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 (!(bh = sb_bread(sb, end - 1)))
131 ext4_warning(sb, "Cannot read last block (%llu)",
133 else if (outside(input->block_bitmap, start, end))
134 ext4_warning(sb, "Block bitmap not in group (block %llu)",
135 (unsigned long long)input->block_bitmap);
136 else if (outside(input->inode_bitmap, start, end))
137 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
138 (unsigned long long)input->inode_bitmap);
139 else if (outside(input->inode_table, start, end) ||
140 outside(itend - 1, start, end))
141 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
142 (unsigned long long)input->inode_table, itend - 1);
143 else if (input->inode_bitmap == input->block_bitmap)
144 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
145 (unsigned long long)input->block_bitmap);
146 else if (inside(input->block_bitmap, input->inode_table, itend))
147 ext4_warning(sb, "Block bitmap (%llu) in inode table "
149 (unsigned long long)input->block_bitmap,
150 (unsigned long long)input->inode_table, itend - 1);
151 else if (inside(input->inode_bitmap, input->inode_table, itend))
152 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
154 (unsigned long long)input->inode_bitmap,
155 (unsigned long long)input->inode_table, itend - 1);
156 else if (inside(input->block_bitmap, start, metaend))
157 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
158 (unsigned long long)input->block_bitmap,
160 else if (inside(input->inode_bitmap, start, metaend))
161 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
162 (unsigned long long)input->inode_bitmap,
164 else if (inside(input->inode_table, start, metaend) ||
165 inside(itend - 1, start, metaend))
166 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
168 (unsigned long long)input->inode_table,
169 itend - 1, start, metaend - 1);
178 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
181 struct ext4_new_flex_group_data {
182 struct ext4_new_group_data *groups; /* new_group_data for groups
184 __u16 *bg_flags; /* block group flags of groups
186 ext4_group_t count; /* number of groups in @groups
191 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
194 * Returns NULL on failure otherwise address of the allocated structure.
196 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
198 struct ext4_new_flex_group_data *flex_gd;
200 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
204 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
206 flex_gd->count = flexbg_size;
208 flex_gd->groups = kmalloc_array(flexbg_size,
209 sizeof(struct ext4_new_group_data),
211 if (flex_gd->groups == NULL)
214 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
216 if (flex_gd->bg_flags == NULL)
222 kfree(flex_gd->groups);
229 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
231 kfree(flex_gd->bg_flags);
232 kfree(flex_gd->groups);
237 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
238 * and inode tables for a flex group.
240 * This function is used by 64bit-resize. Note that this function allocates
241 * group tables from the 1st group of groups contained by @flexgd, which may
242 * be a partial of a flex group.
244 * @sb: super block of fs to which the groups belongs
246 * Returns 0 on a successful allocation of the metadata blocks in the
249 static int ext4_alloc_group_tables(struct super_block *sb,
250 struct ext4_new_flex_group_data *flex_gd,
253 struct ext4_new_group_data *group_data = flex_gd->groups;
254 ext4_fsblk_t start_blk;
255 ext4_fsblk_t last_blk;
256 ext4_group_t src_group;
257 ext4_group_t bb_index = 0;
258 ext4_group_t ib_index = 0;
259 ext4_group_t it_index = 0;
261 ext4_group_t last_group;
263 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
266 BUG_ON(flex_gd->count == 0 || group_data == NULL);
268 src_group = group_data[0].group;
269 last_group = src_group + flex_gd->count - 1;
271 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
272 (last_group & ~(flexbg_size - 1))));
274 group = group_data[0].group;
275 if (src_group >= group_data[0].group + flex_gd->count)
277 start_blk = ext4_group_first_block_no(sb, src_group);
278 last_blk = start_blk + group_data[src_group - group].blocks_count;
280 overhead = ext4_group_overhead_blocks(sb, src_group);
282 start_blk += overhead;
284 /* We collect contiguous blocks as much as possible. */
286 for (; src_group <= last_group; src_group++) {
287 overhead = ext4_group_overhead_blocks(sb, src_group);
289 last_blk += group_data[src_group - group].blocks_count;
294 /* Allocate block bitmaps */
295 for (; bb_index < flex_gd->count; bb_index++) {
296 if (start_blk >= last_blk)
298 group_data[bb_index].block_bitmap = start_blk++;
299 group = ext4_get_group_number(sb, start_blk - 1);
300 group -= group_data[0].group;
301 group_data[group].mdata_blocks++;
302 flex_gd->bg_flags[group] &= uninit_mask;
305 /* Allocate inode bitmaps */
306 for (; ib_index < flex_gd->count; ib_index++) {
307 if (start_blk >= last_blk)
309 group_data[ib_index].inode_bitmap = start_blk++;
310 group = ext4_get_group_number(sb, start_blk - 1);
311 group -= group_data[0].group;
312 group_data[group].mdata_blocks++;
313 flex_gd->bg_flags[group] &= uninit_mask;
316 /* Allocate inode tables */
317 for (; it_index < flex_gd->count; it_index++) {
318 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
319 ext4_fsblk_t next_group_start;
321 if (start_blk + itb > last_blk)
323 group_data[it_index].inode_table = start_blk;
324 group = ext4_get_group_number(sb, start_blk);
325 next_group_start = ext4_group_first_block_no(sb, group + 1);
326 group -= group_data[0].group;
328 if (start_blk + itb > next_group_start) {
329 flex_gd->bg_flags[group + 1] &= uninit_mask;
330 overhead = start_blk + itb - next_group_start;
331 group_data[group + 1].mdata_blocks += overhead;
335 group_data[group].mdata_blocks += itb;
336 flex_gd->bg_flags[group] &= uninit_mask;
337 start_blk += EXT4_SB(sb)->s_itb_per_group;
340 /* Update free clusters count to exclude metadata blocks */
341 for (i = 0; i < flex_gd->count; i++) {
342 group_data[i].free_clusters_count -=
343 EXT4_NUM_B2C(EXT4_SB(sb),
344 group_data[i].mdata_blocks);
347 if (test_opt(sb, DEBUG)) {
349 group = group_data[0].group;
351 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
352 "%d groups, flexbg size is %d:\n", flex_gd->count,
355 for (i = 0; i < flex_gd->count; i++) {
357 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
358 ext4_bg_has_super(sb, group + i) ? "normal" :
359 "no-super", group + i,
360 group_data[i].blocks_count,
361 group_data[i].free_clusters_count,
362 group_data[i].mdata_blocks);
368 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
371 struct buffer_head *bh;
374 bh = sb_getblk(sb, blk);
376 return ERR_PTR(-ENOMEM);
377 BUFFER_TRACE(bh, "get_write_access");
378 if ((err = ext4_journal_get_write_access(handle, bh))) {
382 memset(bh->b_data, 0, sb->s_blocksize);
383 set_buffer_uptodate(bh);
390 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
391 * If that fails, restart the transaction & regain write access for the
392 * buffer head which is used for block_bitmap modifications.
