2 * linux/fs/ext4/resize.c
4 * Support for resizing an ext4 filesystem while it is mounted.
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 * This could probably be made into a module, because it is not often in use.
14 #include <linux/errno.h>
15 #include <linux/slab.h>
17 #include "ext4_jbd2.h"
19 int ext4_resize_begin(struct super_block *sb)
23 if (!capable(CAP_SYS_RESOURCE))
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
36 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
42 void ext4_resize_end(struct super_block *sb)
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
49 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
51 static int verify_group_input(struct super_block *sb,
52 struct ext4_new_group_data *input)
54 struct ext4_sb_info *sbi = EXT4_SB(sb);
55 struct ext4_super_block *es = sbi->s_es;
56 ext4_fsblk_t start = ext4_blocks_count(es);
57 ext4_fsblk_t end = start + input->blocks_count;
58 ext4_group_t group = input->group;
59 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 unsigned overhead = ext4_bg_has_super(sb, group) ?
61 (1 + ext4_bg_num_gdb(sb, group) +
62 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
63 ext4_fsblk_t metaend = start + overhead;
64 struct buffer_head *bh = NULL;
65 ext4_grpblk_t free_blocks_count, offset;
68 input->free_blocks_count = free_blocks_count =
69 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
71 if (test_opt(sb, DEBUG))
72 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
73 "(%d free, %u reserved)\n",
74 ext4_bg_has_super(sb, input->group) ? "normal" :
75 "no-super", input->group, input->blocks_count,
76 free_blocks_count, input->reserved_blocks);
78 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
79 if (group != sbi->s_groups_count)
80 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
81 input->group, sbi->s_groups_count);
83 ext4_warning(sb, "Last group not full");
84 else if (input->reserved_blocks > input->blocks_count / 5)
85 ext4_warning(sb, "Reserved blocks too high (%u)",
86 input->reserved_blocks);
87 else if (free_blocks_count < 0)
88 ext4_warning(sb, "Bad blocks count %u",
90 else if (!(bh = sb_bread(sb, end - 1)))
91 ext4_warning(sb, "Cannot read last block (%llu)",
93 else if (outside(input->block_bitmap, start, end))
94 ext4_warning(sb, "Block bitmap not in group (block %llu)",
95 (unsigned long long)input->block_bitmap);
96 else if (outside(input->inode_bitmap, start, end))
97 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
98 (unsigned long long)input->inode_bitmap);
99 else if (outside(input->inode_table, start, end) ||
100 outside(itend - 1, start, end))
101 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
102 (unsigned long long)input->inode_table, itend - 1);
103 else if (input->inode_bitmap == input->block_bitmap)
104 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
105 (unsigned long long)input->block_bitmap);
106 else if (inside(input->block_bitmap, input->inode_table, itend))
107 ext4_warning(sb, "Block bitmap (%llu) in inode table "
109 (unsigned long long)input->block_bitmap,
110 (unsigned long long)input->inode_table, itend - 1);
111 else if (inside(input->inode_bitmap, input->inode_table, itend))
112 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
114 (unsigned long long)input->inode_bitmap,
115 (unsigned long long)input->inode_table, itend - 1);
116 else if (inside(input->block_bitmap, start, metaend))
117 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
118 (unsigned long long)input->block_bitmap,
120 else if (inside(input->inode_bitmap, start, metaend))
121 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
122 (unsigned long long)input->inode_bitmap,
124 else if (inside(input->inode_table, start, metaend) ||
125 inside(itend - 1, start, metaend))
126 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
128 (unsigned long long)input->inode_table,
129 itend - 1, start, metaend - 1);
138 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
141 struct ext4_new_flex_group_data {
142 struct ext4_new_group_data *groups; /* new_group_data for groups
144 __u16 *bg_flags; /* block group flags of groups
146 ext4_group_t count; /* number of groups in @groups
151 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
154 * Returns NULL on failure otherwise address of the allocated structure.
156 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
158 struct ext4_new_flex_group_data *flex_gd;
160 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
164 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
166 flex_gd->count = flexbg_size;
168 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
169 flexbg_size, GFP_NOFS);
170 if (flex_gd->groups == NULL)
173 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
174 if (flex_gd->bg_flags == NULL)
180 kfree(flex_gd->groups);
187 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
189 kfree(flex_gd->bg_flags);
190 kfree(flex_gd->groups);
195 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
196 * and inode tables for a flex group.
198 * This function is used by 64bit-resize. Note that this function allocates
199 * group tables from the 1st group of groups contained by @flexgd, which may
200 * be a partial of a flex group.
