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"
20 int ext4_resize_begin(struct super_block *sb)
22 struct ext4_sb_info *sbi = EXT4_SB(sb);
25 if (!capable(CAP_SYS_RESOURCE))
29 * If we are not using the primary superblock/GDT copy don't resize,
30 * because the user tools have no way of handling this. Probably a
31 * bad time to do it anyways.
33 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
34 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
35 ext4_warning(sb, "won't resize using backup superblock at %llu",
36 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
41 * We are not allowed to do online-resizing on a filesystem mounted
42 * with error, because it can destroy the filesystem easily.
44 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
45 ext4_warning(sb, "There are errors in the filesystem, "
46 "so online resizing is not allowed");
50 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
51 &EXT4_SB(sb)->s_ext4_flags))
57 void ext4_resize_end(struct super_block *sb)
59 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
60 smp_mb__after_atomic();
63 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
65 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
66 EXT4_DESC_PER_BLOCK_BITS(sb);
69 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
71 group = ext4_meta_bg_first_group(sb, group);
72 return ext4_group_first_block_no(sb, group);
75 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
77 ext4_grpblk_t overhead;
78 overhead = ext4_bg_num_gdb(sb, group);
79 if (ext4_bg_has_super(sb, group))
81 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
85 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
86 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
88 static int verify_group_input(struct super_block *sb,
89 struct ext4_new_group_data *input)
91 struct ext4_sb_info *sbi = EXT4_SB(sb);
92 struct ext4_super_block *es = sbi->s_es;
93 ext4_fsblk_t start = ext4_blocks_count(es);
94 ext4_fsblk_t end = start + input->blocks_count;
95 ext4_group_t group = input->group;
96 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
99 struct buffer_head *bh = NULL;
100 ext4_grpblk_t free_blocks_count, offset;
103 if (group != sbi->s_groups_count) {
104 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
105 input->group, sbi->s_groups_count);
109 overhead = ext4_group_overhead_blocks(sb, group);
110 metaend = start + overhead;
111 input->free_clusters_count = free_blocks_count =
112 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
114 if (test_opt(sb, DEBUG))
115 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
116 "(%d free, %u reserved)\n",
117 ext4_bg_has_super(sb, input->group) ? "normal" :
118 "no-super", input->group, input->blocks_count,
119 free_blocks_count, input->reserved_blocks);
121 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
123 ext4_warning(sb, "Last group not full");
124 else if (input->reserved_blocks > input->blocks_count / 5)
125 ext4_warning(sb, "Reserved blocks too high (%u)",
126 input->reserved_blocks);
127 else if (free_blocks_count < 0)
128 ext4_warning(sb, "Bad blocks count %u",
129 input->blocks_count);
130 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
133 ext4_warning(sb, "Cannot read last block (%llu)",
135 } else if (outside(input->block_bitmap, start, end))
136 ext4_warning(sb, "Block bitmap not in group (block %llu)",
137 (unsigned long long)input->block_bitmap);
138 else if (outside(input->inode_bitmap, start, end))
139 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
140 (unsigned long long)input->inode_bitmap);
141 else if (outside(input->inode_table, start, end) ||
142 outside(itend - 1, start, end))
143 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
144 (unsigned long long)input->inode_table, itend - 1);
145 else if (input->inode_bitmap == input->block_bitmap)
146 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
147 (unsigned long long)input->block_bitmap);
148 else if (inside(input->block_bitmap, input->inode_table, itend))
149 ext4_warning(sb, "Block bitmap (%llu) in inode table "
151 (unsigned long long)input->block_bitmap,
152 (unsigned long long)input->inode_table, itend - 1);
153 else if (inside(input->inode_bitmap, input->inode_table, itend))
154 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
156 (unsigned long long)input->inode_bitmap,
157 (unsigned long long)input->inode_table, itend - 1);
158 else if (inside(input->block_bitmap, start, metaend))
159 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
160 (unsigned long long)input->block_bitmap,
162 else if (inside(input->inode_bitmap, start, metaend))
163 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
164 (unsigned long long)input->inode_bitmap,
166 else if (inside(input->inode_table, start, metaend) ||
167 inside(itend - 1, start, metaend))
168 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
170 (unsigned long long)input->inode_table,
171 itend - 1, start, metaend - 1);
180 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
183 struct ext4_new_flex_group_data {
184 struct ext4_new_group_data *groups; /* new_group_data for groups
186 __u16 *bg_flags; /* block group flags of groups
188 ext4_group_t count; /* number of groups in @groups
193 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
196 * Returns NULL on failure otherwise address of the allocated structure.
198 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
200 struct ext4_new_flex_group_data *flex_gd;
202 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
206 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
208 flex_gd->count = flexbg_size;
210 flex_gd->groups = kmalloc_array(flexbg_size,
211 sizeof(struct ext4_new_group_data),
213 if (flex_gd->groups == NULL)
216 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
218 if (flex_gd->bg_flags == NULL)
224 kfree(flex_gd->groups);
231 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
233 kfree(flex_gd->bg_flags);
234 kfree(flex_gd->groups);
239 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
240 * and inode tables for a flex group.
242 * This function is used by 64bit-resize. Note that this function allocates
243 * group tables from the 1st group of groups contained by @flexgd, which may
244 * be a partial of a flex group.