394 static int extend_or_restart_transaction(handle_t *handle, int thresh)
398 if (ext4_handle_has_enough_credits(handle, thresh))
401 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
405 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
414 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
416 * Helper function for ext4_setup_new_group_blocks() which set .
419 * @handle: journal handle
420 * @flex_gd: flex group data
422 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
423 struct ext4_new_flex_group_data *flex_gd,
424 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
426 struct ext4_sb_info *sbi = EXT4_SB(sb);
427 ext4_group_t count = last_cluster - first_cluster + 1;
430 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
432 for (count2 = count; count > 0;
433 count -= count2, first_cluster += count2) {
435 struct buffer_head *bh;
439 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
440 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
441 group -= flex_gd->groups[0].group;
443 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
447 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
448 BUG_ON(flex_gd->count > 1);
452 err = extend_or_restart_transaction(handle, 1);
456 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
460 BUFFER_TRACE(bh, "get_write_access");
461 err = ext4_journal_get_write_access(handle, bh);
464 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
465 first_cluster, first_cluster - start, count2);
466 ext4_set_bits(bh->b_data, first_cluster - start, count2);
468 err = ext4_handle_dirty_metadata(handle, NULL, bh);
478 * Set up the block and inode bitmaps, and the inode table for the new groups.
479 * This doesn't need to be part of the main transaction, since we are only
480 * changing blocks outside the actual filesystem. We still do journaling to
481 * ensure the recovery is correct in case of a failure just after resize.
482 * If any part of this fails, we simply abort the resize.
484 * setup_new_flex_group_blocks handles a flex group as follow:
485 * 1. copy super block and GDT, and initialize group tables if necessary.
486 * In this step, we only set bits in blocks bitmaps for blocks taken by
487 * super block and GDT.
488 * 2. allocate group tables in block bitmaps, that is, set bits in block
489 * bitmap for blocks taken by group tables.
491 static int setup_new_flex_group_blocks(struct super_block *sb,
492 struct ext4_new_flex_group_data *flex_gd)
494 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
497 struct ext4_sb_info *sbi = EXT4_SB(sb);
498 struct ext4_super_block *es = sbi->s_es;
499 struct ext4_new_group_data *group_data = flex_gd->groups;
500 __u16 *bg_flags = flex_gd->bg_flags;
502 ext4_group_t group, count;
503 struct buffer_head *bh = NULL;
504 int reserved_gdb, i, j, err = 0, err2;
507 BUG_ON(!flex_gd->count || !group_data ||
508 group_data[0].group != sbi->s_groups_count);
510 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
511 meta_bg = ext4_has_feature_meta_bg(sb);
513 /* This transaction may be extended/restarted along the way */
514 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
516 return PTR_ERR(handle);
518 group = group_data[0].group;
519 for (i = 0; i < flex_gd->count; i++, group++) {
520 unsigned long gdblocks;
521 ext4_grpblk_t overhead;
523 gdblocks = ext4_bg_num_gdb(sb, group);
524 start = ext4_group_first_block_no(sb, group);
526 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
530 ext4_group_t first_group;
531 first_group = ext4_meta_bg_first_group(sb, group);
532 if (first_group != group + 1 &&
533 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
537 block = start + ext4_bg_has_super(sb, group);
538 /* Copy all of the GDT blocks into the backup in this group */
539 for (j = 0; j < gdblocks; j++, block++) {
540 struct buffer_head *gdb;
542 ext4_debug("update backup group %#04llx\n", block);
543 err = extend_or_restart_transaction(handle, 1);
547 gdb = sb_getblk(sb, block);
548 if (unlikely(!gdb)) {
553 BUFFER_TRACE(gdb, "get_write_access");
554 err = ext4_journal_get_write_access(handle, gdb);
559 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
561 set_buffer_uptodate(gdb);
563 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
571 /* Zero out all of the reserved backup group descriptor
574 if (ext4_bg_has_super(sb, group)) {
575 err = sb_issue_zeroout(sb, gdblocks + start + 1,
576 reserved_gdb, GFP_NOFS);
582 /* Initialize group tables of the grop @group */
583 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
586 /* Zero out all of the inode table blocks */
587 block = group_data[i].inode_table;
588 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
589 block, sbi->s_itb_per_group);
590 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
596 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
599 /* Initialize block bitmap of the @group */
600 block = group_data[i].block_bitmap;
601 err = extend_or_restart_transaction(handle, 1);
605 bh = bclean(handle, sb, block);
610 overhead = ext4_group_overhead_blocks(sb, group);
612 ext4_debug("mark backup superblock %#04llx (+0)\n",
614 ext4_set_bits(bh->b_data, 0,
615 EXT4_NUM_B2C(sbi, overhead));
617 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
618 sb->s_blocksize * 8, bh->b_data);
619 err = ext4_handle_dirty_metadata(handle, NULL, bh);
625 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
628 /* Initialize inode bitmap of the @group */
629 block = group_data[i].inode_bitmap;
630 err = extend_or_restart_transaction(handle, 1);
633 /* Mark unused entries in inode bitmap used */
634 bh = bclean(handle, sb, block);
