
338 group -= group_data[0].group;
339 group_data[group].mdata_blocks++;
340 flex_gd->bg_flags[group] &= uninit_mask;
343 /* Allocate inode bitmaps */
344 for (; ib_index < flex_gd->count; ib_index++) {
345 if (start_blk >= last_blk)
347 group_data[ib_index].inode_bitmap = start_blk++;
348 group = ext4_get_group_number(sb, start_blk - 1);
349 group -= group_data[0].group;
350 group_data[group].mdata_blocks++;
351 flex_gd->bg_flags[group] &= uninit_mask;
354 /* Allocate inode tables */
355 for (; it_index < flex_gd->count; it_index++) {
356 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
357 ext4_fsblk_t next_group_start;
359 if (start_blk + itb > last_blk)
361 group_data[it_index].inode_table = start_blk;
362 group = ext4_get_group_number(sb, start_blk);
363 next_group_start = ext4_group_first_block_no(sb, group + 1);
364 group -= group_data[0].group;
366 if (start_blk + itb > next_group_start) {
367 flex_gd->bg_flags[group + 1] &= uninit_mask;
368 overhead = start_blk + itb - next_group_start;
369 group_data[group + 1].mdata_blocks += overhead;
373 group_data[group].mdata_blocks += itb;
374 flex_gd->bg_flags[group] &= uninit_mask;
375 start_blk += EXT4_SB(sb)->s_itb_per_group;
378 /* Update free clusters count to exclude metadata blocks */
379 for (i = 0; i < flex_gd->count; i++) {
380 group_data[i].free_clusters_count -=
381 EXT4_NUM_B2C(EXT4_SB(sb),
382 group_data[i].mdata_blocks);
385 if (test_opt(sb, DEBUG)) {
387 group = group_data[0].group;
389 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
390 "%d groups, flexbg size is %d:\n", flex_gd->count,
393 for (i = 0; i < flex_gd->count; i++) {
395 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
396 ext4_bg_has_super(sb, group + i) ? "normal" :
397 "no-super", group + i,
398 group_data[i].blocks_count,
399 group_data[i].free_clusters_count,
400 group_data[i].mdata_blocks);
406 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
409 struct buffer_head *bh;
412 bh = sb_getblk(sb, blk);
414 return ERR_PTR(-ENOMEM);
415 BUFFER_TRACE(bh, "get_write_access");
416 if ((err = ext4_journal_get_write_access(handle, bh))) {
420 memset(bh->b_data, 0, sb->s_blocksize);
421 set_buffer_uptodate(bh);
427 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
429 return ext4_journal_ensure_credits_fn(handle, credits,
430 EXT4_MAX_TRANS_DATA, 0, 0);
434 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
436 * Helper function for ext4_setup_new_group_blocks() which set .
439 * @handle: journal handle
440 * @flex_gd: flex group data
442 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
443 struct ext4_new_flex_group_data *flex_gd,
444 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
446 struct ext4_sb_info *sbi = EXT4_SB(sb);
447 ext4_group_t count = last_cluster - first_cluster + 1;
450 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
452 for (count2 = count; count > 0;
453 count -= count2, first_cluster += count2) {
455 struct buffer_head *bh;
459 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
460 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
461 group -= flex_gd->groups[0].group;
463 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
467 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
468 BUG_ON(flex_gd->count > 1);
472 err = ext4_resize_ensure_credits_batch(handle, 1);
476 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
480 BUFFER_TRACE(bh, "get_write_access");
481 err = ext4_journal_get_write_access(handle, bh);
486 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
487 first_cluster, first_cluster - start, count2);
488 ext4_set_bits(bh->b_data, first_cluster - start, count2);
490 err = ext4_handle_dirty_metadata(handle, NULL, bh);
500 * Set up the block and inode bitmaps, and the inode table for the new groups.
501 * This doesn't need to be part of the main transaction, since we are only
502 * changing blocks outside the actual filesystem. We still do journaling to
503 * ensure the recovery is correct in case of a failure just after resize.
504 * If any part of this fails, we simply abort the resize.
506 * setup_new_flex_group_blocks handles a flex group as follow:
507 * 1. copy super block and GDT, and initialize group tables if necessary.
508 * In this step, we only set bits in blocks bitmaps for blocks taken by
509 * super block and GDT.
510 * 2. allocate group tables in block bitmaps, that is, set bits in block
511 * bitmap for blocks taken by group tables.
