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"
24 static void ext4_rcu_ptr_callback(struct rcu_head *head)
26 struct ext4_rcu_ptr *ptr;
28 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
33 void ext4_kvfree_array_rcu(void *to_free)
35 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
39 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
46 int ext4_resize_begin(struct super_block *sb)
48 struct ext4_sb_info *sbi = EXT4_SB(sb);
51 if (!capable(CAP_SYS_RESOURCE))
55 * If the reserved GDT blocks is non-zero, the resize_inode feature
56 * should always be set.
58 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
59 !ext4_has_feature_resize_inode(sb)) {
60 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
65 * If we are not using the primary superblock/GDT copy don't resize,
66 * because the user tools have no way of handling this. Probably a
67 * bad time to do it anyways.
69 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
70 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
71 ext4_warning(sb, "won't resize using backup superblock at %llu",
72 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
77 * We are not allowed to do online-resizing on a filesystem mounted
78 * with error, because it can destroy the filesystem easily.
80 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
81 ext4_warning(sb, "There are errors in the filesystem, "
82 "so online resizing is not allowed");
86 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
92 void ext4_resize_end(struct super_block *sb)
94 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
95 smp_mb__after_atomic();
98 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
100 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
101 EXT4_DESC_PER_BLOCK_BITS(sb);
104 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
105 ext4_group_t group) {
106 group = ext4_meta_bg_first_group(sb, group);
107 return ext4_group_first_block_no(sb, group);
110 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
111 ext4_group_t group) {
112 ext4_grpblk_t overhead;
113 overhead = ext4_bg_num_gdb(sb, group);
114 if (ext4_bg_has_super(sb, group))
116 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
120 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
121 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
123 static int verify_group_input(struct super_block *sb,
124 struct ext4_new_group_data *input)
126 struct ext4_sb_info *sbi = EXT4_SB(sb);
127 struct ext4_super_block *es = sbi->s_es;
128 ext4_fsblk_t start = ext4_blocks_count(es);
129 ext4_fsblk_t end = start + input->blocks_count;
130 ext4_group_t group = input->group;
131 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
133 ext4_fsblk_t metaend;
134 struct buffer_head *bh = NULL;
135 ext4_grpblk_t free_blocks_count, offset;
138 if (group != sbi->s_groups_count) {
139 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
140 input->group, sbi->s_groups_count);
144 overhead = ext4_group_overhead_blocks(sb, group);
145 metaend = start + overhead;
146 input->free_blocks_count = free_blocks_count =
147 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
149 if (test_opt(sb, DEBUG))
150 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
151 "(%d free, %u reserved)\n",
152 ext4_bg_has_super(sb, input->group) ? "normal" :
153 "no-super", input->group, input->blocks_count,
154 free_blocks_count, input->reserved_blocks);
156 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
158 ext4_warning(sb, "Last group not full");
159 else if (input->reserved_blocks > input->blocks_count / 5)
160 ext4_warning(sb, "Reserved blocks too high (%u)",
161 input->reserved_blocks);
162 else if (free_blocks_count < 0)
163 ext4_warning(sb, "Bad blocks count %u",
164 input->blocks_count);
165 else if (!(bh = sb_bread(sb, end - 1)))
166 ext4_warning(sb, "Cannot read last block (%llu)",
168 else if (outside(input->block_bitmap, start, end))
169 ext4_warning(sb, "Block bitmap not in group (block %llu)",
170 (unsigned long long)input->block_bitmap);
171 else if (outside(input->inode_bitmap, start, end))
172 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
173 (unsigned long long)input->inode_bitmap);
174 else if (outside(input->inode_table, start, end) ||
175 outside(itend - 1, start, end))
176 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
177 (unsigned long long)input->inode_table, itend - 1);
178 else if (input->inode_bitmap == input->block_bitmap)
179 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
180 (unsigned long long)input->block_bitmap);
181 else if (inside(input->block_bitmap, input->inode_table, itend))
182 ext4_warning(sb, "Block bitmap (%llu) in inode table "
184 (unsigned long long)input->block_bitmap,
185 (unsigned long long)input->inode_table, itend - 1);
186 else if (inside(input->inode_bitmap, input->inode_table, itend))
187 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
189 (unsigned long long)input->inode_bitmap,
190 (unsigned long long)input->inode_table, itend - 1);
191 else if (inside(input->block_bitmap, start, metaend))
192 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
193 (unsigned long long)input->block_bitmap,
195 else if (inside(input->inode_bitmap, start, metaend))
196 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
197 (unsigned long long)input->inode_bitmap,
199 else if (inside(input->inode_table, start, metaend) ||
200 inside(itend - 1, start, metaend))
201 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
203 (unsigned long long)input->inode_table,
204 itend - 1, start, metaend - 1);
213 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
216 struct ext4_new_flex_group_data {
217 struct ext4_new_group_data *groups; /* new_group_data for groups
219 __u16 *bg_flags; /* block group flags of groups
221 ext4_group_t count; /* number of groups in @groups
226 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
229 * Returns NULL on failure otherwise address of the allocated structure.