640 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
641 sb->s_blocksize * 8, bh->b_data);
642 err = ext4_handle_dirty_metadata(handle, NULL, bh);
648 /* Mark group tables in block bitmap */
649 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
650 count = group_table_count[j];
651 start = (&group_data[0].block_bitmap)[j];
653 for (i = 1; i < flex_gd->count; i++) {
654 block += group_table_count[j];
655 if (block == (&group_data[i].block_bitmap)[j]) {
656 count += group_table_count[j];
659 err = set_flexbg_block_bitmap(sb, handle,
661 EXT4_B2C(sbi, start),
667 count = group_table_count[j];
668 start = (&group_data[i].block_bitmap)[j];
673 err = set_flexbg_block_bitmap(sb, handle,
675 EXT4_B2C(sbi, start),
685 err2 = ext4_journal_stop(handle);
693 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
694 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
695 * calling this for the first time. In a sparse filesystem it will be the
696 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
697 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
699 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
700 unsigned *five, unsigned *seven)
702 unsigned *min = three;
706 if (!ext4_has_feature_sparse_super(sb)) {
728 * Check that all of the backup GDT blocks are held in the primary GDT block.
729 * It is assumed that they are stored in group order. Returns the number of
730 * groups in current filesystem that have BACKUPS, or -ve error code.
732 static int verify_reserved_gdb(struct super_block *sb,
734 struct buffer_head *primary)
736 const ext4_fsblk_t blk = primary->b_blocknr;
741 __le32 *p = (__le32 *)primary->b_data;
744 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
745 if (le32_to_cpu(*p++) !=
746 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
747 ext4_warning(sb, "reserved GDT %llu"
748 " missing grp %d (%llu)",
751 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
755 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
763 * Called when we need to bring a reserved group descriptor table block into
764 * use from the resize inode. The primary copy of the new GDT block currently
765 * is an indirect block (under the double indirect block in the resize inode).
766 * The new backup GDT blocks will be stored as leaf blocks in this indirect
767 * block, in group order. Even though we know all the block numbers we need,
768 * we check to ensure that the resize inode has actually reserved these blocks.
770 * Don't need to update the block bitmaps because the blocks are still in use.
772 * We get all of the error cases out of the way, so that we are sure to not
773 * fail once we start modifying the data on disk, because JBD has no rollback.
775 static int add_new_gdb(handle_t *handle, struct inode *inode,
778 struct super_block *sb = inode->i_sb;
779 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
780 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
781 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
782 struct buffer_head **o_group_desc, **n_group_desc;
783 struct buffer_head *dind;
784 struct buffer_head *gdb_bh;
786 struct ext4_iloc iloc;
790 if (test_opt(sb, DEBUG))
792 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
795 gdb_bh = sb_bread(sb, gdblock);
799 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
805 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
806 dind = sb_bread(sb, le32_to_cpu(*data));
812 data = (__le32 *)dind->b_data;
813 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
814 ext4_warning(sb, "new group %u GDT block %llu not reserved",
820 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
821 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
825 BUFFER_TRACE(gdb_bh, "get_write_access");
826 err = ext4_journal_get_write_access(handle, gdb_bh);
830 BUFFER_TRACE(dind, "get_write_access");
831 err = ext4_journal_get_write_access(handle, dind);
833 ext4_std_error(sb, err);
835 /* ext4_reserve_inode_write() gets a reference on the iloc */
836 err = ext4_reserve_inode_write(handle, inode, &iloc);
840 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
841 sizeof(struct buffer_head *),
845 ext4_warning(sb, "not enough memory for %lu groups",
851 * Finally, we have all of the possible failures behind us...
853 * Remove new GDT block from inode double-indirect block and clear out
854 * the new GDT block for use (which also "frees" the backup GDT blocks
855 * from the reserved inode). We don't need to change the bitmaps for
856 * these blocks, because they are marked as in-use from being in the
857 * reserved inode, and will become GDT blocks (primary and backup).
859 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
860 err = ext4_handle_dirty_metadata(handle, NULL, dind);
862 ext4_std_error(sb, err);
865 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
866 (9 - EXT4_SB(sb)->s_cluster_bits);
867 ext4_mark_iloc_dirty(handle, inode, &iloc);
868 memset(gdb_bh->b_data, 0, sb->s_blocksize);
869 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
871 ext4_std_error(sb, err);
876 o_group_desc = EXT4_SB(sb)->s_group_desc;
877 memcpy(n_group_desc, o_group_desc,
878 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
879 n_group_desc[gdb_num] = gdb_bh;
880 EXT4_SB(sb)->s_group_desc = n_group_desc;
881 EXT4_SB(sb)->s_gdb_count++;
882 kvfree(o_group_desc);
884 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
885 err = ext4_handle_dirty_super(handle, sb);
887 ext4_std_error(sb, err);
892 kvfree(n_group_desc);
899 ext4_debug("leaving with error %d\n", err);
904 * add_new_gdb_meta_bg is the sister of add_new_gdb.