246 * @sb: super block of fs to which the groups belongs
248 * Returns 0 on a successful allocation of the metadata blocks in the
251 static int ext4_alloc_group_tables(struct super_block *sb,
252 struct ext4_new_flex_group_data *flex_gd,
255 struct ext4_new_group_data *group_data = flex_gd->groups;
256 ext4_fsblk_t start_blk;
257 ext4_fsblk_t last_blk;
258 ext4_group_t src_group;
259 ext4_group_t bb_index = 0;
260 ext4_group_t ib_index = 0;
261 ext4_group_t it_index = 0;
263 ext4_group_t last_group;
265 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
268 BUG_ON(flex_gd->count == 0 || group_data == NULL);
270 src_group = group_data[0].group;
271 last_group = src_group + flex_gd->count - 1;
273 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
274 (last_group & ~(flexbg_size - 1))));
276 group = group_data[0].group;
277 if (src_group >= group_data[0].group + flex_gd->count)
279 start_blk = ext4_group_first_block_no(sb, src_group);
280 last_blk = start_blk + group_data[src_group - group].blocks_count;
282 overhead = ext4_group_overhead_blocks(sb, src_group);
284 start_blk += overhead;
286 /* We collect contiguous blocks as much as possible. */
288 for (; src_group <= last_group; src_group++) {
289 overhead = ext4_group_overhead_blocks(sb, src_group);
291 last_blk += group_data[src_group - group].blocks_count;
296 /* Allocate block bitmaps */
297 for (; bb_index < flex_gd->count; bb_index++) {
298 if (start_blk >= last_blk)
300 group_data[bb_index].block_bitmap = start_blk++;
301 group = ext4_get_group_number(sb, start_blk - 1);
302 group -= group_data[0].group;
303 group_data[group].mdata_blocks++;
304 flex_gd->bg_flags[group] &= uninit_mask;
307 /* Allocate inode bitmaps */
308 for (; ib_index < flex_gd->count; ib_index++) {
309 if (start_blk >= last_blk)
311 group_data[ib_index].inode_bitmap = start_blk++;
312 group = ext4_get_group_number(sb, start_blk - 1);
313 group -= group_data[0].group;
314 group_data[group].mdata_blocks++;
315 flex_gd->bg_flags[group] &= uninit_mask;
318 /* Allocate inode tables */
319 for (; it_index < flex_gd->count; it_index++) {
320 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
321 ext4_fsblk_t next_group_start;
323 if (start_blk + itb > last_blk)
325 group_data[it_index].inode_table = start_blk;
326 group = ext4_get_group_number(sb, start_blk);
327 next_group_start = ext4_group_first_block_no(sb, group + 1);
328 group -= group_data[0].group;
330 if (start_blk + itb > next_group_start) {
331 flex_gd->bg_flags[group + 1] &= uninit_mask;
332 overhead = start_blk + itb - next_group_start;
333 group_data[group + 1].mdata_blocks += overhead;
337 group_data[group].mdata_blocks += itb;
338 flex_gd->bg_flags[group] &= uninit_mask;
339 start_blk += EXT4_SB(sb)->s_itb_per_group;
342 /* Update free clusters count to exclude metadata blocks */
343 for (i = 0; i < flex_gd->count; i++) {
344 group_data[i].free_clusters_count -=
345 EXT4_NUM_B2C(EXT4_SB(sb),
346 group_data[i].mdata_blocks);
349 if (test_opt(sb, DEBUG)) {
351 group = group_data[0].group;
353 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
354 "%d groups, flexbg size is %d:\n", flex_gd->count,
357 for (i = 0; i < flex_gd->count; i++) {
359 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
360 ext4_bg_has_super(sb, group + i) ? "normal" :
361 "no-super", group + i,
362 group_data[i].blocks_count,
363 group_data[i].free_clusters_count,
364 group_data[i].mdata_blocks);
370 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
373 struct buffer_head *bh;
376 bh = sb_getblk(sb, blk);
378 return ERR_PTR(-ENOMEM);
379 BUFFER_TRACE(bh, "get_write_access");
380 if ((err = ext4_journal_get_write_access(handle, bh))) {
384 memset(bh->b_data, 0, sb->s_blocksize);
385 set_buffer_uptodate(bh);
391 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
393 return ext4_journal_ensure_credits_fn(handle, credits,
394 EXT4_MAX_TRANS_DATA, 0, 0);
398 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
400 * Helper function for ext4_setup_new_group_blocks() which set .
403 * @handle: journal handle
404 * @flex_gd: flex group data
406 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
407 struct ext4_new_flex_group_data *flex_gd,
408 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
410 struct ext4_sb_info *sbi = EXT4_SB(sb);
411 ext4_group_t count = last_cluster - first_cluster + 1;
414 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
416 for (count2 = count; count > 0;
417 count -= count2, first_cluster += count2) {
419 struct buffer_head *bh;
423 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
424 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
425 group -= flex_gd->groups[0].group;
427 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
431 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
432 BUG_ON(flex_gd->count > 1);
436 err = ext4_resize_ensure_credits_batch(handle, 1);
440 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
444 BUFFER_TRACE(bh, "get_write_access");
445 err = ext4_journal_get_write_access(handle, bh);
450 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
451 first_cluster, first_cluster - start, count2);
452 ext4_set_bits(bh->b_data, first_cluster - start, count2);
454 err = ext4_handle_dirty_metadata(handle, NULL, bh);
464 * Set up the block and inode bitmaps, and the inode table for the new groups.
465 * This doesn't need to be part of the main transaction, since we are only
466 * changing blocks outside the actual filesystem. We still do journaling to
467 * ensure the recovery is correct in case of a failure just after resize.
468 * If any part of this fails, we simply abort the resize.
470 * setup_new_flex_group_blocks handles a flex group as follow:
471 * 1. copy super block and GDT, and initialize group tables if necessary.
472 * In this step, we only set bits in blocks bitmaps for blocks taken by
473 * super block and GDT.
474 * 2. allocate group tables in block bitmaps, that is, set bits in block
475 * bitmap for blocks taken by group tables.