202 * @sb: super block of fs to which the groups belongs
204 * Returns 0 on a successful allocation of the metadata blocks in the
207 static int ext4_alloc_group_tables(struct super_block *sb,
208 struct ext4_new_flex_group_data *flex_gd,
211 struct ext4_new_group_data *group_data = flex_gd->groups;
212 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
213 ext4_fsblk_t start_blk;
214 ext4_fsblk_t last_blk;
215 ext4_group_t src_group;
216 ext4_group_t bb_index = 0;
217 ext4_group_t ib_index = 0;
218 ext4_group_t it_index = 0;
220 ext4_group_t last_group;
223 BUG_ON(flex_gd->count == 0 || group_data == NULL);
225 src_group = group_data[0].group;
226 last_group = src_group + flex_gd->count - 1;
228 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
229 (last_group & ~(flexbg_size - 1))));
231 group = group_data[0].group;
232 if (src_group >= group_data[0].group + flex_gd->count)
234 start_blk = ext4_group_first_block_no(sb, src_group);
235 last_blk = start_blk + group_data[src_group - group].blocks_count;
237 overhead = ext4_bg_has_super(sb, src_group) ?
238 (1 + ext4_bg_num_gdb(sb, src_group) +
239 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
241 start_blk += overhead;
243 /* We collect contiguous blocks as much as possible. */
245 for (; src_group <= last_group; src_group++)
246 if (!ext4_bg_has_super(sb, src_group))
247 last_blk += group_data[src_group - group].blocks_count;
251 /* Allocate block bitmaps */
252 for (; bb_index < flex_gd->count; bb_index++) {
253 if (start_blk >= last_blk)
255 group_data[bb_index].block_bitmap = start_blk++;
256 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
257 group -= group_data[0].group;
258 group_data[group].free_blocks_count--;
260 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
263 /* Allocate inode bitmaps */
264 for (; ib_index < flex_gd->count; ib_index++) {
265 if (start_blk >= last_blk)
267 group_data[ib_index].inode_bitmap = start_blk++;
268 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
269 group -= group_data[0].group;
270 group_data[group].free_blocks_count--;
272 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
275 /* Allocate inode tables */
276 for (; it_index < flex_gd->count; it_index++) {
277 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
279 group_data[it_index].inode_table = start_blk;
280 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
281 group -= group_data[0].group;
282 group_data[group].free_blocks_count -=
283 EXT4_SB(sb)->s_itb_per_group;
285 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
287 start_blk += EXT4_SB(sb)->s_itb_per_group;
290 if (test_opt(sb, DEBUG)) {
292 group = group_data[0].group;
294 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
295 "%d groups, flexbg size is %d:\n", flex_gd->count,
298 for (i = 0; i < flex_gd->count; i++) {
299 printk(KERN_DEBUG "adding %s group %u: %u "
300 "blocks (%d free)\n",
301 ext4_bg_has_super(sb, group + i) ? "normal" :
302 "no-super", group + i,
303 group_data[i].blocks_count,
304 group_data[i].free_blocks_count);
310 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
313 struct buffer_head *bh;
316 bh = sb_getblk(sb, blk);
318 return ERR_PTR(-EIO);
319 if ((err = ext4_journal_get_write_access(handle, bh))) {
323 memset(bh->b_data, 0, sb->s_blocksize);
324 set_buffer_uptodate(bh);
331 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
332 * If that fails, restart the transaction & regain write access for the
333 * buffer head which is used for block_bitmap modifications.
335 static int extend_or_restart_transaction(handle_t *handle, int thresh)
339 if (ext4_handle_has_enough_credits(handle, thresh))
342 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
346 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
355 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
357 * Helper function for ext4_setup_new_group_blocks() which set .
360 * @handle: journal handle
361 * @flex_gd: flex group data
363 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
364 struct ext4_new_flex_group_data *flex_gd,
365 ext4_fsblk_t block, ext4_group_t count)
369 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
370 for (count2 = count; count > 0; count -= count2, block += count2) {
372 struct buffer_head *bh;
376 ext4_get_group_no_and_offset(sb, block, &group, NULL);
377 start = ext4_group_first_block_no(sb, group);
378 group -= flex_gd->groups[0].group;
380 count2 = sb->s_blocksize * 8 - (block - start);
384 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
385 BUG_ON(flex_gd->count > 1);
389 err = extend_or_restart_transaction(handle, 1);
393 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
397 err = ext4_journal_get_write_access(handle, bh);
400 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
401 block - start, count2);
402 ext4_set_bits(bh->b_data, block - start, count2);
404 err = ext4_handle_dirty_metadata(handle, NULL, bh);
414 * Set up the block and inode bitmaps, and the inode table for the new groups.
415 * This doesn't need to be part of the main transaction, since we are only
416 * changing blocks outside the actual filesystem. We still do journaling to
417 * ensure the recovery is correct in case of a failure just after resize.
418 * If any part of this fails, we simply abort the resize.
420 * setup_new_flex_group_blocks handles a flex group as follow:
421 * 1. copy super block and GDT, and initialize group tables if necessary.
422 * In this step, we only set bits in blocks bitmaps for blocks taken by
423 * super block and GDT.
424 * 2. allocate group tables in block bitmaps, that is, set bits in block
425 * bitmap for blocks taken by group tables.