906 static int add_new_gdb_meta_bg(struct super_block *sb,
907 handle_t *handle, ext4_group_t group) {
908 ext4_fsblk_t gdblock;
909 struct buffer_head *gdb_bh;
910 struct buffer_head **o_group_desc, **n_group_desc;
911 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
914 gdblock = ext4_meta_bg_first_block_no(sb, group) +
915 ext4_bg_has_super(sb, group);
916 gdb_bh = sb_bread(sb, gdblock);
919 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
920 sizeof(struct buffer_head *),
924 ext4_warning(sb, "not enough memory for %lu groups",
929 o_group_desc = EXT4_SB(sb)->s_group_desc;
930 memcpy(n_group_desc, o_group_desc,
931 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
932 n_group_desc[gdb_num] = gdb_bh;
933 EXT4_SB(sb)->s_group_desc = n_group_desc;
934 EXT4_SB(sb)->s_gdb_count++;
935 kvfree(o_group_desc);
936 BUFFER_TRACE(gdb_bh, "get_write_access");
937 err = ext4_journal_get_write_access(handle, gdb_bh);
944 * Called when we are adding a new group which has a backup copy of each of
945 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
946 * We need to add these reserved backup GDT blocks to the resize inode, so
947 * that they are kept for future resizing and not allocated to files.
949 * Each reserved backup GDT block will go into a different indirect block.
950 * The indirect blocks are actually the primary reserved GDT blocks,
951 * so we know in advance what their block numbers are. We only get the
952 * double-indirect block to verify it is pointing to the primary reserved
953 * GDT blocks so we don't overwrite a data block by accident. The reserved
954 * backup GDT blocks are stored in their reserved primary GDT block.
956 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
959 struct super_block *sb = inode->i_sb;
960 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
961 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
962 struct buffer_head **primary;
963 struct buffer_head *dind;
964 struct ext4_iloc iloc;
971 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
975 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
976 dind = sb_bread(sb, le32_to_cpu(*data));
982 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
983 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
984 EXT4_ADDR_PER_BLOCK(sb));
985 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
987 /* Get each reserved primary GDT block and verify it holds backups */
988 for (res = 0; res < reserved_gdb; res++, blk++) {
989 if (le32_to_cpu(*data) != blk) {
990 ext4_warning(sb, "reserved block %llu"
991 " not at offset %ld",
993 (long)(data - (__le32 *)dind->b_data));
997 primary[res] = sb_bread(sb, blk);
1002 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1003 if (gdbackups < 0) {
1004 brelse(primary[res]);
1009 data = (__le32 *)dind->b_data;
1012 for (i = 0; i < reserved_gdb; i++) {
1013 BUFFER_TRACE(primary[i], "get_write_access");
1014 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1018 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1022 * Finally we can add each of the reserved backup GDT blocks from
1023 * the new group to its reserved primary GDT block.
1025 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1026 for (i = 0; i < reserved_gdb; i++) {
1028 data = (__le32 *)primary[i]->b_data;
1029 /* printk("reserving backup %lu[%u] = %lu\n",
1030 primary[i]->b_blocknr, gdbackups,
1031 blk + primary[i]->b_blocknr); */
1032 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1033 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1038 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1039 ext4_mark_iloc_dirty(handle, inode, &iloc);
1043 brelse(primary[res]);
1053 * Update the backup copies of the ext4 metadata. These don't need to be part
1054 * of the main resize transaction, because e2fsck will re-write them if there
1055 * is a problem (basically only OOM will cause a problem). However, we
1056 * _should_ update the backups if possible, in case the primary gets trashed
1057 * for some reason and we need to run e2fsck from a backup superblock. The
1058 * important part is that the new block and inode counts are in the backup
1059 * superblocks, and the location of the new group metadata in the GDT backups.
1061 * We do not need take the s_resize_lock for this, because these
1062 * blocks are not otherwise touched by the filesystem code when it is
1063 * mounted. We don't need to worry about last changing from
1064 * sbi->s_groups_count, because the worst that can happen is that we
1065 * do not copy the full number of backups at this time. The resize
1066 * which changed s_groups_count will backup again.
1068 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1069 int size, int meta_bg)
1071 struct ext4_sb_info *sbi = EXT4_SB(sb);
1073 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1077 ext4_group_t group = 0;
1078 int rest = sb->s_blocksize - size;
1082 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1083 if (IS_ERR(handle)) {
1085 err = PTR_ERR(handle);
1090 group = ext4_list_backups(sb, &three, &five, &seven);
1091 last = sbi->s_groups_count;
1093 group = ext4_get_group_number(sb, blk_off) + 1;
1094 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1097 while (group < sbi->s_groups_count) {
1098 struct buffer_head *bh;
1099 ext4_fsblk_t backup_block;
1101 /* Out of journal space, and can't get more - abort - so sad */
1102 if (ext4_handle_valid(handle) &&
1103 handle->h_buffer_credits == 0 &&
1104 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1105 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1109 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1111 backup_block = (ext4_group_first_block_no(sb, group) +
1112 ext4_bg_has_super(sb, group));
1114 bh = sb_getblk(sb, backup_block);
1115 if (unlikely(!bh)) {
1119 ext4_debug("update metadata backup %llu(+%llu)\n",
1120 backup_block, backup_block -
1121 ext4_group_first_block_no(sb, group));
1122 BUFFER_TRACE(bh, "get_write_access");
1123 if ((err = ext4_journal_get_write_access(handle, bh)))
1126 memcpy(bh->b_data, data, size);
1128 memset(bh->b_data + size, 0, rest);
1129 set_buffer_uptodate(bh);
1131 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1133 ext4_std_error(sb, err);
1137 group = ext4_list_backups(sb, &three, &five, &seven);
1138 else if (group == last)
1143 if ((err2 = ext4_journal_stop(handle)) && !err)
1147 * Ugh! Need to have e2fsck write the backup copies. It is too
1148 * late to revert the resize, we shouldn't fail just because of
1149 * the backup copies (they are only needed in case of corruption).