477 static int setup_new_flex_group_blocks(struct super_block *sb,
478 struct ext4_new_flex_group_data *flex_gd)
480 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
483 struct ext4_sb_info *sbi = EXT4_SB(sb);
484 struct ext4_super_block *es = sbi->s_es;
485 struct ext4_new_group_data *group_data = flex_gd->groups;
486 __u16 *bg_flags = flex_gd->bg_flags;
488 ext4_group_t group, count;
489 struct buffer_head *bh = NULL;
490 int reserved_gdb, i, j, err = 0, err2;
493 BUG_ON(!flex_gd->count || !group_data ||
494 group_data[0].group != sbi->s_groups_count);
496 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
497 meta_bg = ext4_has_feature_meta_bg(sb);
499 /* This transaction may be extended/restarted along the way */
500 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
502 return PTR_ERR(handle);
504 group = group_data[0].group;
505 for (i = 0; i < flex_gd->count; i++, group++) {
506 unsigned long gdblocks;
507 ext4_grpblk_t overhead;
509 gdblocks = ext4_bg_num_gdb(sb, group);
510 start = ext4_group_first_block_no(sb, group);
512 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
516 ext4_group_t first_group;
517 first_group = ext4_meta_bg_first_group(sb, group);
518 if (first_group != group + 1 &&
519 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
523 block = start + ext4_bg_has_super(sb, group);
524 /* Copy all of the GDT blocks into the backup in this group */
525 for (j = 0; j < gdblocks; j++, block++) {
526 struct buffer_head *gdb;
528 ext4_debug("update backup group %#04llx\n", block);
529 err = ext4_resize_ensure_credits_batch(handle, 1);
533 gdb = sb_getblk(sb, block);
534 if (unlikely(!gdb)) {
539 BUFFER_TRACE(gdb, "get_write_access");
540 err = ext4_journal_get_write_access(handle, gdb);
545 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
547 set_buffer_uptodate(gdb);
549 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
557 /* Zero out all of the reserved backup group descriptor
560 if (ext4_bg_has_super(sb, group)) {
561 err = sb_issue_zeroout(sb, gdblocks + start + 1,
562 reserved_gdb, GFP_NOFS);
568 /* Initialize group tables of the grop @group */
569 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
572 /* Zero out all of the inode table blocks */
573 block = group_data[i].inode_table;
574 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
575 block, sbi->s_itb_per_group);
576 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
582 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
585 /* Initialize block bitmap of the @group */
586 block = group_data[i].block_bitmap;
587 err = ext4_resize_ensure_credits_batch(handle, 1);
591 bh = bclean(handle, sb, block);
596 overhead = ext4_group_overhead_blocks(sb, group);
598 ext4_debug("mark backup superblock %#04llx (+0)\n",
600 ext4_set_bits(bh->b_data, 0,
601 EXT4_NUM_B2C(sbi, overhead));
603 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
604 sb->s_blocksize * 8, bh->b_data);
605 err = ext4_handle_dirty_metadata(handle, NULL, bh);
611 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
614 /* Initialize inode bitmap of the @group */
615 block = group_data[i].inode_bitmap;
616 err = ext4_resize_ensure_credits_batch(handle, 1);
619 /* Mark unused entries in inode bitmap used */
620 bh = bclean(handle, sb, block);
626 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
627 sb->s_blocksize * 8, bh->b_data);
628 err = ext4_handle_dirty_metadata(handle, NULL, bh);
634 /* Mark group tables in block bitmap */
635 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
636 count = group_table_count[j];
637 start = (&group_data[0].block_bitmap)[j];
639 for (i = 1; i < flex_gd->count; i++) {
640 block += group_table_count[j];
641 if (block == (&group_data[i].block_bitmap)[j]) {
642 count += group_table_count[j];
645 err = set_flexbg_block_bitmap(sb, handle,
647 EXT4_B2C(sbi, start),
653 count = group_table_count[j];
654 start = (&group_data[i].block_bitmap)[j];
659 err = set_flexbg_block_bitmap(sb, handle,
661 EXT4_B2C(sbi, start),
671 err2 = ext4_journal_stop(handle);
679 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
680 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
681 * calling this for the first time. In a sparse filesystem it will be the
682 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
683 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
685 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
686 unsigned *five, unsigned *seven)
688 unsigned *min = three;
692 if (!ext4_has_feature_sparse_super(sb)) {
714 * Check that all of the backup GDT blocks are held in the primary GDT block.
715 * It is assumed that they are stored in group order. Returns the number of
716 * groups in current filesystem that have BACKUPS, or -ve error code.
718 static int verify_reserved_gdb(struct super_block *sb,
720 struct buffer_head *primary)
722 const ext4_fsblk_t blk = primary->b_blocknr;
727 __le32 *p = (__le32 *)primary->b_data;
730 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
731 if (le32_to_cpu(*p++) !=
732 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
733 ext4_warning(sb, "reserved GDT %llu"
734 " missing grp %d (%llu)",
737 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
741 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
749 * Called when we need to bring a reserved group descriptor table block into
750 * use from the resize inode. The primary copy of the new GDT block currently
751 * is an indirect block (under the double indirect block in the resize inode).
752 * The new backup GDT blocks will be stored as leaf blocks in this indirect
753 * block, in group order. Even though we know all the block numbers we need,
754 * we check to ensure that the resize inode has actually reserved these blocks.
756 * Don't need to update the block bitmaps because the blocks are still in use.
758 * We get all of the error cases out of the way, so that we are sure to not
759 * fail once we start modifying the data on disk, because JBD has no rollback.
761 static int add_new_gdb(handle_t *handle, struct inode *inode,
764 struct super_block *sb = inode->i_sb;
765 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
766 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
767 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
768 struct buffer_head **o_group_desc, **n_group_desc = NULL;
769 struct buffer_head *dind = NULL;
770 struct buffer_head *gdb_bh = NULL;
772 struct ext4_iloc iloc = { .bh = NULL };
776 if (test_opt(sb, DEBUG))
778 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
781 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
783 return PTR_ERR(gdb_bh);
785 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
791 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
792 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
799 data = (__le32 *)dind->b_data;
800 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
801 ext4_warning(sb, "new group %u GDT block %llu not reserved",
807 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
808 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
812 BUFFER_TRACE(gdb_bh, "get_write_access");
813 err = ext4_journal_get_write_access(handle, gdb_bh);
817 BUFFER_TRACE(dind, "get_write_access");
818 err = ext4_journal_get_write_access(handle, dind);
820 ext4_std_error(sb, err);
822 /* ext4_reserve_inode_write() gets a reference on the iloc */
823 err = ext4_reserve_inode_write(handle, inode, &iloc);
827 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
831 ext4_warning(sb, "not enough memory for %lu groups",
837 * Finally, we have all of the possible failures behind us...