427 static int setup_new_flex_group_blocks(struct super_block *sb,
428 struct ext4_new_flex_group_data *flex_gd)
430 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
433 struct ext4_sb_info *sbi = EXT4_SB(sb);
434 struct ext4_super_block *es = sbi->s_es;
435 struct ext4_new_group_data *group_data = flex_gd->groups;
436 __u16 *bg_flags = flex_gd->bg_flags;
438 ext4_group_t group, count;
439 struct buffer_head *bh = NULL;
440 int reserved_gdb, i, j, err = 0, err2;
442 BUG_ON(!flex_gd->count || !group_data ||
443 group_data[0].group != sbi->s_groups_count);
445 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
447 /* This transaction may be extended/restarted along the way */
448 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
450 return PTR_ERR(handle);
452 group = group_data[0].group;
453 for (i = 0; i < flex_gd->count; i++, group++) {
454 unsigned long gdblocks;
456 gdblocks = ext4_bg_num_gdb(sb, group);
457 start = ext4_group_first_block_no(sb, group);
459 /* Copy all of the GDT blocks into the backup in this group */
460 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
461 struct buffer_head *gdb;
463 ext4_debug("update backup group %#04llx\n", block);
464 err = extend_or_restart_transaction(handle, 1);
468 gdb = sb_getblk(sb, block);
474 err = ext4_journal_get_write_access(handle, gdb);
479 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
481 set_buffer_uptodate(gdb);
483 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
491 /* Zero out all of the reserved backup group descriptor
494 if (ext4_bg_has_super(sb, group)) {
495 err = sb_issue_zeroout(sb, gdblocks + start + 1,
496 reserved_gdb, GFP_NOFS);
501 /* Initialize group tables of the grop @group */
502 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
505 /* Zero out all of the inode table blocks */
506 block = group_data[i].inode_table;
507 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
508 block, sbi->s_itb_per_group);
509 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
515 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
518 /* Initialize block bitmap of the @group */
519 block = group_data[i].block_bitmap;
520 err = extend_or_restart_transaction(handle, 1);
524 bh = bclean(handle, sb, block);
529 if (ext4_bg_has_super(sb, group)) {
530 ext4_debug("mark backup superblock %#04llx (+0)\n",
532 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
535 ext4_mark_bitmap_end(group_data[i].blocks_count,
536 sb->s_blocksize * 8, bh->b_data);
537 err = ext4_handle_dirty_metadata(handle, NULL, bh);
543 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
546 /* Initialize inode bitmap of the @group */
547 block = group_data[i].inode_bitmap;
548 err = extend_or_restart_transaction(handle, 1);
551 /* Mark unused entries in inode bitmap used */
552 bh = bclean(handle, sb, block);
558 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
559 sb->s_blocksize * 8, bh->b_data);
560 err = ext4_handle_dirty_metadata(handle, NULL, bh);
567 /* Mark group tables in block bitmap */
568 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
569 count = group_table_count[j];
570 start = (&group_data[0].block_bitmap)[j];
572 for (i = 1; i < flex_gd->count; i++) {
573 block += group_table_count[j];
574 if (block == (&group_data[i].block_bitmap)[j]) {
575 count += group_table_count[j];
578 err = set_flexbg_block_bitmap(sb, handle,
579 flex_gd, start, count);
582 count = group_table_count[j];
583 start = group_data[i].block_bitmap;
588 err = set_flexbg_block_bitmap(sb, handle,
589 flex_gd, start, count);
597 err2 = ext4_journal_stop(handle);
605 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
606 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
607 * calling this for the first time. In a sparse filesystem it will be the
608 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
609 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
611 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
612 unsigned *five, unsigned *seven)
614 unsigned *min = three;
618 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
619 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
641 * Check that all of the backup GDT blocks are held in the primary GDT block.
642 * It is assumed that they are stored in group order. Returns the number of
643 * groups in current filesystem that have BACKUPS, or -ve error code.
645 static int verify_reserved_gdb(struct super_block *sb,
647 struct buffer_head *primary)
649 const ext4_fsblk_t blk = primary->b_blocknr;
654 __le32 *p = (__le32 *)primary->b_data;
657 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
658 if (le32_to_cpu(*p++) !=
659 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
660 ext4_warning(sb, "reserved GDT %llu"
661 " missing grp %d (%llu)",
664 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
668 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
676 * Called when we need to bring a reserved group descriptor table block into
677 * use from the resize inode. The primary copy of the new GDT block currently
678 * is an indirect block (under the double indirect block in the resize inode).
679 * The new backup GDT blocks will be stored as leaf blocks in this indirect
680 * block, in group order. Even though we know all the block numbers we need,
681 * we check to ensure that the resize inode has actually reserved these blocks.
683 * Don't need to update the block bitmaps because the blocks are still in use.
685 * We get all of the error cases out of the way, so that we are sure to not
686 * fail once we start modifying the data on disk, because JBD has no rollback.