513 static int setup_new_flex_group_blocks(struct super_block *sb,
514 struct ext4_new_flex_group_data *flex_gd)
516 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
519 struct ext4_sb_info *sbi = EXT4_SB(sb);
520 struct ext4_super_block *es = sbi->s_es;
521 struct ext4_new_group_data *group_data = flex_gd->groups;
522 __u16 *bg_flags = flex_gd->bg_flags;
524 ext4_group_t group, count;
525 struct buffer_head *bh = NULL;
526 int reserved_gdb, i, j, err = 0, err2;
529 BUG_ON(!flex_gd->count || !group_data ||
530 group_data[0].group != sbi->s_groups_count);
532 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
533 meta_bg = ext4_has_feature_meta_bg(sb);
535 /* This transaction may be extended/restarted along the way */
536 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
538 return PTR_ERR(handle);
540 group = group_data[0].group;
541 for (i = 0; i < flex_gd->count; i++, group++) {
542 unsigned long gdblocks;
543 ext4_grpblk_t overhead;
545 gdblocks = ext4_bg_num_gdb(sb, group);
546 start = ext4_group_first_block_no(sb, group);
548 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
552 ext4_group_t first_group;
553 first_group = ext4_meta_bg_first_group(sb, group);
554 if (first_group != group + 1 &&
555 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
559 block = start + ext4_bg_has_super(sb, group);
560 /* Copy all of the GDT blocks into the backup in this group */
561 for (j = 0; j < gdblocks; j++, block++) {
562 struct buffer_head *gdb;
564 ext4_debug("update backup group %#04llx\n", block);
565 err = ext4_resize_ensure_credits_batch(handle, 1);
569 gdb = sb_getblk(sb, block);
570 if (unlikely(!gdb)) {
575 BUFFER_TRACE(gdb, "get_write_access");
576 err = ext4_journal_get_write_access(handle, gdb);
581 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
582 s_group_desc, j)->b_data, gdb->b_size);
583 set_buffer_uptodate(gdb);
585 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
593 /* Zero out all of the reserved backup group descriptor
596 if (ext4_bg_has_super(sb, group)) {
597 err = sb_issue_zeroout(sb, gdblocks + start + 1,
598 reserved_gdb, GFP_NOFS);
604 /* Initialize group tables of the grop @group */
605 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
608 /* Zero out all of the inode table blocks */
609 block = group_data[i].inode_table;
610 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
611 block, sbi->s_itb_per_group);
612 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
618 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
621 /* Initialize block bitmap of the @group */
622 block = group_data[i].block_bitmap;
623 err = ext4_resize_ensure_credits_batch(handle, 1);
627 bh = bclean(handle, sb, block);
632 overhead = ext4_group_overhead_blocks(sb, group);
634 ext4_debug("mark backup superblock %#04llx (+0)\n",
636 ext4_set_bits(bh->b_data, 0,
637 EXT4_NUM_B2C(sbi, overhead));
639 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
640 sb->s_blocksize * 8, bh->b_data);
641 err = ext4_handle_dirty_metadata(handle, NULL, bh);
647 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
650 /* Initialize inode bitmap of the @group */
651 block = group_data[i].inode_bitmap;
652 err = ext4_resize_ensure_credits_batch(handle, 1);
655 /* Mark unused entries in inode bitmap used */
656 bh = bclean(handle, sb, block);
662 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
663 sb->s_blocksize * 8, bh->b_data);
664 err = ext4_handle_dirty_metadata(handle, NULL, bh);
670 /* Mark group tables in block bitmap */
671 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
672 count = group_table_count[j];
673 start = (&group_data[0].block_bitmap)[j];
675 for (i = 1; i < flex_gd->count; i++) {
676 block += group_table_count[j];
677 if (block == (&group_data[i].block_bitmap)[j]) {
678 count += group_table_count[j];
681 err = set_flexbg_block_bitmap(sb, handle,
683 EXT4_B2C(sbi, start),
689 count = group_table_count[j];
690 start = (&group_data[i].block_bitmap)[j];
695 err = set_flexbg_block_bitmap(sb, handle,
697 EXT4_B2C(sbi, start),
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 = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
869 ext4_warning(sb, "not enough memory for %lu groups",
875 * Finally, we have all of the possible failures behind us...
877 * Remove new GDT block from inode double-indirect block and clear out
878 * the new GDT block for use (which also "frees" the backup GDT blocks
879 * from the reserved inode). We don't need to change the bitmaps for
880 * these blocks, because they are marked as in-use from being in the
881 * reserved inode, and will become GDT blocks (primary and backup).
883 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
884 err = ext4_handle_dirty_metadata(handle, NULL, dind);
886 ext4_std_error(sb, err);
889 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
890 (9 - EXT4_SB(sb)->s_cluster_bits);
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 lock_buffer(EXT4_SB(sb)->s_sbh);
912 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
913 ext4_superblock_csum_set(sb);
914 unlock_buffer(EXT4_SB(sb)->s_sbh);
915 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
917 ext4_std_error(sb, err);
920 kvfree(n_group_desc);
925 ext4_debug("leaving with error %d\n", err);
930 * add_new_gdb_meta_bg is the sister of add_new_gdb.