1151 * However, if we got here we have a journal problem too, so we
1152 * can't really start a transaction to mark the superblock.
1153 * Chicken out and just set the flag on the hope it will be written
1154 * to disk, and if not - we will simply wait until next fsck.
1158 ext4_warning(sb, "can't update backup for group %u (err %d), "
1159 "forcing fsck on next reboot", group, err);
1160 sbi->s_mount_state &= ~EXT4_VALID_FS;
1161 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1162 mark_buffer_dirty(sbi->s_sbh);
1167 * ext4_add_new_descs() adds @count group descriptor of groups
1168 * starting at @group
1170 * @handle: journal handle
1172 * @group: the group no. of the first group desc to be added
1173 * @resize_inode: the resize inode
1174 * @count: number of group descriptors to be added
1176 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1177 ext4_group_t group, struct inode *resize_inode,
1180 struct ext4_sb_info *sbi = EXT4_SB(sb);
1181 struct ext4_super_block *es = sbi->s_es;
1182 struct buffer_head *gdb_bh;
1183 int i, gdb_off, gdb_num, err = 0;
1186 meta_bg = ext4_has_feature_meta_bg(sb);
1187 for (i = 0; i < count; i++, group++) {
1188 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1189 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1191 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1192 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1195 * We will only either add reserved group blocks to a backup group
1196 * or remove reserved blocks for the first group in a new group block.
1197 * Doing both would be mean more complex code, and sane people don't
1198 * use non-sparse filesystems anymore. This is already checked above.
1201 gdb_bh = sbi->s_group_desc[gdb_num];
1202 BUFFER_TRACE(gdb_bh, "get_write_access");
1203 err = ext4_journal_get_write_access(handle, gdb_bh);
1205 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1206 err = reserve_backup_gdb(handle, resize_inode, group);
1207 } else if (meta_bg != 0) {
1208 err = add_new_gdb_meta_bg(sb, handle, group);
1210 err = add_new_gdb(handle, resize_inode, group);
1218 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1220 struct buffer_head *bh = sb_getblk(sb, block);
1223 if (!bh_uptodate_or_lock(bh)) {
1224 if (bh_submit_read(bh) < 0) {
1233 static int ext4_set_bitmap_checksums(struct super_block *sb,
1235 struct ext4_group_desc *gdp,
1236 struct ext4_new_group_data *group_data)
1238 struct buffer_head *bh;
1240 if (!ext4_has_metadata_csum(sb))
1243 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1246 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1247 EXT4_INODES_PER_GROUP(sb) / 8);
1250 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1253 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1260 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1262 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1263 struct ext4_new_flex_group_data *flex_gd)
1265 struct ext4_new_group_data *group_data = flex_gd->groups;
1266 struct ext4_group_desc *gdp;
1267 struct ext4_sb_info *sbi = EXT4_SB(sb);
1268 struct buffer_head *gdb_bh;
1270 __u16 *bg_flags = flex_gd->bg_flags;
1271 int i, gdb_off, gdb_num, err = 0;
1274 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1275 group = group_data->group;
1277 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1278 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1281 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1283 gdb_bh = sbi->s_group_desc[gdb_num];
1284 /* Update group descriptor block for new group */
1285 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1286 gdb_off * EXT4_DESC_SIZE(sb));
1288 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1289 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1290 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1291 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1293 ext4_std_error(sb, err);
1297 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1298 ext4_free_group_clusters_set(sb, gdp,
1299 group_data->free_clusters_count);
1300 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1301 if (ext4_has_group_desc_csum(sb))
1302 ext4_itable_unused_set(sb, gdp,
1303 EXT4_INODES_PER_GROUP(sb));
1304 gdp->bg_flags = cpu_to_le16(*bg_flags);
1305 ext4_group_desc_csum_set(sb, group, gdp);
1307 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1308 if (unlikely(err)) {
1309 ext4_std_error(sb, err);
1314 * We can allocate memory for mb_alloc based on the new group
1317 err = ext4_mb_add_groupinfo(sb, group, gdp);
1325 * ext4_update_super() updates the super block so that the newly added
1326 * groups can be seen by the filesystem.
1329 * @flex_gd: new added groups
1331 static void ext4_update_super(struct super_block *sb,
1332 struct ext4_new_flex_group_data *flex_gd)
1334 ext4_fsblk_t blocks_count = 0;
1335 ext4_fsblk_t free_blocks = 0;
1336 ext4_fsblk_t reserved_blocks = 0;
1337 struct ext4_new_group_data *group_data = flex_gd->groups;
1338 struct ext4_sb_info *sbi = EXT4_SB(sb);
1339 struct ext4_super_block *es = sbi->s_es;
1342 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1344 * Make the new blocks and inodes valid next. We do this before
1345 * increasing the group count so that once the group is enabled,
1346 * all of its blocks and inodes are already valid.
1348 * We always allocate group-by-group, then block-by-block or
1349 * inode-by-inode within a group, so enabling these
1350 * blocks/inodes before the group is live won't actually let us
1351 * allocate the new space yet.