688 static int add_new_gdb(handle_t *handle, struct inode *inode,
691 struct super_block *sb = inode->i_sb;
692 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
693 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
694 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
695 struct buffer_head **o_group_desc, **n_group_desc;
696 struct buffer_head *dind;
697 struct buffer_head *gdb_bh;
699 struct ext4_iloc iloc;
703 if (test_opt(sb, DEBUG))
705 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
709 * If we are not using the primary superblock/GDT copy don't resize,
710 * because the user tools have no way of handling this. Probably a
711 * bad time to do it anyways.
713 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
714 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
715 ext4_warning(sb, "won't resize using backup superblock at %llu",
716 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
720 gdb_bh = sb_bread(sb, gdblock);
724 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
730 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
731 dind = sb_bread(sb, le32_to_cpu(*data));
737 data = (__le32 *)dind->b_data;
738 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
739 ext4_warning(sb, "new group %u GDT block %llu not reserved",
745 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
749 err = ext4_journal_get_write_access(handle, gdb_bh);
753 err = ext4_journal_get_write_access(handle, dind);
755 ext4_std_error(sb, err);
757 /* ext4_reserve_inode_write() gets a reference on the iloc */
758 err = ext4_reserve_inode_write(handle, inode, &iloc);
762 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
763 sizeof(struct buffer_head *),
767 ext4_warning(sb, "not enough memory for %lu groups",
773 * Finally, we have all of the possible failures behind us...
775 * Remove new GDT block from inode double-indirect block and clear out
776 * the new GDT block for use (which also "frees" the backup GDT blocks
777 * from the reserved inode). We don't need to change the bitmaps for
778 * these blocks, because they are marked as in-use from being in the
779 * reserved inode, and will become GDT blocks (primary and backup).
781 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
782 err = ext4_handle_dirty_metadata(handle, NULL, dind);
784 ext4_std_error(sb, err);
787 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
788 ext4_mark_iloc_dirty(handle, inode, &iloc);
789 memset(gdb_bh->b_data, 0, sb->s_blocksize);
790 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
792 ext4_std_error(sb, err);
797 o_group_desc = EXT4_SB(sb)->s_group_desc;
798 memcpy(n_group_desc, o_group_desc,
799 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
800 n_group_desc[gdb_num] = gdb_bh;
801 EXT4_SB(sb)->s_group_desc = n_group_desc;
802 EXT4_SB(sb)->s_gdb_count++;
803 ext4_kvfree(o_group_desc);
805 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
806 err = ext4_handle_dirty_super(handle, sb);
808 ext4_std_error(sb, err);
813 ext4_kvfree(n_group_desc);
814 /* ext4_handle_release_buffer(handle, iloc.bh); */
817 /* ext4_handle_release_buffer(handle, dind); */
819 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
825 ext4_debug("leaving with error %d\n", err);
830 * Called when we are adding a new group which has a backup copy of each of
831 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
832 * We need to add these reserved backup GDT blocks to the resize inode, so
833 * that they are kept for future resizing and not allocated to files.
835 * Each reserved backup GDT block will go into a different indirect block.
836 * The indirect blocks are actually the primary reserved GDT blocks,
837 * so we know in advance what their block numbers are. We only get the
838 * double-indirect block to verify it is pointing to the primary reserved
839 * GDT blocks so we don't overwrite a data block by accident. The reserved
840 * backup GDT blocks are stored in their reserved primary GDT block.
842 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
845 struct super_block *sb = inode->i_sb;
846 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
847 struct buffer_head **primary;
848 struct buffer_head *dind;
849 struct ext4_iloc iloc;
856 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
860 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
861 dind = sb_bread(sb, le32_to_cpu(*data));
867 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
868 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
869 EXT4_ADDR_PER_BLOCK(sb));
870 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
872 /* Get each reserved primary GDT block and verify it holds backups */
873 for (res = 0; res < reserved_gdb; res++, blk++) {
874 if (le32_to_cpu(*data) != blk) {
875 ext4_warning(sb, "reserved block %llu"
876 " not at offset %ld",
878 (long)(data - (__le32 *)dind->b_data));
882 primary[res] = sb_bread(sb, blk);
887 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
889 brelse(primary[res]);
894 data = (__le32 *)dind->b_data;
897 for (i = 0; i < reserved_gdb; i++) {
898 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
901 for (j = 0; j < i; j++)
902 ext4_handle_release_buffer(handle, primary[j]);
908 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
912 * Finally we can add each of the reserved backup GDT blocks from
913 * the new group to its reserved primary GDT block.
915 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
916 for (i = 0; i < reserved_gdb; i++) {
918 data = (__le32 *)primary[i]->b_data;
919 /* printk("reserving backup %lu[%u] = %lu\n",
920 primary[i]->b_blocknr, gdbackups,
921 blk + primary[i]->b_blocknr); */
922 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
923 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
927 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
928 ext4_mark_iloc_dirty(handle, inode, &iloc);
932 brelse(primary[res]);
942 * Update the backup copies of the ext4 metadata. These don't need to be part
943 * of the main resize transaction, because e2fsck will re-write them if there
944 * is a problem (basically only OOM will cause a problem). However, we
945 * _should_ update the backups if possible, in case the primary gets trashed
946 * for some reason and we need to run e2fsck from a backup superblock. The
947 * important part is that the new block and inode counts are in the backup
948 * superblocks, and the location of the new group metadata in the GDT backups.