932 static int add_new_gdb_meta_bg(struct super_block *sb,
933 handle_t *handle, ext4_group_t group) {
934 ext4_fsblk_t gdblock;
935 struct buffer_head *gdb_bh;
936 struct buffer_head **o_group_desc, **n_group_desc;
937 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
940 gdblock = ext4_meta_bg_first_block_no(sb, group) +
941 ext4_bg_has_super(sb, group);
942 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
944 return PTR_ERR(gdb_bh);
945 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
950 ext4_warning(sb, "not enough memory for %lu groups",
956 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
957 memcpy(n_group_desc, o_group_desc,
958 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
960 n_group_desc[gdb_num] = gdb_bh;
962 BUFFER_TRACE(gdb_bh, "get_write_access");
963 err = ext4_journal_get_write_access(handle, gdb_bh);
965 kvfree(n_group_desc);
970 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
971 EXT4_SB(sb)->s_gdb_count++;
972 ext4_kvfree_array_rcu(o_group_desc);
977 * Called when we are adding a new group which has a backup copy of each of
978 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
979 * We need to add these reserved backup GDT blocks to the resize inode, so
980 * that they are kept for future resizing and not allocated to files.
982 * Each reserved backup GDT block will go into a different indirect block.
983 * The indirect blocks are actually the primary reserved GDT blocks,
984 * so we know in advance what their block numbers are. We only get the
985 * double-indirect block to verify it is pointing to the primary reserved
986 * GDT blocks so we don't overwrite a data block by accident. The reserved
987 * backup GDT blocks are stored in their reserved primary GDT block.
989 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
992 struct super_block *sb = inode->i_sb;
993 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
994 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
995 struct buffer_head **primary;
996 struct buffer_head *dind;
997 struct ext4_iloc iloc;
1004 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1008 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1009 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1011 err = PTR_ERR(dind);
1016 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1017 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1018 EXT4_ADDR_PER_BLOCK(sb));
1019 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1021 /* Get each reserved primary GDT block and verify it holds backups */
1022 for (res = 0; res < reserved_gdb; res++, blk++) {
1023 if (le32_to_cpu(*data) != blk) {
1024 ext4_warning(sb, "reserved block %llu"
1025 " not at offset %ld",
1027 (long)(data - (__le32 *)dind->b_data));
1031 primary[res] = ext4_sb_bread(sb, blk, 0);
1032 if (IS_ERR(primary[res])) {
1033 err = PTR_ERR(primary[res]);
1034 primary[res] = NULL;
1037 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1038 if (gdbackups < 0) {
1039 brelse(primary[res]);
1044 data = (__le32 *)dind->b_data;
1047 for (i = 0; i < reserved_gdb; i++) {
1048 BUFFER_TRACE(primary[i], "get_write_access");
1049 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1053 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1057 * Finally we can add each of the reserved backup GDT blocks from
1058 * the new group to its reserved primary GDT block.
1060 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1061 for (i = 0; i < reserved_gdb; i++) {
1063 data = (__le32 *)primary[i]->b_data;
1064 /* printk("reserving backup %lu[%u] = %lu\n",
1065 primary[i]->b_blocknr, gdbackups,
1066 blk + primary[i]->b_blocknr); */
1067 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1068 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1073 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1074 ext4_mark_iloc_dirty(handle, inode, &iloc);
1078 brelse(primary[res]);
1088 * Update the backup copies of the ext4 metadata. These don't need to be part
1089 * of the main resize transaction, because e2fsck will re-write them if there
1090 * is a problem (basically only OOM will cause a problem). However, we
1091 * _should_ update the backups if possible, in case the primary gets trashed
1092 * for some reason and we need to run e2fsck from a backup superblock. The
1093 * important part is that the new block and inode counts are in the backup
1094 * superblocks, and the location of the new group metadata in the GDT backups.
1096 * We do not need take the s_resize_lock for this, because these
1097 * blocks are not otherwise touched by the filesystem code when it is
1098 * mounted. We don't need to worry about last changing from
1099 * sbi->s_groups_count, because the worst that can happen is that we
1100 * do not copy the full number of backups at this time. The resize
1101 * which changed s_groups_count will backup again.