839 * Remove new GDT block from inode double-indirect block and clear out
840 * the new GDT block for use (which also "frees" the backup GDT blocks
841 * from the reserved inode). We don't need to change the bitmaps for
842 * these blocks, because they are marked as in-use from being in the
843 * reserved inode, and will become GDT blocks (primary and backup).
845 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
846 err = ext4_handle_dirty_metadata(handle, NULL, dind);
848 ext4_std_error(sb, err);
851 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
852 (9 - EXT4_SB(sb)->s_cluster_bits);
853 ext4_mark_iloc_dirty(handle, inode, &iloc);
854 memset(gdb_bh->b_data, 0, sb->s_blocksize);
855 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
857 ext4_std_error(sb, err);
863 o_group_desc = EXT4_SB(sb)->s_group_desc;
864 memcpy(n_group_desc, o_group_desc,
865 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
866 n_group_desc[gdb_num] = gdb_bh;
867 EXT4_SB(sb)->s_group_desc = n_group_desc;
868 EXT4_SB(sb)->s_gdb_count++;
869 kvfree(o_group_desc);
871 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
872 err = ext4_handle_dirty_super(handle, sb);
874 ext4_std_error(sb, err);
877 kvfree(n_group_desc);
882 ext4_debug("leaving with error %d\n", err);
887 * add_new_gdb_meta_bg is the sister of add_new_gdb.
889 static int add_new_gdb_meta_bg(struct super_block *sb,
890 handle_t *handle, ext4_group_t group) {
891 ext4_fsblk_t gdblock;
892 struct buffer_head *gdb_bh;
893 struct buffer_head **o_group_desc, **n_group_desc;
894 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
897 gdblock = ext4_meta_bg_first_block_no(sb, group) +
898 ext4_bg_has_super(sb, group);
899 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
901 return PTR_ERR(gdb_bh);
902 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
907 ext4_warning(sb, "not enough memory for %lu groups",
912 o_group_desc = 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 *));
915 n_group_desc[gdb_num] = gdb_bh;
917 BUFFER_TRACE(gdb_bh, "get_write_access");
918 err = ext4_journal_get_write_access(handle, gdb_bh);
920 kvfree(n_group_desc);
925 EXT4_SB(sb)->s_group_desc = n_group_desc;
926 EXT4_SB(sb)->s_gdb_count++;
927 kvfree(o_group_desc);
932 * Called when we are adding a new group which has a backup copy of each of
933 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
934 * We need to add these reserved backup GDT blocks to the resize inode, so
935 * that they are kept for future resizing and not allocated to files.
937 * Each reserved backup GDT block will go into a different indirect block.
938 * The indirect blocks are actually the primary reserved GDT blocks,
939 * so we know in advance what their block numbers are. We only get the
940 * double-indirect block to verify it is pointing to the primary reserved
941 * GDT blocks so we don't overwrite a data block by accident. The reserved
942 * backup GDT blocks are stored in their reserved primary GDT block.
944 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
947 struct super_block *sb = inode->i_sb;
948 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
949 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
950 struct buffer_head **primary;
951 struct buffer_head *dind;
952 struct ext4_iloc iloc;
959 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
963 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
964 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
971 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
972 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
973 EXT4_ADDR_PER_BLOCK(sb));
974 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
976 /* Get each reserved primary GDT block and verify it holds backups */
977 for (res = 0; res < reserved_gdb; res++, blk++) {
978 if (le32_to_cpu(*data) != blk) {
979 ext4_warning(sb, "reserved block %llu"
980 " not at offset %ld",
982 (long)(data - (__le32 *)dind->b_data));
986 primary[res] = ext4_sb_bread(sb, blk, 0);
987 if (IS_ERR(primary[res])) {
988 err = PTR_ERR(primary[res]);
992 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
994 brelse(primary[res]);
999 data = (__le32 *)dind->b_data;
1002 for (i = 0; i < reserved_gdb; i++) {
1003 BUFFER_TRACE(primary[i], "get_write_access");
1004 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1008 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1012 * Finally we can add each of the reserved backup GDT blocks from
1013 * the new group to its reserved primary GDT block.
1015 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1016 for (i = 0; i < reserved_gdb; i++) {
1018 data = (__le32 *)primary[i]->b_data;
1019 /* printk("reserving backup %lu[%u] = %lu\n",
1020 primary[i]->b_blocknr, gdbackups,
1021 blk + primary[i]->b_blocknr); */
1022 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1023 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1028 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1029 ext4_mark_iloc_dirty(handle, inode, &iloc);
1033 brelse(primary[res]);
1043 * Update the backup copies of the ext4 metadata. These don't need to be part
1044 * of the main resize transaction, because e2fsck will re-write them if there
1045 * is a problem (basically only OOM will cause a problem). However, we
1046 * _should_ update the backups if possible, in case the primary gets trashed
1047 * for some reason and we need to run e2fsck from a backup superblock. The
1048 * important part is that the new block and inode counts are in the backup
1049 * superblocks, and the location of the new group metadata in the GDT backups.
1051 * We do not need take the s_resize_lock for this, because these
1052 * blocks are not otherwise touched by the filesystem code when it is
1053 * mounted. We don't need to worry about last changing from
1054 * sbi->s_groups_count, because the worst that can happen is that we
1055 * do not copy the full number of backups at this time. The resize
1056 * which changed s_groups_count will backup again.