950 * We do not need take the s_resize_lock for this, because these
951 * blocks are not otherwise touched by the filesystem code when it is
952 * mounted. We don't need to worry about last changing from
953 * sbi->s_groups_count, because the worst that can happen is that we
954 * do not copy the full number of backups at this time. The resize
955 * which changed s_groups_count will backup again.
957 static void update_backups(struct super_block *sb,
958 int blk_off, char *data, int size)
960 struct ext4_sb_info *sbi = EXT4_SB(sb);
961 const ext4_group_t last = sbi->s_groups_count;
962 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
967 int rest = sb->s_blocksize - size;
971 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
972 if (IS_ERR(handle)) {
974 err = PTR_ERR(handle);
978 ext4_superblock_csum_set(sb, (struct ext4_super_block *)data);
980 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
981 struct buffer_head *bh;
983 /* Out of journal space, and can't get more - abort - so sad */
984 if (ext4_handle_valid(handle) &&
985 handle->h_buffer_credits == 0 &&
986 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
987 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
990 bh = sb_getblk(sb, group * bpg + blk_off);
995 ext4_debug("update metadata backup %#04lx\n",
996 (unsigned long)bh->b_blocknr);
997 if ((err = ext4_journal_get_write_access(handle, bh)))
1000 memcpy(bh->b_data, data, size);
1002 memset(bh->b_data + size, 0, rest);
1003 set_buffer_uptodate(bh);
1005 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1007 ext4_std_error(sb, err);
1010 if ((err2 = ext4_journal_stop(handle)) && !err)
1014 * Ugh! Need to have e2fsck write the backup copies. It is too
1015 * late to revert the resize, we shouldn't fail just because of
1016 * the backup copies (they are only needed in case of corruption).
1018 * However, if we got here we have a journal problem too, so we
1019 * can't really start a transaction to mark the superblock.
1020 * Chicken out and just set the flag on the hope it will be written
1021 * to disk, and if not - we will simply wait until next fsck.
1025 ext4_warning(sb, "can't update backup for group %u (err %d), "
1026 "forcing fsck on next reboot", group, err);
1027 sbi->s_mount_state &= ~EXT4_VALID_FS;
1028 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1029 mark_buffer_dirty(sbi->s_sbh);
1034 * ext4_add_new_descs() adds @count group descriptor of groups
1035 * starting at @group
1037 * @handle: journal handle
1039 * @group: the group no. of the first group desc to be added
1040 * @resize_inode: the resize inode
1041 * @count: number of group descriptors to be added
1043 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1044 ext4_group_t group, struct inode *resize_inode,
1047 struct ext4_sb_info *sbi = EXT4_SB(sb);
1048 struct ext4_super_block *es = sbi->s_es;
1049 struct buffer_head *gdb_bh;
1050 int i, gdb_off, gdb_num, err = 0;
1052 for (i = 0; i < count; i++, group++) {
1053 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1054 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1056 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1057 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1060 * We will only either add reserved group blocks to a backup group
1061 * or remove reserved blocks for the first group in a new group block.
1062 * Doing both would be mean more complex code, and sane people don't
1063 * use non-sparse filesystems anymore. This is already checked above.
1066 gdb_bh = sbi->s_group_desc[gdb_num];
1067 err = ext4_journal_get_write_access(handle, gdb_bh);
1069 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1070 err = reserve_backup_gdb(handle, resize_inode, group);
1072 err = add_new_gdb(handle, resize_inode, group);
1079 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1081 struct buffer_head *bh = sb_getblk(sb, block);
1085 if (bitmap_uptodate(bh))
1089 if (bh_submit_read(bh) < 0) {
1099 static int ext4_set_bitmap_checksums(struct super_block *sb,
1101 struct ext4_group_desc *gdp,
1102 struct ext4_new_group_data *group_data)
1104 struct buffer_head *bh;
1106 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1110 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1113 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1114 EXT4_INODES_PER_GROUP(sb) / 8);
1117 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1120 ext4_block_bitmap_csum_set(sb, group, gdp, bh,
1121 EXT4_BLOCKS_PER_GROUP(sb) / 8);
1128 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1130 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1131 struct ext4_new_flex_group_data *flex_gd)
1133 struct ext4_new_group_data *group_data = flex_gd->groups;
1134 struct ext4_group_desc *gdp;
1135 struct ext4_sb_info *sbi = EXT4_SB(sb);
1136 struct buffer_head *gdb_bh;
1138 __u16 *bg_flags = flex_gd->bg_flags;
1139 int i, gdb_off, gdb_num, err = 0;
1142 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1143 group = group_data->group;
1145 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1146 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1149 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1151 gdb_bh = sbi->s_group_desc[gdb_num];
1152 /* Update group descriptor block for new group */
1153 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1154 gdb_off * EXT4_DESC_SIZE(sb));
1156 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1157 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1158 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1159 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1161 ext4_std_error(sb, err);
1165 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1166 ext4_free_group_clusters_set(sb, gdp,
1167 EXT4_B2C(sbi, group_data->free_blocks_count));
1168 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1169 gdp->bg_flags = cpu_to_le16(*bg_flags);
1170 ext4_group_desc_csum_set(sb, group, gdp);
1172 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1173 if (unlikely(err)) {
1174 ext4_std_error(sb, err);
1179 * We can allocate memory for mb_alloc based on the new group
1182 err = ext4_mb_add_groupinfo(sb, group, gdp);
1190 * ext4_update_super() updates the super block so that the newly added
1191 * groups can be seen by the filesystem.