1103 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1104 int size, int meta_bg)
1106 struct ext4_sb_info *sbi = EXT4_SB(sb);
1108 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1112 ext4_group_t group = 0;
1113 int rest = sb->s_blocksize - size;
1117 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1118 if (IS_ERR(handle)) {
1120 err = PTR_ERR(handle);
1125 group = ext4_list_backups(sb, &three, &five, &seven);
1126 last = sbi->s_groups_count;
1128 group = ext4_get_group_number(sb, blk_off) + 1;
1129 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1132 while (group < sbi->s_groups_count) {
1133 struct buffer_head *bh;
1134 ext4_fsblk_t backup_block;
1136 /* Out of journal space, and can't get more - abort - so sad */
1137 err = ext4_resize_ensure_credits_batch(handle, 1);
1142 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1144 backup_block = (ext4_group_first_block_no(sb, group) +
1145 ext4_bg_has_super(sb, group));
1147 bh = sb_getblk(sb, backup_block);
1148 if (unlikely(!bh)) {
1152 ext4_debug("update metadata backup %llu(+%llu)\n",
1153 backup_block, backup_block -
1154 ext4_group_first_block_no(sb, group));
1155 BUFFER_TRACE(bh, "get_write_access");
1156 if ((err = ext4_journal_get_write_access(handle, bh))) {
1161 memcpy(bh->b_data, data, size);
1163 memset(bh->b_data + size, 0, rest);
1164 set_buffer_uptodate(bh);
1166 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1168 ext4_std_error(sb, err);
1172 group = ext4_list_backups(sb, &three, &five, &seven);
1173 else if (group == last)
1178 if ((err2 = ext4_journal_stop(handle)) && !err)
1182 * Ugh! Need to have e2fsck write the backup copies. It is too
1183 * late to revert the resize, we shouldn't fail just because of
1184 * the backup copies (they are only needed in case of corruption).
1186 * However, if we got here we have a journal problem too, so we
1187 * can't really start a transaction to mark the superblock.
1188 * Chicken out and just set the flag on the hope it will be written
1189 * to disk, and if not - we will simply wait until next fsck.
1193 ext4_warning(sb, "can't update backup for group %u (err %d), "
1194 "forcing fsck on next reboot", group, err);
1195 sbi->s_mount_state &= ~EXT4_VALID_FS;
1196 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1197 mark_buffer_dirty(sbi->s_sbh);
1202 * ext4_add_new_descs() adds @count group descriptor of groups
1203 * starting at @group
1205 * @handle: journal handle
1207 * @group: the group no. of the first group desc to be added
1208 * @resize_inode: the resize inode
1209 * @count: number of group descriptors to be added
1211 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1212 ext4_group_t group, struct inode *resize_inode,
1215 struct ext4_sb_info *sbi = EXT4_SB(sb);
1216 struct ext4_super_block *es = sbi->s_es;
1217 struct buffer_head *gdb_bh;
1218 int i, gdb_off, gdb_num, err = 0;
1221 meta_bg = ext4_has_feature_meta_bg(sb);
1222 for (i = 0; i < count; i++, group++) {
1223 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1224 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1226 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1227 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1230 * We will only either add reserved group blocks to a backup group
1231 * or remove reserved blocks for the first group in a new group block.
1232 * Doing both would be mean more complex code, and sane people don't
1233 * use non-sparse filesystems anymore. This is already checked above.
1236 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1238 BUFFER_TRACE(gdb_bh, "get_write_access");
1239 err = ext4_journal_get_write_access(handle, gdb_bh);
1241 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1242 err = reserve_backup_gdb(handle, resize_inode, group);
1243 } else if (meta_bg != 0) {
1244 err = add_new_gdb_meta_bg(sb, handle, group);
1246 err = add_new_gdb(handle, resize_inode, group);
1254 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1256 struct buffer_head *bh = sb_getblk(sb, block);
1259 if (!bh_uptodate_or_lock(bh)) {
1260 if (ext4_read_bh(bh, 0, NULL) < 0) {
1269 static int ext4_set_bitmap_checksums(struct super_block *sb,
1271 struct ext4_group_desc *gdp,
1272 struct ext4_new_group_data *group_data)
1274 struct buffer_head *bh;
1276 if (!ext4_has_metadata_csum(sb))
1279 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1282 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1283 EXT4_INODES_PER_GROUP(sb) / 8);
1286 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1289 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1296 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1298 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1299 struct ext4_new_flex_group_data *flex_gd)
1301 struct ext4_new_group_data *group_data = flex_gd->groups;
1302 struct ext4_group_desc *gdp;
1303 struct ext4_sb_info *sbi = EXT4_SB(sb);
1304 struct buffer_head *gdb_bh;
1306 __u16 *bg_flags = flex_gd->bg_flags;
1307 int i, gdb_off, gdb_num, err = 0;
1310 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1311 group = group_data->group;
1313 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1314 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1317 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1319 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1320 /* Update group descriptor block for new group */
1321 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1322 gdb_off * EXT4_DESC_SIZE(sb));
1324 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1325 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1326 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1327 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1329 ext4_std_error(sb, err);
1333 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1334 ext4_free_group_clusters_set(sb, gdp,
1335 group_data->free_clusters_count);
1336 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1337 if (ext4_has_group_desc_csum(sb))
1338 ext4_itable_unused_set(sb, gdp,
1339 EXT4_INODES_PER_GROUP(sb));
1340 gdp->bg_flags = cpu_to_le16(*bg_flags);
1341 ext4_group_desc_csum_set(sb, group, gdp);
1343 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1344 if (unlikely(err)) {
1345 ext4_std_error(sb, err);
1350 * We can allocate memory for mb_alloc based on the new group
1353 err = ext4_mb_add_groupinfo(sb, group, gdp);
1361 * ext4_update_super() updates the super block so that the newly added
1362 * groups can be seen by the filesystem.