1058 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1059 int size, int meta_bg)
1061 struct ext4_sb_info *sbi = EXT4_SB(sb);
1063 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1067 ext4_group_t group = 0;
1068 int rest = sb->s_blocksize - size;
1072 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1073 if (IS_ERR(handle)) {
1075 err = PTR_ERR(handle);
1080 group = ext4_list_backups(sb, &three, &five, &seven);
1081 last = sbi->s_groups_count;
1083 group = ext4_get_group_number(sb, blk_off) + 1;
1084 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1087 while (group < sbi->s_groups_count) {
1088 struct buffer_head *bh;
1089 ext4_fsblk_t backup_block;
1091 /* Out of journal space, and can't get more - abort - so sad */
1092 err = ext4_resize_ensure_credits_batch(handle, 1);
1097 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1099 backup_block = (ext4_group_first_block_no(sb, group) +
1100 ext4_bg_has_super(sb, group));
1102 bh = sb_getblk(sb, backup_block);
1103 if (unlikely(!bh)) {
1107 ext4_debug("update metadata backup %llu(+%llu)\n",
1108 backup_block, backup_block -
1109 ext4_group_first_block_no(sb, group));
1110 BUFFER_TRACE(bh, "get_write_access");
1111 if ((err = ext4_journal_get_write_access(handle, bh))) {
1116 memcpy(bh->b_data, data, size);
1118 memset(bh->b_data + size, 0, rest);
1119 set_buffer_uptodate(bh);
1121 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1123 ext4_std_error(sb, err);
1127 group = ext4_list_backups(sb, &three, &five, &seven);
1128 else if (group == last)
1133 if ((err2 = ext4_journal_stop(handle)) && !err)
1137 * Ugh! Need to have e2fsck write the backup copies. It is too
1138 * late to revert the resize, we shouldn't fail just because of
1139 * the backup copies (they are only needed in case of corruption).
1141 * However, if we got here we have a journal problem too, so we
1142 * can't really start a transaction to mark the superblock.
1143 * Chicken out and just set the flag on the hope it will be written
1144 * to disk, and if not - we will simply wait until next fsck.
1148 ext4_warning(sb, "can't update backup for group %u (err %d), "
1149 "forcing fsck on next reboot", group, err);
1150 sbi->s_mount_state &= ~EXT4_VALID_FS;
1151 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1152 mark_buffer_dirty(sbi->s_sbh);
1157 * ext4_add_new_descs() adds @count group descriptor of groups
1158 * starting at @group
1160 * @handle: journal handle
1162 * @group: the group no. of the first group desc to be added
1163 * @resize_inode: the resize inode
1164 * @count: number of group descriptors to be added
1166 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1167 ext4_group_t group, struct inode *resize_inode,
1170 struct ext4_sb_info *sbi = EXT4_SB(sb);
1171 struct ext4_super_block *es = sbi->s_es;
1172 struct buffer_head *gdb_bh;
1173 int i, gdb_off, gdb_num, err = 0;
1176 meta_bg = ext4_has_feature_meta_bg(sb);
1177 for (i = 0; i < count; i++, group++) {
1178 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1179 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1181 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1182 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1185 * We will only either add reserved group blocks to a backup group
1186 * or remove reserved blocks for the first group in a new group block.
1187 * Doing both would be mean more complex code, and sane people don't
1188 * use non-sparse filesystems anymore. This is already checked above.
1191 gdb_bh = sbi->s_group_desc[gdb_num];
1192 BUFFER_TRACE(gdb_bh, "get_write_access");
1193 err = ext4_journal_get_write_access(handle, gdb_bh);
1195 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1196 err = reserve_backup_gdb(handle, resize_inode, group);
1197 } else if (meta_bg != 0) {
1198 err = add_new_gdb_meta_bg(sb, handle, group);
1200 err = add_new_gdb(handle, resize_inode, group);
1208 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1210 struct buffer_head *bh = sb_getblk(sb, block);
1213 if (!bh_uptodate_or_lock(bh)) {
1214 if (bh_submit_read(bh) < 0) {
1223 static int ext4_set_bitmap_checksums(struct super_block *sb,
1225 struct ext4_group_desc *gdp,
1226 struct ext4_new_group_data *group_data)
1228 struct buffer_head *bh;
1230 if (!ext4_has_metadata_csum(sb))
1233 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1236 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1237 EXT4_INODES_PER_GROUP(sb) / 8);
1240 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1243 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1250 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1252 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1253 struct ext4_new_flex_group_data *flex_gd)
1255 struct ext4_new_group_data *group_data = flex_gd->groups;
1256 struct ext4_group_desc *gdp;
1257 struct ext4_sb_info *sbi = EXT4_SB(sb);
1258 struct buffer_head *gdb_bh;
1260 __u16 *bg_flags = flex_gd->bg_flags;
1261 int i, gdb_off, gdb_num, err = 0;
1264 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1265 group = group_data->group;
1267 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1268 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1271 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1273 gdb_bh = sbi->s_group_desc[gdb_num];
1274 /* Update group descriptor block for new group */
1275 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1276 gdb_off * EXT4_DESC_SIZE(sb));
1278 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1279 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1280 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1281 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1283 ext4_std_error(sb, err);
1287 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1288 ext4_free_group_clusters_set(sb, gdp,
1289 group_data->free_clusters_count);
1290 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1291 if (ext4_has_group_desc_csum(sb))
1292 ext4_itable_unused_set(sb, gdp,
1293 EXT4_INODES_PER_GROUP(sb));
1294 gdp->bg_flags = cpu_to_le16(*bg_flags);
1295 ext4_group_desc_csum_set(sb, group, gdp);
1297 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1298 if (unlikely(err)) {
1299 ext4_std_error(sb, err);
1304 * We can allocate memory for mb_alloc based on the new group
1307 err = ext4_mb_add_groupinfo(sb, group, gdp);
1315 * ext4_update_super() updates the super block so that the newly added
1316 * groups can be seen by the filesystem.