1194 * @flex_gd: new added groups
1196 static void ext4_update_super(struct super_block *sb,
1197 struct ext4_new_flex_group_data *flex_gd)
1199 ext4_fsblk_t blocks_count = 0;
1200 ext4_fsblk_t free_blocks = 0;
1201 ext4_fsblk_t reserved_blocks = 0;
1202 struct ext4_new_group_data *group_data = flex_gd->groups;
1203 struct ext4_sb_info *sbi = EXT4_SB(sb);
1204 struct ext4_super_block *es = sbi->s_es;
1207 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1209 * Make the new blocks and inodes valid next. We do this before
1210 * increasing the group count so that once the group is enabled,
1211 * all of its blocks and inodes are already valid.
1213 * We always allocate group-by-group, then block-by-block or
1214 * inode-by-inode within a group, so enabling these
1215 * blocks/inodes before the group is live won't actually let us
1216 * allocate the new space yet.
1218 for (i = 0; i < flex_gd->count; i++) {
1219 blocks_count += group_data[i].blocks_count;
1220 free_blocks += group_data[i].free_blocks_count;
1223 reserved_blocks = ext4_r_blocks_count(es) * 100;
1224 do_div(reserved_blocks, ext4_blocks_count(es));
1225 reserved_blocks *= blocks_count;
1226 do_div(reserved_blocks, 100);
1228 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1229 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1230 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1232 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1236 * We need to protect s_groups_count against other CPUs seeing
1237 * inconsistent state in the superblock.
1239 * The precise rules we use are:
1241 * * Writers must perform a smp_wmb() after updating all
1242 * dependent data and before modifying the groups count
1244 * * Readers must perform an smp_rmb() after reading the groups
1245 * count and before reading any dependent data.
1247 * NB. These rules can be relaxed when checking the group count
1248 * while freeing data, as we can only allocate from a block
1249 * group after serialising against the group count, and we can
1250 * only then free after serialising in turn against that
1255 /* Update the global fs size fields */
1256 sbi->s_groups_count += flex_gd->count;
1258 /* Update the reserved block counts only once the new group is
1260 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1263 /* Update the free space counts */
1264 percpu_counter_add(&sbi->s_freeclusters_counter,
1265 EXT4_B2C(sbi, free_blocks));
1266 percpu_counter_add(&sbi->s_freeinodes_counter,
1267 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1269 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1270 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1271 sbi->s_log_groups_per_flex) {
1272 ext4_group_t flex_group;
1273 flex_group = ext4_flex_group(sbi, group_data[0].group);
1274 atomic_add(EXT4_B2C(sbi, free_blocks),
1275 &sbi->s_flex_groups[flex_group].free_clusters);
1276 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1277 &sbi->s_flex_groups[flex_group].free_inodes);
1281 * Update the fs overhead information
1283 ext4_calculate_overhead(sb);
1285 if (test_opt(sb, DEBUG))
1286 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1287 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1288 blocks_count, free_blocks, reserved_blocks);
1291 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1292 * _before_ we start modifying the filesystem, because we cannot abort the
1293 * transaction and not have it write the data to disk.
1295 static int ext4_flex_group_add(struct super_block *sb,
1296 struct inode *resize_inode,
1297 struct ext4_new_flex_group_data *flex_gd)
1299 struct ext4_sb_info *sbi = EXT4_SB(sb);
1300 struct ext4_super_block *es = sbi->s_es;
1301 ext4_fsblk_t o_blocks_count;
1305 unsigned reserved_gdb;
1306 int err = 0, err2 = 0, credit;
1308 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1310 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1311 o_blocks_count = ext4_blocks_count(es);
1312 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1315 err = setup_new_flex_group_blocks(sb, flex_gd);
1319 * We will always be modifying at least the superblock and GDT
1320 * block. If we are adding a group past the last current GDT block,
1321 * we will also modify the inode and the dindirect block. If we
1322 * are adding a group with superblock/GDT backups we will also
1323 * modify each of the reserved GDT dindirect blocks.