1353 for (i = 0; i < flex_gd->count; i++) {
1354 blocks_count += group_data[i].blocks_count;
1355 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1358 reserved_blocks = ext4_r_blocks_count(es) * 100;
1359 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1360 reserved_blocks *= blocks_count;
1361 do_div(reserved_blocks, 100);
1363 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1364 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1365 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1367 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1370 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1372 * We need to protect s_groups_count against other CPUs seeing
1373 * inconsistent state in the superblock.
1375 * The precise rules we use are:
1377 * * Writers must perform a smp_wmb() after updating all
1378 * dependent data and before modifying the groups count
1380 * * Readers must perform an smp_rmb() after reading the groups
1381 * count and before reading any dependent data.
1383 * NB. These rules can be relaxed when checking the group count
1384 * while freeing data, as we can only allocate from a block
1385 * group after serialising against the group count, and we can
1386 * only then free after serialising in turn against that
1391 /* Update the global fs size fields */
1392 sbi->s_groups_count += flex_gd->count;
1393 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1394 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1396 /* Update the reserved block counts only once the new group is
1398 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1401 /* Update the free space counts */
1402 percpu_counter_add(&sbi->s_freeclusters_counter,
1403 EXT4_NUM_B2C(sbi, free_blocks));
1404 percpu_counter_add(&sbi->s_freeinodes_counter,
1405 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1407 ext4_debug("free blocks count %llu",
1408 percpu_counter_read(&sbi->s_freeclusters_counter));
1409 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1410 ext4_group_t flex_group;
1411 flex_group = ext4_flex_group(sbi, group_data[0].group);
1412 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1413 &sbi->s_flex_groups[flex_group].free_clusters);
1414 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1415 &sbi->s_flex_groups[flex_group].free_inodes);
1419 * Update the fs overhead information
1421 ext4_calculate_overhead(sb);
1423 if (test_opt(sb, DEBUG))
1424 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1425 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1426 blocks_count, free_blocks, reserved_blocks);
1429 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1430 * _before_ we start modifying the filesystem, because we cannot abort the
1431 * transaction and not have it write the data to disk.
1433 static int ext4_flex_group_add(struct super_block *sb,
1434 struct inode *resize_inode,
1435 struct ext4_new_flex_group_data *flex_gd)
1437 struct ext4_sb_info *sbi = EXT4_SB(sb);
1438 struct ext4_super_block *es = sbi->s_es;
1439 ext4_fsblk_t o_blocks_count;
1443 unsigned reserved_gdb;
1444 int err = 0, err2 = 0, credit;
1446 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1448 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1449 o_blocks_count = ext4_blocks_count(es);
1450 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1453 err = setup_new_flex_group_blocks(sb, flex_gd);
1457 * We will always be modifying at least the superblock and GDT
1458 * blocks. If we are adding a group past the last current GDT block,
1459 * we will also modify the inode and the dindirect block. If we
1460 * are adding a group with superblock/GDT backups we will also
1461 * modify each of the reserved GDT dindirect blocks.
1463 credit = 3; /* sb, resize inode, resize inode dindirect */
1465 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1466 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1467 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1468 if (IS_ERR(handle)) {
1469 err = PTR_ERR(handle);
1473 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1474 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1478 group = flex_gd->groups[0].group;
1479 BUG_ON(group != sbi->s_groups_count);
1480 err = ext4_add_new_descs(handle, sb, group,
1481 resize_inode, flex_gd->count);
1485 err = ext4_setup_new_descs(handle, sb, flex_gd);
1489 ext4_update_super(sb, flex_gd);
1491 err = ext4_handle_dirty_super(handle, sb);
1494 err2 = ext4_journal_stop(handle);
1499 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1500 int gdb_num_end = ((group + flex_gd->count - 1) /
1501 EXT4_DESC_PER_BLOCK(sb));
1502 int meta_bg = ext4_has_feature_meta_bg(sb);
1503 sector_t old_gdb = 0;
1505 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1506 sizeof(struct ext4_super_block), 0);
1507 for (; gdb_num <= gdb_num_end; gdb_num++) {
1508 struct buffer_head *gdb_bh;
1510 gdb_bh = sbi->s_group_desc[gdb_num];
1511 if (old_gdb == gdb_bh->b_blocknr)
1513 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1514 gdb_bh->b_size, meta_bg);
1515 old_gdb = gdb_bh->b_blocknr;
1522 static int ext4_setup_next_flex_gd(struct super_block *sb,
1523 struct ext4_new_flex_group_data *flex_gd,
1524 ext4_fsblk_t n_blocks_count,
1525 unsigned long flexbg_size)
1527 struct ext4_sb_info *sbi = EXT4_SB(sb);
1528 struct ext4_super_block *es = sbi->s_es;
1529 struct ext4_new_group_data *group_data = flex_gd->groups;
1530 ext4_fsblk_t o_blocks_count;
1531 ext4_group_t n_group;
1533 ext4_group_t last_group;
1535 ext4_grpblk_t clusters_per_group;
1538 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1540 o_blocks_count = ext4_blocks_count(es);
1542 if (o_blocks_count == n_blocks_count)
1545 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1547 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1549 last_group = group | (flexbg_size - 1);
1550 if (last_group > n_group)
1551 last_group = n_group;
1553 flex_gd->count = last_group - group + 1;
1555 for (i = 0; i < flex_gd->count; i++) {
1558 group_data[i].group = group + i;
1559 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1560 overhead = ext4_group_overhead_blocks(sb, group + i);
1561 group_data[i].mdata_blocks = overhead;
1562 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1563 if (ext4_has_group_desc_csum(sb)) {
1564 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1565 EXT4_BG_INODE_UNINIT;
1566 if (!test_opt(sb, INIT_INODE_TABLE))
1567 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1569 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1572 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1573 /* We need to initialize block bitmap of last group. */
1574 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1576 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1577 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1578 group_data[i - 1].free_clusters_count -= clusters_per_group -
1585 /* Add group descriptor data to an existing or new group descriptor block.