1319 * @flex_gd: new added groups
1321 static void ext4_update_super(struct super_block *sb,
1322 struct ext4_new_flex_group_data *flex_gd)
1324 ext4_fsblk_t blocks_count = 0;
1325 ext4_fsblk_t free_blocks = 0;
1326 ext4_fsblk_t reserved_blocks = 0;
1327 struct ext4_new_group_data *group_data = flex_gd->groups;
1328 struct ext4_sb_info *sbi = EXT4_SB(sb);
1329 struct ext4_super_block *es = sbi->s_es;
1332 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1334 * Make the new blocks and inodes valid next. We do this before
1335 * increasing the group count so that once the group is enabled,
1336 * all of its blocks and inodes are already valid.
1338 * We always allocate group-by-group, then block-by-block or
1339 * inode-by-inode within a group, so enabling these
1340 * blocks/inodes before the group is live won't actually let us
1341 * allocate the new space yet.
1343 for (i = 0; i < flex_gd->count; i++) {
1344 blocks_count += group_data[i].blocks_count;
1345 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1348 reserved_blocks = ext4_r_blocks_count(es) * 100;
1349 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1350 reserved_blocks *= blocks_count;
1351 do_div(reserved_blocks, 100);
1353 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1354 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1355 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1357 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1360 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1362 * We need to protect s_groups_count against other CPUs seeing
1363 * inconsistent state in the superblock.
1365 * The precise rules we use are:
1367 * * Writers must perform a smp_wmb() after updating all
1368 * dependent data and before modifying the groups count
1370 * * Readers must perform an smp_rmb() after reading the groups
1371 * count and before reading any dependent data.
1373 * NB. These rules can be relaxed when checking the group count
1374 * while freeing data, as we can only allocate from a block
1375 * group after serialising against the group count, and we can
1376 * only then free after serialising in turn against that
1381 /* Update the global fs size fields */
1382 sbi->s_groups_count += flex_gd->count;
1383 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1384 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1386 /* Update the reserved block counts only once the new group is
1388 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1391 /* Update the free space counts */
1392 percpu_counter_add(&sbi->s_freeclusters_counter,
1393 EXT4_NUM_B2C(sbi, free_blocks));
1394 percpu_counter_add(&sbi->s_freeinodes_counter,
1395 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1397 ext4_debug("free blocks count %llu",
1398 percpu_counter_read(&sbi->s_freeclusters_counter));
1399 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1400 ext4_group_t flex_group;
1401 flex_group = ext4_flex_group(sbi, group_data[0].group);
1402 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1403 &sbi->s_flex_groups[flex_group].free_clusters);
1404 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1405 &sbi->s_flex_groups[flex_group].free_inodes);
1409 * Update the fs overhead information
1411 ext4_calculate_overhead(sb);
1413 if (test_opt(sb, DEBUG))
1414 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1415 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1416 blocks_count, free_blocks, reserved_blocks);
1419 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1420 * _before_ we start modifying the filesystem, because we cannot abort the
1421 * transaction and not have it write the data to disk.
1423 static int ext4_flex_group_add(struct super_block *sb,
1424 struct inode *resize_inode,
1425 struct ext4_new_flex_group_data *flex_gd)
1427 struct ext4_sb_info *sbi = EXT4_SB(sb);
1428 struct ext4_super_block *es = sbi->s_es;
1429 ext4_fsblk_t o_blocks_count;
1433 unsigned reserved_gdb;
1434 int err = 0, err2 = 0, credit;
1436 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1438 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1439 o_blocks_count = ext4_blocks_count(es);
1440 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1443 err = setup_new_flex_group_blocks(sb, flex_gd);
1447 * We will always be modifying at least the superblock and GDT
1448 * blocks. If we are adding a group past the last current GDT block,
1449 * we will also modify the inode and the dindirect block. If we
1450 * are adding a group with superblock/GDT backups we will also
1451 * modify each of the reserved GDT dindirect blocks.
1453 credit = 3; /* sb, resize inode, resize inode dindirect */
1455 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1456 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1457 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1458 if (IS_ERR(handle)) {
1459 err = PTR_ERR(handle);
1463 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1464 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1468 group = flex_gd->groups[0].group;
1469 BUG_ON(group != sbi->s_groups_count);
1470 err = ext4_add_new_descs(handle, sb, group,
1471 resize_inode, flex_gd->count);
1475 err = ext4_setup_new_descs(handle, sb, flex_gd);
1479 ext4_update_super(sb, flex_gd);
1481 err = ext4_handle_dirty_super(handle, sb);
1484 err2 = ext4_journal_stop(handle);
1489 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1490 int gdb_num_end = ((group + flex_gd->count - 1) /
1491 EXT4_DESC_PER_BLOCK(sb));
1492 int meta_bg = ext4_has_feature_meta_bg(sb);
1493 sector_t old_gdb = 0;
1495 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1496 sizeof(struct ext4_super_block), 0);
1497 for (; gdb_num <= gdb_num_end; gdb_num++) {
1498 struct buffer_head *gdb_bh;
1500 gdb_bh = sbi->s_group_desc[gdb_num];
1501 if (old_gdb == gdb_bh->b_blocknr)
1503 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1504 gdb_bh->b_size, meta_bg);
1505 old_gdb = gdb_bh->b_blocknr;
1512 static int ext4_setup_next_flex_gd(struct super_block *sb,
1513 struct ext4_new_flex_group_data *flex_gd,
1514 ext4_fsblk_t n_blocks_count,
1515 unsigned long flexbg_size)
1517 struct ext4_sb_info *sbi = EXT4_SB(sb);
1518 struct ext4_super_block *es = sbi->s_es;
1519 struct ext4_new_group_data *group_data = flex_gd->groups;
1520 ext4_fsblk_t o_blocks_count;
1521 ext4_group_t n_group;
1523 ext4_group_t last_group;
1525 ext4_grpblk_t clusters_per_group;
1528 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1530 o_blocks_count = ext4_blocks_count(es);
1532 if (o_blocks_count == n_blocks_count)
1535 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1537 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1539 last_group = group | (flexbg_size - 1);
1540 if (last_group > n_group)
1541 last_group = n_group;
1543 flex_gd->count = last_group - group + 1;
1545 for (i = 0; i < flex_gd->count; i++) {
1548 group_data[i].group = group + i;
1549 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1550 overhead = ext4_group_overhead_blocks(sb, group + i);
1551 group_data[i].mdata_blocks = overhead;
1552 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1553 if (ext4_has_group_desc_csum(sb)) {
1554 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1555 EXT4_BG_INODE_UNINIT;
1556 if (!test_opt(sb, INIT_INODE_TABLE))
1557 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1559 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1562 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1563 /* We need to initialize block bitmap of last group. */
1564 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1566 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1567 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1568 group_data[i - 1].free_clusters_count -= clusters_per_group -
1575 /* Add group descriptor data to an existing or new group descriptor block.