1325 credit = flex_gd->count * 4 + reserved_gdb;
1326 handle = ext4_journal_start_sb(sb, credit);
1327 if (IS_ERR(handle)) {
1328 err = PTR_ERR(handle);
1332 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1336 group = flex_gd->groups[0].group;
1337 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1338 err = ext4_add_new_descs(handle, sb, group,
1339 resize_inode, flex_gd->count);
1343 err = ext4_setup_new_descs(handle, sb, flex_gd);
1347 ext4_update_super(sb, flex_gd);
1349 err = ext4_handle_dirty_super(handle, sb);
1352 err2 = ext4_journal_stop(handle);
1358 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1359 sizeof(struct ext4_super_block));
1360 for (i = 0; i < flex_gd->count; i++, group++) {
1361 struct buffer_head *gdb_bh;
1363 gdb_num = group / EXT4_BLOCKS_PER_GROUP(sb);
1364 gdb_bh = sbi->s_group_desc[gdb_num];
1365 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1373 static int ext4_setup_next_flex_gd(struct super_block *sb,
1374 struct ext4_new_flex_group_data *flex_gd,
1375 ext4_fsblk_t n_blocks_count,
1376 unsigned long flexbg_size)
1378 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1379 struct ext4_new_group_data *group_data = flex_gd->groups;
1380 ext4_fsblk_t o_blocks_count;
1381 ext4_group_t n_group;
1383 ext4_group_t last_group;
1385 ext4_grpblk_t blocks_per_group;
1388 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1390 o_blocks_count = ext4_blocks_count(es);
1392 if (o_blocks_count == n_blocks_count)
1395 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1397 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1399 last_group = group | (flexbg_size - 1);
1400 if (last_group > n_group)
1401 last_group = n_group;
1403 flex_gd->count = last_group - group + 1;
1405 for (i = 0; i < flex_gd->count; i++) {
1408 group_data[i].group = group + i;
1409 group_data[i].blocks_count = blocks_per_group;
1410 overhead = ext4_bg_has_super(sb, group + i) ?
1411 (1 + ext4_bg_num_gdb(sb, group + i) +
1412 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
1413 group_data[i].free_blocks_count = blocks_per_group - overhead;
1414 if (ext4_has_group_desc_csum(sb))
1415 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1416 EXT4_BG_INODE_UNINIT;
1418 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1421 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1422 /* We need to initialize block bitmap of last group. */
1423 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1425 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1426 group_data[i - 1].blocks_count = last + 1;
1427 group_data[i - 1].free_blocks_count -= blocks_per_group-
1434 /* Add group descriptor data to an existing or new group descriptor block.
1435 * Ensure we handle all possible error conditions _before_ we start modifying
1436 * the filesystem, because we cannot abort the transaction and not have it
1437 * write the data to disk.
1439 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1440 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1442 * We only need to hold the superblock lock while we are actually adding
1443 * in the new group's counts to the superblock. Prior to that we have
1444 * not really "added" the group at all. We re-check that we are still
1445 * adding in the last group in case things have changed since verifying.
1447 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1449 struct ext4_new_flex_group_data flex_gd;
1450 struct ext4_sb_info *sbi = EXT4_SB(sb);
1451 struct ext4_super_block *es = sbi->s_es;
1452 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1453 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1454 struct inode *inode = NULL;
1455 int gdb_off, gdb_num;
1459 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1460 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1462 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1463 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1464 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1468 if (ext4_blocks_count(es) + input->blocks_count <
1469 ext4_blocks_count(es)) {
1470 ext4_warning(sb, "blocks_count overflow");
1474 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1475 le32_to_cpu(es->s_inodes_count)) {
1476 ext4_warning(sb, "inodes_count overflow");
1480 if (reserved_gdb || gdb_off == 0) {
1481 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1482 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1483 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1485 "No reserved GDT blocks, can't resize");
1488 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1489 if (IS_ERR(inode)) {
1490 ext4_warning(sb, "Error opening resize inode");
1491 return PTR_ERR(inode);
1496 err = verify_group_input(sb, input);
1501 flex_gd.groups = input;
1502 flex_gd.bg_flags = &bg_flags;
1503 err = ext4_flex_group_add(sb, inode, &flex_gd);
1507 } /* ext4_group_add */
1510 * extend a group without checking assuming that checking has been done.
1512 static int ext4_group_extend_no_check(struct super_block *sb,
1513 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1515 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1519 /* We will update the superblock, one block bitmap, and
1520 * one group descriptor via ext4_group_add_blocks().
1522 handle = ext4_journal_start_sb(sb, 3);
1523 if (IS_ERR(handle)) {
1524 err = PTR_ERR(handle);
1525 ext4_warning(sb, "error %d on journal start", err);
1529 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1531 ext4_warning(sb, "error %d on journal write access", err);
1535 ext4_blocks_count_set(es, o_blocks_count + add);
1536 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1537 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1538 o_blocks_count + add);
1539 /* We add the blocks to the bitmap and set the group need init bit */
1540 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1543 ext4_handle_dirty_super(handle, sb);
1544 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1545 o_blocks_count + add);
1547 err2 = ext4_journal_stop(handle);
1552 if (test_opt(sb, DEBUG))
1553 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1554 "blocks\n", ext4_blocks_count(es));
1555 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1556 sizeof(struct ext4_super_block));
1562 * Extend the filesystem to the new number of blocks specified. This entry
1563 * point is only used to extend the current filesystem to the end of the last
1564 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1565 * for emergencies (because it has no dependencies on reserved blocks).