231 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
233 struct ext4_new_flex_group_data *flex_gd;
235 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
239 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
241 flex_gd->count = flexbg_size;
243 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
244 flexbg_size, GFP_NOFS);
245 if (flex_gd->groups == NULL)
248 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
249 if (flex_gd->bg_flags == NULL)
255 kfree(flex_gd->groups);
262 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
264 kfree(flex_gd->bg_flags);
265 kfree(flex_gd->groups);
270 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
271 * and inode tables for a flex group.
273 * This function is used by 64bit-resize. Note that this function allocates
274 * group tables from the 1st group of groups contained by @flexgd, which may
275 * be a partial of a flex group.
277 * @sb: super block of fs to which the groups belongs
279 * Returns 0 on a successful allocation of the metadata blocks in the
282 static int ext4_alloc_group_tables(struct super_block *sb,
283 struct ext4_new_flex_group_data *flex_gd,
286 struct ext4_new_group_data *group_data = flex_gd->groups;
287 ext4_fsblk_t start_blk;
288 ext4_fsblk_t last_blk;
289 ext4_group_t src_group;
290 ext4_group_t bb_index = 0;
291 ext4_group_t ib_index = 0;
292 ext4_group_t it_index = 0;
294 ext4_group_t last_group;
296 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
298 BUG_ON(flex_gd->count == 0 || group_data == NULL);
300 src_group = group_data[0].group;
301 last_group = src_group + flex_gd->count - 1;
303 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
304 (last_group & ~(flexbg_size - 1))));
306 group = group_data[0].group;
307 if (src_group >= group_data[0].group + flex_gd->count)
309 start_blk = ext4_group_first_block_no(sb, src_group);
310 last_blk = start_blk + group_data[src_group - group].blocks_count;
312 overhead = ext4_group_overhead_blocks(sb, src_group);
314 start_blk += overhead;
316 /* We collect contiguous blocks as much as possible. */
318 for (; src_group <= last_group; src_group++) {
319 overhead = ext4_group_overhead_blocks(sb, src_group);
321 last_blk += group_data[src_group - group].blocks_count;
326 /* Allocate block bitmaps */
327 for (; bb_index < flex_gd->count; bb_index++) {
328 if (start_blk >= last_blk)
330 group_data[bb_index].block_bitmap = start_blk++;
331 group = ext4_get_group_number(sb, start_blk - 1);
332 group -= group_data[0].group;
333 group_data[group].free_blocks_count--;
334 flex_gd->bg_flags[group] &= uninit_mask;
337 /* Allocate inode bitmaps */
338 for (; ib_index < flex_gd->count; ib_index++) {
339 if (start_blk >= last_blk)
341 group_data[ib_index].inode_bitmap = start_blk++;
342 group = ext4_get_group_number(sb, start_blk - 1);
343 group -= group_data[0].group;
344 group_data[group].free_blocks_count--;
345 flex_gd->bg_flags[group] &= uninit_mask;
348 /* Allocate inode tables */
349 for (; it_index < flex_gd->count; it_index++) {
350 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
351 ext4_fsblk_t next_group_start;
353 if (start_blk + itb > last_blk)
355 group_data[it_index].inode_table = start_blk;
356 group = ext4_get_group_number(sb, start_blk);
357 next_group_start = ext4_group_first_block_no(sb, group + 1);
358 group -= group_data[0].group;
360 if (start_blk + itb > next_group_start) {