1576 * Ensure we handle all possible error conditions _before_ we start modifying
1577 * the filesystem, because we cannot abort the transaction and not have it
1578 * write the data to disk.
1580 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1581 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1583 * We only need to hold the superblock lock while we are actually adding
1584 * in the new group's counts to the superblock. Prior to that we have
1585 * not really "added" the group at all. We re-check that we are still
1586 * adding in the last group in case things have changed since verifying.
1588 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1590 struct ext4_new_flex_group_data flex_gd;
1591 struct ext4_sb_info *sbi = EXT4_SB(sb);
1592 struct ext4_super_block *es = sbi->s_es;
1593 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1594 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1595 struct inode *inode = NULL;
1600 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1602 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1603 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1607 if (ext4_blocks_count(es) + input->blocks_count <
1608 ext4_blocks_count(es)) {
1609 ext4_warning(sb, "blocks_count overflow");
1613 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1614 le32_to_cpu(es->s_inodes_count)) {
1615 ext4_warning(sb, "inodes_count overflow");
1619 if (reserved_gdb || gdb_off == 0) {
1620 if (!ext4_has_feature_resize_inode(sb) ||
1621 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1623 "No reserved GDT blocks, can't resize");
1626 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1627 if (IS_ERR(inode)) {
1628 ext4_warning(sb, "Error opening resize inode");
1629 return PTR_ERR(inode);
1634 err = verify_group_input(sb, input);
1638 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1642 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1647 flex_gd.groups = input;
1648 flex_gd.bg_flags = &bg_flags;
1649 err = ext4_flex_group_add(sb, inode, &flex_gd);
1653 } /* ext4_group_add */
1656 * extend a group without checking assuming that checking has been done.
1658 static int ext4_group_extend_no_check(struct super_block *sb,
1659 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1661 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1665 /* We will update the superblock, one block bitmap, and
1666 * one group descriptor via ext4_group_add_blocks().
1668 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1669 if (IS_ERR(handle)) {
1670 err = PTR_ERR(handle);
1671 ext4_warning(sb, "error %d on journal start", err);
1675 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1676 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1678 ext4_warning(sb, "error %d on journal write access", err);
1682 ext4_blocks_count_set(es, o_blocks_count + add);
1683 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1684 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1685 o_blocks_count + add);
1686 /* We add the blocks to the bitmap and set the group need init bit */
1687 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1690 ext4_handle_dirty_super(handle, sb);
1691 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1692 o_blocks_count + add);
1694 err2 = ext4_journal_stop(handle);
1699 if (test_opt(sb, DEBUG))
1700 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1701 "blocks\n", ext4_blocks_count(es));
1702 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1703 (char *)es, sizeof(struct ext4_super_block), 0);
1709 * Extend the filesystem to the new number of blocks specified. This entry
1710 * point is only used to extend the current filesystem to the end of the last
1711 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1712 * for emergencies (because it has no dependencies on reserved blocks).
1714 * If we _really_ wanted, we could use default values to call ext4_group_add()
1715 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1716 * GDT blocks are reserved to grow to the desired size.
1718 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1719 ext4_fsblk_t n_blocks_count)
1721 ext4_fsblk_t o_blocks_count;
1724 struct buffer_head *bh;
1728 o_blocks_count = ext4_blocks_count(es);
1730 if (test_opt(sb, DEBUG))
1731 ext4_msg(sb, KERN_DEBUG,
1732 "extending last group from %llu to %llu blocks",
1733 o_blocks_count, n_blocks_count);
1735 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1738 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1739 ext4_msg(sb, KERN_ERR,
1740 "filesystem too large to resize to %llu blocks safely",
1745 if (n_blocks_count < o_blocks_count) {
1746 ext4_warning(sb, "can't shrink FS - resize aborted");
1750 /* Handle the remaining blocks in the last group only. */
1751 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1754 ext4_warning(sb, "need to use ext2online to resize further");
1758 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1760 if (o_blocks_count + add < o_blocks_count) {
1761 ext4_warning(sb, "blocks_count overflow");
1765 if (o_blocks_count + add > n_blocks_count)
1766 add = n_blocks_count - o_blocks_count;
1768 if (o_blocks_count + add < n_blocks_count)
1769 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1770 o_blocks_count + add, add);
1772 /* See if the device is actually as big as what was requested */
1773 bh = sb_bread(sb, o_blocks_count + add - 1);
1775 ext4_warning(sb, "can't read last block, resize aborted");
1780 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1782 } /* ext4_group_extend */
1785 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1787 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1791 * Release the resize inode and drop the resize_inode feature if there
1792 * are no more reserved gdt blocks, and then convert the file system
1795 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1798 struct ext4_sb_info *sbi = EXT4_SB(sb);
1799 struct ext4_super_block *es = sbi->s_es;
1800 struct ext4_inode_info *ei = EXT4_I(inode);
1802 int i, ret, err = 0;