1567 * If we _really_ wanted, we could use default values to call ext4_group_add()
1568 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1569 * GDT blocks are reserved to grow to the desired size.
1571 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1572 ext4_fsblk_t n_blocks_count)
1574 ext4_fsblk_t o_blocks_count;
1577 struct buffer_head *bh;
1581 o_blocks_count = ext4_blocks_count(es);
1583 if (test_opt(sb, DEBUG))
1584 ext4_msg(sb, KERN_DEBUG,
1585 "extending last group from %llu to %llu blocks",
1586 o_blocks_count, n_blocks_count);
1588 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1591 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1592 ext4_msg(sb, KERN_ERR,
1593 "filesystem too large to resize to %llu blocks safely",
1595 if (sizeof(sector_t) < 8)
1596 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1600 if (n_blocks_count < o_blocks_count) {
1601 ext4_warning(sb, "can't shrink FS - resize aborted");
1605 /* Handle the remaining blocks in the last group only. */
1606 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1609 ext4_warning(sb, "need to use ext2online to resize further");
1613 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1615 if (o_blocks_count + add < o_blocks_count) {
1616 ext4_warning(sb, "blocks_count overflow");
1620 if (o_blocks_count + add > n_blocks_count)
1621 add = n_blocks_count - o_blocks_count;
1623 if (o_blocks_count + add < n_blocks_count)
1624 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1625 o_blocks_count + add, add);
1627 /* See if the device is actually as big as what was requested */
1628 bh = sb_bread(sb, o_blocks_count + add - 1);
1630 ext4_warning(sb, "can't read last block, resize aborted");
1635 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1637 } /* ext4_group_extend */
1640 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1642 * @sb: super block of the fs to be resized
1643 * @n_blocks_count: the number of blocks resides in the resized fs
1645 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1647 struct ext4_new_flex_group_data *flex_gd = NULL;
1648 struct ext4_sb_info *sbi = EXT4_SB(sb);
1649 struct ext4_super_block *es = sbi->s_es;
1650 struct buffer_head *bh;
1651 struct inode *resize_inode;
1652 ext4_fsblk_t o_blocks_count;
1653 ext4_group_t o_group;
1654 ext4_group_t n_group;
1655 ext4_grpblk_t offset, add;
1656 unsigned long n_desc_blocks;
1657 unsigned long o_desc_blocks;
1658 unsigned long desc_blocks;
1659 int err = 0, flexbg_size = 1;
1661 o_blocks_count = ext4_blocks_count(es);
1663 if (test_opt(sb, DEBUG))
1664 ext4_msg(sb, KERN_DEBUG, "resizing filesystem from %llu "
1665 "to %llu blocks", o_blocks_count, n_blocks_count);
1667 if (n_blocks_count < o_blocks_count) {
1668 /* On-line shrinking not supported */
1669 ext4_warning(sb, "can't shrink FS - resize aborted");
1673 if (n_blocks_count == o_blocks_count)
1674 /* Nothing need to do */
1677 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
1678 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1680 n_desc_blocks = (n_group + EXT4_DESC_PER_BLOCK(sb)) /
1681 EXT4_DESC_PER_BLOCK(sb);
1682 o_desc_blocks = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1683 EXT4_DESC_PER_BLOCK(sb);
1684 desc_blocks = n_desc_blocks - o_desc_blocks;
1687 (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE) ||
1688 le16_to_cpu(es->s_reserved_gdt_blocks) < desc_blocks)) {
1689 ext4_warning(sb, "No reserved GDT blocks, can't resize");
1693 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1694 if (IS_ERR(resize_inode)) {
1695 ext4_warning(sb, "Error opening resize inode");
1696 return PTR_ERR(resize_inode);
1699 /* See if the device is actually as big as what was requested */
1700 bh = sb_bread(sb, n_blocks_count - 1);
1702 ext4_warning(sb, "can't read last block, resize aborted");
1707 /* extend the last group */
1708 if (n_group == o_group)
1709 add = n_blocks_count - o_blocks_count;
1711 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1713 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1718 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1719 es->s_log_groups_per_flex)
1720 flexbg_size = 1 << es->s_log_groups_per_flex;
1722 if (ext4_blocks_count(es) == n_blocks_count)
1725 flex_gd = alloc_flex_gd(flexbg_size);
1726 if (flex_gd == NULL) {
1731 /* Add flex groups. Note that a regular group is a
1732 * flex group with 1 group.
1734 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1736 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1738 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1745 free_flex_gd(flex_gd);
1748 if (test_opt(sb, DEBUG))
1749 ext4_msg(sb, KERN_DEBUG, "resized filesystem from %llu "
1750 "upto %llu blocks", o_blocks_count, n_blocks_count);