361 flex_gd->bg_flags[group + 1] &= uninit_mask;
362 overhead = start_blk + itb - next_group_start;
363 group_data[group + 1].free_blocks_count -= overhead;
367 group_data[group].free_blocks_count -= itb;
368 flex_gd->bg_flags[group] &= uninit_mask;
369 start_blk += EXT4_SB(sb)->s_itb_per_group;
372 if (test_opt(sb, DEBUG)) {
374 group = group_data[0].group;
376 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
377 "%d groups, flexbg size is %d:\n", flex_gd->count,
380 for (i = 0; i < flex_gd->count; i++) {
381 printk(KERN_DEBUG "adding %s group %u: %u "
382 "blocks (%d free)\n",
383 ext4_bg_has_super(sb, group + i) ? "normal" :
384 "no-super", group + i,
385 group_data[i].blocks_count,
386 group_data[i].free_blocks_count);
392 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
395 struct buffer_head *bh;
398 bh = sb_getblk(sb, blk);
400 return ERR_PTR(-ENOMEM);
401 BUFFER_TRACE(bh, "get_write_access");
402 if ((err = ext4_journal_get_write_access(handle, bh))) {
406 memset(bh->b_data, 0, sb->s_blocksize);
407 set_buffer_uptodate(bh);
414 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
415 * If that fails, restart the transaction & regain write access for the
416 * buffer head which is used for block_bitmap modifications.
418 static int extend_or_restart_transaction(handle_t *handle, int thresh)
422 if (ext4_handle_has_enough_credits(handle, thresh))
425 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
429 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
438 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
440 * Helper function for ext4_setup_new_group_blocks() which set .
443 * @handle: journal handle
444 * @flex_gd: flex group data
446 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
447 struct ext4_new_flex_group_data *flex_gd,
448 ext4_fsblk_t block, ext4_group_t count)
452 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
453 for (count2 = count; count > 0; count -= count2, block += count2) {
455 struct buffer_head *bh;
459 group = ext4_get_group_number(sb, block);
460 start = ext4_group_first_block_no(sb, group);
461 group -= flex_gd->groups[0].group;
463 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
467 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
468 BUG_ON(flex_gd->count > 1);
472 err = extend_or_restart_transaction(handle, 1);
476 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
480 BUFFER_TRACE(bh, "get_write_access");
481 err = ext4_journal_get_write_access(handle, bh);
486 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
487 block - start, count2);
488 ext4_set_bits(bh->b_data, block - start, count2);
490 err = ext4_handle_dirty_metadata(handle, NULL, bh);
500 * Set up the block and inode bitmaps, and the inode table for the new groups.
501 * This doesn't need to be part of the main transaction, since we are only
502 * changing blocks outside the actual filesystem. We still do journaling to
503 * ensure the recovery is correct in case of a failure just after resize.
504 * If any part of this fails, we simply abort the resize.
506 * setup_new_flex_group_blocks handles a flex group as follow:
507 * 1. copy super block and GDT, and initialize group tables if necessary.
508 * In this step, we only set bits in blocks bitmaps for blocks taken by
509 * super block and GDT.
510 * 2. allocate group tables in block bitmaps, that is, set bits in block
511 * bitmap for blocks taken by group tables.