1805 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1807 if (es->s_reserved_gdt_blocks) {
1808 ext4_error(sb, "Unexpected non-zero "
1809 "s_reserved_gdt_blocks");
1813 /* Do a quick sanity check of the resize inode */
1814 if (inode->i_blocks != 1 << (inode->i_blkbits -
1815 (9 - sbi->s_cluster_bits)))
1816 goto invalid_resize_inode;
1817 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1818 if (i == EXT4_DIND_BLOCK) {
1822 goto invalid_resize_inode;
1825 goto invalid_resize_inode;
1827 credits += 3; /* block bitmap, bg descriptor, resize inode */
1830 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1832 return PTR_ERR(handle);
1834 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1835 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1839 ext4_clear_feature_resize_inode(sb);
1840 ext4_set_feature_meta_bg(sb);
1841 sbi->s_es->s_first_meta_bg =
1842 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1844 err = ext4_handle_dirty_super(handle, sb);
1846 ext4_std_error(sb, err);
1851 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1852 ext4_free_blocks(handle, inode, NULL, nr, 1,
1853 EXT4_FREE_BLOCKS_METADATA |
1854 EXT4_FREE_BLOCKS_FORGET);
1855 ei->i_data[EXT4_DIND_BLOCK] = 0;
1856 inode->i_blocks = 0;
1858 err = ext4_mark_inode_dirty(handle, inode);
1860 ext4_std_error(sb, err);
1864 ret = ext4_journal_stop(handle);
1869 invalid_resize_inode:
1870 ext4_error(sb, "corrupted/inconsistent resize inode");
1875 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1877 * @sb: super block of the fs to be resized
1878 * @n_blocks_count: the number of blocks resides in the resized fs
1880 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1882 struct ext4_new_flex_group_data *flex_gd = NULL;
1883 struct ext4_sb_info *sbi = EXT4_SB(sb);
1884 struct ext4_super_block *es = sbi->s_es;
1885 struct buffer_head *bh;
1886 struct inode *resize_inode = NULL;
1887 ext4_grpblk_t add, offset;
1888 unsigned long n_desc_blocks;
1889 unsigned long o_desc_blocks;
1890 ext4_group_t o_group;
1891 ext4_group_t n_group;
1892 ext4_fsblk_t o_blocks_count;
1893 ext4_fsblk_t n_blocks_count_retry = 0;
1894 unsigned long last_update_time = 0;
1895 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1898 /* See if the device is actually as big as what was requested */
1899 bh = sb_bread(sb, n_blocks_count - 1);
1901 ext4_warning(sb, "can't read last block, resize aborted");
1907 o_blocks_count = ext4_blocks_count(es);
1909 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1910 "to %llu blocks", o_blocks_count, n_blocks_count);
1912 if (n_blocks_count < o_blocks_count) {
1913 /* On-line shrinking not supported */
1914 ext4_warning(sb, "can't shrink FS - resize aborted");
1918 if (n_blocks_count == o_blocks_count)
1919 /* Nothing need to do */
1922 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1923 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1924 ext4_warning(sb, "resize would cause inodes_count overflow");
1927 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1929 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1930 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1932 meta_bg = ext4_has_feature_meta_bg(sb);
1934 if (ext4_has_feature_resize_inode(sb)) {
1936 ext4_error(sb, "resize_inode and meta_bg enabled "
1940 if (n_desc_blocks > o_desc_blocks +
1941 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1942 n_blocks_count_retry = n_blocks_count;
1943 n_desc_blocks = o_desc_blocks +
1944 le16_to_cpu(es->s_reserved_gdt_blocks);
1945 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1946 n_blocks_count = (ext4_fsblk_t)n_group *
1947 EXT4_BLOCKS_PER_GROUP(sb) +
1948 le32_to_cpu(es->s_first_data_block);
1949 n_group--; /* set to last group number */
1953 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1955 if (IS_ERR(resize_inode)) {
1956 ext4_warning(sb, "Error opening resize inode");
1957 return PTR_ERR(resize_inode);
1961 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1962 err = ext4_convert_meta_bg(sb, resize_inode);
1967 resize_inode = NULL;
1969 if (n_blocks_count_retry) {
1970 n_blocks_count = n_blocks_count_retry;
1971 n_blocks_count_retry = 0;
1977 * Make sure the last group has enough space so that it's
1978 * guaranteed to have enough space for all metadata blocks
1979 * that it might need to hold. (We might not need to store
1980 * the inode table blocks in the last block group, but there
1981 * will be cases where this might be needed.)
1983 if ((ext4_group_first_block_no(sb, n_group) +
1984 ext4_group_overhead_blocks(sb, n_group) + 2 +
1985 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
1986 n_blocks_count = ext4_group_first_block_no(sb, n_group);
1988 n_blocks_count_retry = 0;
1991 resize_inode = NULL;
1996 /* extend the last group */
1997 if (n_group == o_group)
1998 add = n_blocks_count - o_blocks_count;
2000 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2002 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2007 if (ext4_blocks_count(es) == n_blocks_count)
2010 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2014 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2018 flex_gd = alloc_flex_gd(flexbg_size);
2019 if (flex_gd == NULL) {
2024 /* Add flex groups. Note that a regular group is a
2025 * flex group with 1 group.
2027 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2029 if (jiffies - last_update_time > HZ * 10) {
2030 if (last_update_time)
2031 ext4_msg(sb, KERN_INFO,
2032 "resized to %llu blocks",
2033 ext4_blocks_count(es));
2034 last_update_time = jiffies;
2036 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2038 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2043 if (!err && n_blocks_count_retry) {
2044 n_blocks_count = n_blocks_count_retry;
2045 n_blocks_count_retry = 0;
2046 free_flex_gd(flex_gd);
2050 resize_inode = NULL;
2057 free_flex_gd(flex_gd);
2058 if (resize_inode != NULL)
2061 ext4_warning(sb, "error (%d) occurred during "
2062 "file system resize", err);
2063 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2064 ext4_blocks_count(es));