1365 * @flex_gd: new added groups
1367 static void ext4_update_super(struct super_block *sb,
1368 struct ext4_new_flex_group_data *flex_gd)
1370 ext4_fsblk_t blocks_count = 0;
1371 ext4_fsblk_t free_blocks = 0;
1372 ext4_fsblk_t reserved_blocks = 0;
1373 struct ext4_new_group_data *group_data = flex_gd->groups;
1374 struct ext4_sb_info *sbi = EXT4_SB(sb);
1375 struct ext4_super_block *es = sbi->s_es;
1378 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1380 * Make the new blocks and inodes valid next. We do this before
1381 * increasing the group count so that once the group is enabled,
1382 * all of its blocks and inodes are already valid.
1384 * We always allocate group-by-group, then block-by-block or
1385 * inode-by-inode within a group, so enabling these
1386 * blocks/inodes before the group is live won't actually let us
1387 * allocate the new space yet.
1389 for (i = 0; i < flex_gd->count; i++) {
1390 blocks_count += group_data[i].blocks_count;
1391 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1394 reserved_blocks = ext4_r_blocks_count(es) * 100;
1395 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1396 reserved_blocks *= blocks_count;
1397 do_div(reserved_blocks, 100);
1399 lock_buffer(sbi->s_sbh);
1400 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1401 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1402 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1404 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1407 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1409 * We need to protect s_groups_count against other CPUs seeing
1410 * inconsistent state in the superblock.
1412 * The precise rules we use are:
1414 * * Writers must perform a smp_wmb() after updating all
1415 * dependent data and before modifying the groups count
1417 * * Readers must perform an smp_rmb() after reading the groups
1418 * count and before reading any dependent data.
1420 * NB. These rules can be relaxed when checking the group count
1421 * while freeing data, as we can only allocate from a block
1422 * group after serialising against the group count, and we can
1423 * only then free after serialising in turn against that
1428 /* Update the global fs size fields */
1429 sbi->s_groups_count += flex_gd->count;
1430 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1431 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1433 /* Update the reserved block counts only once the new group is
1435 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1437 ext4_superblock_csum_set(sb);
1438 unlock_buffer(sbi->s_sbh);
1440 /* Update the free space counts */
1441 percpu_counter_add(&sbi->s_freeclusters_counter,
1442 EXT4_NUM_B2C(sbi, free_blocks));
1443 percpu_counter_add(&sbi->s_freeinodes_counter,
1444 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1446 ext4_debug("free blocks count %llu",
1447 percpu_counter_read(&sbi->s_freeclusters_counter));
1448 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1449 ext4_group_t flex_group;
1450 struct flex_groups *fg;
1452 flex_group = ext4_flex_group(sbi, group_data[0].group);
1453 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1454 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1455 &fg->free_clusters);
1456 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1461 * Update the fs overhead information
1463 ext4_calculate_overhead(sb);
1465 if (test_opt(sb, DEBUG))
1466 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1467 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1468 blocks_count, free_blocks, reserved_blocks);
1471 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1472 * _before_ we start modifying the filesystem, because we cannot abort the
1473 * transaction and not have it write the data to disk.
1475 static int ext4_flex_group_add(struct super_block *sb,
1476 struct inode *resize_inode,
1477 struct ext4_new_flex_group_data *flex_gd)
1479 struct ext4_sb_info *sbi = EXT4_SB(sb);
1480 struct ext4_super_block *es = sbi->s_es;
1481 ext4_fsblk_t o_blocks_count;
1485 unsigned reserved_gdb;
1486 int err = 0, err2 = 0, credit;
1488 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1490 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1491 o_blocks_count = ext4_blocks_count(es);
1492 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1495 err = setup_new_flex_group_blocks(sb, flex_gd);
1499 * We will always be modifying at least the superblock and GDT
1500 * blocks. If we are adding a group past the last current GDT block,
1501 * we will also modify the inode and the dindirect block. If we
1502 * are adding a group with superblock/GDT backups we will also
1503 * modify each of the reserved GDT dindirect blocks.