1586 * Ensure we handle all possible error conditions _before_ we start modifying
1587 * the filesystem, because we cannot abort the transaction and not have it
1588 * write the data to disk.
1590 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1591 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1593 * We only need to hold the superblock lock while we are actually adding
1594 * in the new group's counts to the superblock. Prior to that we have
1595 * not really "added" the group at all. We re-check that we are still
1596 * adding in the last group in case things have changed since verifying.
1598 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1600 struct ext4_new_flex_group_data flex_gd;
1601 struct ext4_sb_info *sbi = EXT4_SB(sb);
1602 struct ext4_super_block *es = sbi->s_es;
1603 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1604 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1605 struct inode *inode = NULL;
1610 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1612 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1613 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1617 if (ext4_blocks_count(es) + input->blocks_count <
1618 ext4_blocks_count(es)) {
1619 ext4_warning(sb, "blocks_count overflow");
1623 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1624 le32_to_cpu(es->s_inodes_count)) {
1625 ext4_warning(sb, "inodes_count overflow");
1629 if (reserved_gdb || gdb_off == 0) {
1630 if (ext4_has_feature_resize_inode(sb) ||
1631 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1633 "No reserved GDT blocks, can't resize");
1636 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1637 if (IS_ERR(inode)) {
1638 ext4_warning(sb, "Error opening resize inode");
1639 return PTR_ERR(inode);
1644 err = verify_group_input(sb, input);
1648 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1652 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1657 flex_gd.groups = input;
1658 flex_gd.bg_flags = &bg_flags;
1659 err = ext4_flex_group_add(sb, inode, &flex_gd);
1663 } /* ext4_group_add */
1666 * extend a group without checking assuming that checking has been done.
1668 static int ext4_group_extend_no_check(struct super_block *sb,
1669 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1671 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1675 /* We will update the superblock, one block bitmap, and
1676 * one group descriptor via ext4_group_add_blocks().
1678 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1679 if (IS_ERR(handle)) {
1680 err = PTR_ERR(handle);
1681 ext4_warning(sb, "error %d on journal start", err);
1685 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1686 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1688 ext4_warning(sb, "error %d on journal write access", err);
1692 ext4_blocks_count_set(es, o_blocks_count + add);
1693 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1694 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1695 o_blocks_count + add);
1696 /* We add the blocks to the bitmap and set the group need init bit */
1697 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1700 ext4_handle_dirty_super(handle, sb);
1701 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1702 o_blocks_count + add);
1704 err2 = ext4_journal_stop(handle);
1709 if (test_opt(sb, DEBUG))
1710 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1711 "blocks\n", ext4_blocks_count(es));
1712 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1713 (char *)es, sizeof(struct ext4_super_block), 0);
1719 * Extend the filesystem to the new number of blocks specified. This entry
1720 * point is only used to extend the current filesystem to the end of the last
1721 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1722 * for emergencies (because it has no dependencies on reserved blocks).
1724 * If we _really_ wanted, we could use default values to call ext4_group_add()
1725 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1726 * GDT blocks are reserved to grow to the desired size.
1728 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1729 ext4_fsblk_t n_blocks_count)
1731 ext4_fsblk_t o_blocks_count;
1734 struct buffer_head *bh;
1738 o_blocks_count = ext4_blocks_count(es);
1740 if (test_opt(sb, DEBUG))
1741 ext4_msg(sb, KERN_DEBUG,
1742 "extending last group from %llu to %llu blocks",
1743 o_blocks_count, n_blocks_count);
1745 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1748 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1749 ext4_msg(sb, KERN_ERR,
1750 "filesystem too large to resize to %llu blocks safely",
1752 if (sizeof(sector_t) < 8)
1753 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1757 if (n_blocks_count < o_blocks_count) {
1758 ext4_warning(sb, "can't shrink FS - resize aborted");
1762 /* Handle the remaining blocks in the last group only. */
1763 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1766 ext4_warning(sb, "need to use ext2online to resize further");
1770 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1772 if (o_blocks_count + add < o_blocks_count) {
1773 ext4_warning(sb, "blocks_count overflow");
1777 if (o_blocks_count + add > n_blocks_count)
1778 add = n_blocks_count - o_blocks_count;
1780 if (o_blocks_count + add < n_blocks_count)
1781 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1782 o_blocks_count + add, add);
1784 /* See if the device is actually as big as what was requested */
1785 bh = sb_bread(sb, o_blocks_count + add - 1);
1787 ext4_warning(sb, "can't read last block, resize aborted");
1792 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1794 } /* ext4_group_extend */
1797 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1799 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1803 * Release the resize inode and drop the resize_inode feature if there
1804 * are no more reserved gdt blocks, and then convert the file system
1807 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1810 struct ext4_sb_info *sbi = EXT4_SB(sb);