1505 credit = 3; /* sb, resize inode, resize inode dindirect */
1507 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1508 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1509 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1510 if (IS_ERR(handle)) {
1511 err = PTR_ERR(handle);
1515 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1516 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1520 group = flex_gd->groups[0].group;
1521 BUG_ON(group != sbi->s_groups_count);
1522 err = ext4_add_new_descs(handle, sb, group,
1523 resize_inode, flex_gd->count);
1527 err = ext4_setup_new_descs(handle, sb, flex_gd);
1531 ext4_update_super(sb, flex_gd);
1533 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1536 err2 = ext4_journal_stop(handle);
1541 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1542 int gdb_num_end = ((group + flex_gd->count - 1) /
1543 EXT4_DESC_PER_BLOCK(sb));
1544 int meta_bg = ext4_has_feature_meta_bg(sb);
1545 sector_t old_gdb = 0;
1547 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1548 sizeof(struct ext4_super_block), 0);
1549 for (; gdb_num <= gdb_num_end; gdb_num++) {
1550 struct buffer_head *gdb_bh;
1552 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1554 if (old_gdb == gdb_bh->b_blocknr)
1556 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1557 gdb_bh->b_size, meta_bg);
1558 old_gdb = gdb_bh->b_blocknr;
1565 static int ext4_setup_next_flex_gd(struct super_block *sb,
1566 struct ext4_new_flex_group_data *flex_gd,
1567 ext4_fsblk_t n_blocks_count,
1568 unsigned long flexbg_size)
1570 struct ext4_sb_info *sbi = EXT4_SB(sb);
1571 struct ext4_super_block *es = sbi->s_es;
1572 struct ext4_new_group_data *group_data = flex_gd->groups;
1573 ext4_fsblk_t o_blocks_count;
1574 ext4_group_t n_group;
1576 ext4_group_t last_group;
1578 ext4_grpblk_t clusters_per_group;
1581 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1583 o_blocks_count = ext4_blocks_count(es);
1585 if (o_blocks_count == n_blocks_count)
1588 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1590 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1592 last_group = group | (flexbg_size - 1);
1593 if (last_group > n_group)
1594 last_group = n_group;
1596 flex_gd->count = last_group - group + 1;
1598 for (i = 0; i < flex_gd->count; i++) {
1601 group_data[i].group = group + i;
1602 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1603 overhead = ext4_group_overhead_blocks(sb, group + i);
1604 group_data[i].mdata_blocks = overhead;
1605 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1606 if (ext4_has_group_desc_csum(sb)) {
1607 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1608 EXT4_BG_INODE_UNINIT;
1609 if (!test_opt(sb, INIT_INODE_TABLE))
1610 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1612 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1615 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1616 /* We need to initialize block bitmap of last group. */
1617 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1619 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1620 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1621 group_data[i - 1].free_clusters_count -= clusters_per_group -
1628 /* Add group descriptor data to an existing or new group descriptor block.
1629 * Ensure we handle all possible error conditions _before_ we start modifying
1630 * the filesystem, because we cannot abort the transaction and not have it
1631 * write the data to disk.
1633 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1634 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1636 * We only need to hold the superblock lock while we are actually adding
1637 * in the new group's counts to the superblock. Prior to that we have
1638 * not really "added" the group at all. We re-check that we are still
1639 * adding in the last group in case things have changed since verifying.
1641 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1643 struct ext4_new_flex_group_data flex_gd;
1644 struct ext4_sb_info *sbi = EXT4_SB(sb);
1645 struct ext4_super_block *es = sbi->s_es;
1646 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1647 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1648 struct inode *inode = NULL;
1653 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1655 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1656 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1660 if (ext4_blocks_count(es) + input->blocks_count <
1661 ext4_blocks_count(es)) {
1662 ext4_warning(sb, "blocks_count overflow");
1666 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1667 le32_to_cpu(es->s_inodes_count)) {
1668 ext4_warning(sb, "inodes_count overflow");
1672 if (reserved_gdb || gdb_off == 0) {
1673 if (!ext4_has_feature_resize_inode(sb) ||
1674 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1676 "No reserved GDT blocks, can't resize");
1679 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1680 if (IS_ERR(inode)) {
1681 ext4_warning(sb, "Error opening resize inode");
1682 return PTR_ERR(inode);
1687 err = verify_group_input(sb, input);
1691 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1695 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1700 flex_gd.groups = input;
1701 flex_gd.bg_flags = &bg_flags;
1702 err = ext4_flex_group_add(sb, inode, &flex_gd);
1706 } /* ext4_group_add */
1709 * extend a group without checking assuming that checking has been done.
1711 static int ext4_group_extend_no_check(struct super_block *sb,
1712 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1714 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1718 /* We will update the superblock, one block bitmap, and
1719 * one group descriptor via ext4_group_add_blocks().
1721 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1722 if (IS_ERR(handle)) {
1723 err = PTR_ERR(handle);
1724 ext4_warning(sb, "error %d on journal start", err);
1728 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1729 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1731 ext4_warning(sb, "error %d on journal write access", err);
1735 lock_buffer(EXT4_SB(sb)->s_sbh);
1736 ext4_blocks_count_set(es, o_blocks_count + add);
1737 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1738 ext4_superblock_csum_set(sb);
1739 unlock_buffer(EXT4_SB(sb)->s_sbh);
1740 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1741 o_blocks_count + add);
1742 /* We add the blocks to the bitmap and set the group need init bit */
1743 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1746 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1747 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1748 o_blocks_count + add);
1750 err2 = ext4_journal_stop(handle);
1755 if (test_opt(sb, DEBUG))
1756 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1757 "blocks\n", ext4_blocks_count(es));
1758 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1759 (char *)es, sizeof(struct ext4_super_block), 0);
1765 * Extend the filesystem to the new number of blocks specified. This entry
1766 * point is only used to extend the current filesystem to the end of the last
1767 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1768 * for emergencies (because it has no dependencies on reserved blocks).
1770 * If we _really_ wanted, we could use default values to call ext4_group_add()
1771 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1772 * GDT blocks are reserved to grow to the desired size.
1774 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1775 ext4_fsblk_t n_blocks_count)
1777 ext4_fsblk_t o_blocks_count;
1780 struct buffer_head *bh;
1784 o_blocks_count = ext4_blocks_count(es);
1786 if (test_opt(sb, DEBUG))
1787 ext4_msg(sb, KERN_DEBUG,
1788 "extending last group from %llu to %llu blocks",
1789 o_blocks_count, n_blocks_count);
1791 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1794 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1795 ext4_msg(sb, KERN_ERR,
1796 "filesystem too large to resize to %llu blocks safely",
1801 if (n_blocks_count < o_blocks_count) {
1802 ext4_warning(sb, "can't shrink FS - resize aborted");
1806 /* Handle the remaining blocks in the last group only. */
1807 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1810 ext4_warning(sb, "need to use ext2online to resize further");
1814 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1816 if (o_blocks_count + add < o_blocks_count) {
1817 ext4_warning(sb, "blocks_count overflow");
1821 if (o_blocks_count + add > n_blocks_count)
1822 add = n_blocks_count - o_blocks_count;
1824 if (o_blocks_count + add < n_blocks_count)
1825 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1826 o_blocks_count + add, add);
1828 /* See if the device is actually as big as what was requested */
1829 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1831 ext4_warning(sb, "can't read last block, resize aborted");
1836 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1838 } /* ext4_group_extend */
1841 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1843 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1847 * Release the resize inode and drop the resize_inode feature if there
1848 * are no more reserved gdt blocks, and then convert the file system
1851 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1854 struct ext4_sb_info *sbi = EXT4_SB(sb);
1855 struct ext4_super_block *es = sbi->s_es;
1856 struct ext4_inode_info *ei = EXT4_I(inode);
1858 int i, ret, err = 0;
1861 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1863 if (es->s_reserved_gdt_blocks) {
1864 ext4_error(sb, "Unexpected non-zero "
1865 "s_reserved_gdt_blocks");
1869 /* Do a quick sanity check of the resize inode */
1870 if (inode->i_blocks != 1 << (inode->i_blkbits -
1871 (9 - sbi->s_cluster_bits)))
1872 goto invalid_resize_inode;
1873 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1874 if (i == EXT4_DIND_BLOCK) {
1878 goto invalid_resize_inode;
1881 goto invalid_resize_inode;
1883 credits += 3; /* block bitmap, bg descriptor, resize inode */
1886 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1888 return PTR_ERR(handle);
1890 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1891 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1895 lock_buffer(sbi->s_sbh);
1896 ext4_clear_feature_resize_inode(sb);
1897 ext4_set_feature_meta_bg(sb);
1898 sbi->s_es->s_first_meta_bg =
1899 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1900 ext4_superblock_csum_set(sb);
1901 unlock_buffer(sbi->s_sbh);
1903 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1905 ext4_std_error(sb, err);