1811 struct ext4_super_block *es = sbi->s_es;
1812 struct ext4_inode_info *ei = EXT4_I(inode);
1814 int i, ret, err = 0;
1817 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1819 if (es->s_reserved_gdt_blocks) {
1820 ext4_error(sb, "Unexpected non-zero "
1821 "s_reserved_gdt_blocks");
1825 /* Do a quick sanity check of the resize inode */
1826 if (inode->i_blocks != 1 << (inode->i_blkbits -
1827 (9 - sbi->s_cluster_bits)))
1828 goto invalid_resize_inode;
1829 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1830 if (i == EXT4_DIND_BLOCK) {
1834 goto invalid_resize_inode;
1837 goto invalid_resize_inode;
1839 credits += 3; /* block bitmap, bg descriptor, resize inode */
1842 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1844 return PTR_ERR(handle);
1846 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1847 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1851 ext4_clear_feature_resize_inode(sb);
1852 ext4_set_feature_meta_bg(sb);
1853 sbi->s_es->s_first_meta_bg =
1854 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1856 err = ext4_handle_dirty_super(handle, sb);
1858 ext4_std_error(sb, err);
1863 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1864 ext4_free_blocks(handle, inode, NULL, nr, 1,
1865 EXT4_FREE_BLOCKS_METADATA |
1866 EXT4_FREE_BLOCKS_FORGET);
1867 ei->i_data[EXT4_DIND_BLOCK] = 0;
1868 inode->i_blocks = 0;
1870 err = ext4_mark_inode_dirty(handle, inode);
1872 ext4_std_error(sb, err);
1876 ret = ext4_journal_stop(handle);
1881 invalid_resize_inode:
1882 ext4_error(sb, "corrupted/inconsistent resize inode");
1887 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1889 * @sb: super block of the fs to be resized
1890 * @n_blocks_count: the number of blocks resides in the resized fs
1892 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1894 struct ext4_new_flex_group_data *flex_gd = NULL;
1895 struct ext4_sb_info *sbi = EXT4_SB(sb);
1896 struct ext4_super_block *es = sbi->s_es;
1897 struct buffer_head *bh;
1898 struct inode *resize_inode = NULL;
1899 ext4_grpblk_t add, offset;
1900 unsigned long n_desc_blocks;
1901 unsigned long o_desc_blocks;
1902 ext4_group_t o_group;
1903 ext4_group_t n_group;
1904 ext4_fsblk_t o_blocks_count;
1905 ext4_fsblk_t n_blocks_count_retry = 0;
1906 unsigned long last_update_time = 0;
1907 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1910 /* See if the device is actually as big as what was requested */
1911 bh = sb_bread(sb, n_blocks_count - 1);
1913 ext4_warning(sb, "can't read last block, resize aborted");
1919 o_blocks_count = ext4_blocks_count(es);
1921 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1922 "to %llu blocks", o_blocks_count, n_blocks_count);
1924 if (n_blocks_count < o_blocks_count) {
1925 /* On-line shrinking not supported */
1926 ext4_warning(sb, "can't shrink FS - resize aborted");
1930 if (n_blocks_count == o_blocks_count)
1931 /* Nothing need to do */
1934 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1935 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1936 ext4_warning(sb, "resize would cause inodes_count overflow");
1939 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1941 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1942 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1944 meta_bg = ext4_has_feature_meta_bg(sb);
1946 if (ext4_has_feature_resize_inode(sb)) {
1948 ext4_error(sb, "resize_inode and meta_bg enabled "
1952 if (n_desc_blocks > o_desc_blocks +
1953 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1954 n_blocks_count_retry = n_blocks_count;
1955 n_desc_blocks = o_desc_blocks +
1956 le16_to_cpu(es->s_reserved_gdt_blocks);
1957 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1958 n_blocks_count = (ext4_fsblk_t)n_group *
1959 EXT4_BLOCKS_PER_GROUP(sb);
1960 n_group--; /* set to last group number */
1964 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1965 if (IS_ERR(resize_inode)) {
1966 ext4_warning(sb, "Error opening resize inode");
1967 return PTR_ERR(resize_inode);
1971 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1972 err = ext4_convert_meta_bg(sb, resize_inode);
1977 resize_inode = NULL;
1979 if (n_blocks_count_retry) {
1980 n_blocks_count = n_blocks_count_retry;
1981 n_blocks_count_retry = 0;
1987 * Make sure the last group has enough space so that it's
1988 * guaranteed to have enough space for all metadata blocks
1989 * that it might need to hold. (We might not need to store
1990 * the inode table blocks in the last block group, but there
1991 * will be cases where this might be needed.)
1993 if ((ext4_group_first_block_no(sb, n_group) +
1994 ext4_group_overhead_blocks(sb, n_group) + 2 +
1995 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
1996 n_blocks_count = ext4_group_first_block_no(sb, n_group);
1998 n_blocks_count_retry = 0;
2001 resize_inode = NULL;
2006 /* extend the last group */
2007 if (n_group == o_group)
2008 add = n_blocks_count - o_blocks_count;
2010 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2012 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2017 if (ext4_blocks_count(es) == n_blocks_count)
2020 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2024 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2028 flex_gd = alloc_flex_gd(flexbg_size);
2029 if (flex_gd == NULL) {
2034 /* Add flex groups. Note that a regular group is a
2035 * flex group with 1 group.
2037 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2039 if (jiffies - last_update_time > HZ * 10) {
2040 if (last_update_time)
2041 ext4_msg(sb, KERN_INFO,
2042 "resized to %llu blocks",
2043 ext4_blocks_count(es));
2044 last_update_time = jiffies;
2046 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2048 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2053 if (!err && n_blocks_count_retry) {
2054 n_blocks_count = n_blocks_count_retry;
2055 n_blocks_count_retry = 0;
2056 free_flex_gd(flex_gd);
2063 free_flex_gd(flex_gd);
2064 if (resize_inode != NULL)
2066 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);