1910 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1911 ext4_free_blocks(handle, inode, NULL, nr, 1,
1912 EXT4_FREE_BLOCKS_METADATA |
1913 EXT4_FREE_BLOCKS_FORGET);
1914 ei->i_data[EXT4_DIND_BLOCK] = 0;
1915 inode->i_blocks = 0;
1917 err = ext4_mark_inode_dirty(handle, inode);
1919 ext4_std_error(sb, err);
1923 ret = ext4_journal_stop(handle);
1928 invalid_resize_inode:
1929 ext4_error(sb, "corrupted/inconsistent resize inode");
1934 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1936 * @sb: super block of the fs to be resized
1937 * @n_blocks_count: the number of blocks resides in the resized fs
1939 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1941 struct ext4_new_flex_group_data *flex_gd = NULL;
1942 struct ext4_sb_info *sbi = EXT4_SB(sb);
1943 struct ext4_super_block *es = sbi->s_es;
1944 struct buffer_head *bh;
1945 struct inode *resize_inode = NULL;
1946 ext4_grpblk_t add, offset;
1947 unsigned long n_desc_blocks;
1948 unsigned long o_desc_blocks;
1949 ext4_group_t o_group;
1950 ext4_group_t n_group;
1951 ext4_fsblk_t o_blocks_count;
1952 ext4_fsblk_t n_blocks_count_retry = 0;
1953 unsigned long last_update_time = 0;
1954 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1957 /* See if the device is actually as big as what was requested */
1958 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1960 ext4_warning(sb, "can't read last block, resize aborted");
1966 o_blocks_count = ext4_blocks_count(es);
1968 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1969 "to %llu blocks", o_blocks_count, n_blocks_count);
1971 if (n_blocks_count < o_blocks_count) {
1972 /* On-line shrinking not supported */
1973 ext4_warning(sb, "can't shrink FS - resize aborted");
1977 if (n_blocks_count == o_blocks_count)
1978 /* Nothing need to do */
1981 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1982 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1983 ext4_warning(sb, "resize would cause inodes_count overflow");
1986 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1988 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1989 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1991 meta_bg = ext4_has_feature_meta_bg(sb);
1993 if (ext4_has_feature_resize_inode(sb)) {
1995 ext4_error(sb, "resize_inode and meta_bg enabled "
1999 if (n_desc_blocks > o_desc_blocks +
2000 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2001 n_blocks_count_retry = n_blocks_count;
2002 n_desc_blocks = o_desc_blocks +
2003 le16_to_cpu(es->s_reserved_gdt_blocks);
2004 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2005 n_blocks_count = (ext4_fsblk_t)n_group *
2006 EXT4_BLOCKS_PER_GROUP(sb) +
2007 le32_to_cpu(es->s_first_data_block);
2008 n_group--; /* set to last group number */
2012 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2014 if (IS_ERR(resize_inode)) {
2015 ext4_warning(sb, "Error opening resize inode");
2016 return PTR_ERR(resize_inode);
2020 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2021 err = ext4_convert_meta_bg(sb, resize_inode);
2026 resize_inode = NULL;
2028 if (n_blocks_count_retry) {
2029 n_blocks_count = n_blocks_count_retry;
2030 n_blocks_count_retry = 0;
2036 * Make sure the last group has enough space so that it's
2037 * guaranteed to have enough space for all metadata blocks
2038 * that it might need to hold. (We might not need to store
2039 * the inode table blocks in the last block group, but there
2040 * will be cases where this might be needed.)
2042 if ((ext4_group_first_block_no(sb, n_group) +
2043 ext4_group_overhead_blocks(sb, n_group) + 2 +
2044 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2045 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2047 n_blocks_count_retry = 0;
2050 resize_inode = NULL;
2055 /* extend the last group */
2056 if (n_group == o_group)
2057 add = n_blocks_count - o_blocks_count;
2059 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2061 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2066 if (ext4_blocks_count(es) == n_blocks_count)
2069 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2073 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2077 flex_gd = alloc_flex_gd(flexbg_size);
2078 if (flex_gd == NULL) {
2083 /* Add flex groups. Note that a regular group is a
2084 * flex group with 1 group.
2086 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2088 if (jiffies - last_update_time > HZ * 10) {
2089 if (last_update_time)
2090 ext4_msg(sb, KERN_INFO,
2091 "resized to %llu blocks",
2092 ext4_blocks_count(es));
2093 last_update_time = jiffies;
2095 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2097 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2102 if (!err && n_blocks_count_retry) {
2103 n_blocks_count = n_blocks_count_retry;
2104 n_blocks_count_retry = 0;
2105 free_flex_gd(flex_gd);
2109 resize_inode = NULL;
2116 free_flex_gd(flex_gd);
2117 if (resize_inode != NULL)
2120 ext4_warning(sb, "error (%d) occurred during "
2121 "file system resize", err);
2122 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2123 ext4_blocks_count(es));