513 static int setup_new_flex_group_blocks(struct super_block *sb,
514 struct ext4_new_flex_group_data *flex_gd)
516 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
519 struct ext4_sb_info *sbi = EXT4_SB(sb);
520 struct ext4_super_block *es = sbi->s_es;
521 struct ext4_new_group_data *group_data = flex_gd->groups;
522 __u16 *bg_flags = flex_gd->bg_flags;
524 ext4_group_t group, count;
525 struct buffer_head *bh = NULL;
526 int reserved_gdb, i, j, err = 0, err2;
529 BUG_ON(!flex_gd->count || !group_data ||
530 group_data[0].group != sbi->s_groups_count);
532 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
533 meta_bg = ext4_has_feature_meta_bg(sb);
535 /* This transaction may be extended/restarted along the way */
536 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
538 return PTR_ERR(handle);
540 group = group_data[0].group;
541 for (i = 0; i < flex_gd->count; i++, group++) {
542 unsigned long gdblocks;
543 ext4_grpblk_t overhead;
545 gdblocks = ext4_bg_num_gdb(sb, group);
546 start = ext4_group_first_block_no(sb, group);
548 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
552 ext4_group_t first_group;
553 first_group = ext4_meta_bg_first_group(sb, group);
554 if (first_group != group + 1 &&
555 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
559 block = start + ext4_bg_has_super(sb, group);
560 /* Copy all of the GDT blocks into the backup in this group */
561 for (j = 0; j < gdblocks; j++, block++) {
562 struct buffer_head *gdb;
564 ext4_debug("update backup group %#04llx\n", block);
565 err = extend_or_restart_transaction(handle, 1);
569 gdb = sb_getblk(sb, block);
570 if (unlikely(!gdb)) {
575 BUFFER_TRACE(gdb, "get_write_access");
576 err = ext4_journal_get_write_access(handle, gdb);
581 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
582 s_group_desc, j)->b_data, gdb->b_size);
583 set_buffer_uptodate(gdb);
585 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
593 /* Zero out all of the reserved backup group descriptor
596 if (ext4_bg_has_super(sb, group)) {
597 err = sb_issue_zeroout(sb, gdblocks + start + 1,
598 reserved_gdb, GFP_NOFS);
604 /* Initialize group tables of the grop @group */
605 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
608 /* Zero out all of the inode table blocks */
609 block = group_data[i].inode_table;
610 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
611 block, sbi->s_itb_per_group);
612 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
618 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
621 /* Initialize block bitmap of the @group */
622 block = group_data[i].block_bitmap;
623 err = extend_or_restart_transaction(handle, 1);
627 bh = bclean(handle, sb, block);
632 overhead = ext4_group_overhead_blocks(sb, group);
634 ext4_debug("mark backup superblock %#04llx (+0)\n",
636 ext4_set_bits(bh->b_data, 0, overhead);
638 ext4_mark_bitmap_end(group_data[i].blocks_count,
639 sb->s_blocksize * 8, bh->b_data);
640 err = ext4_handle_dirty_metadata(handle, NULL, bh);
646 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
649 /* Initialize inode bitmap of the @group */
650 block = group_data[i].inode_bitmap;
651 err = extend_or_restart_transaction(handle, 1);
654 /* Mark unused entries in inode bitmap used */
655 bh = bclean(handle, sb, block);
661 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
662 sb->s_blocksize * 8, bh->b_data);
663 err = ext4_handle_dirty_metadata(handle, NULL, bh);
669 /* Mark group tables in block bitmap */
670 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
671 count = group_table_count[j];
672 start = (&group_data[0].block_bitmap)[j];
674 for (i = 1; i < flex_gd->count; i++) {
675 block += group_table_count[j];
676 if (block == (&group_data[i].block_bitmap)[j]) {
677 count += group_table_count[j];
680 err = set_flexbg_block_bitmap(sb, handle,
681 flex_gd, start, count);
684 count = group_table_count[j];
685 start = (&group_data[i].block_bitmap)[j];
690 err = set_flexbg_block_bitmap(sb, handle,
691 flex_gd, start, count);
698 err2 = ext4_journal_stop(handle);
706 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
707 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
708 * calling this for the first time. In a sparse filesystem it will be the
709 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
710 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
712 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
713 unsigned *five, unsigned *seven)
715 unsigned *min = three;
719 if (!ext4_has_feature_sparse_super(sb)) {
741 * Check that all of the backup GDT blocks are held in the primary GDT block.
742 * It is assumed that they are stored in group order. Returns the number of
743 * groups in current filesystem that have BACKUPS, or -ve error code.
745 static int verify_reserved_gdb(struct super_block *sb,
747 struct buffer_head *primary)
749 const ext4_fsblk_t blk = primary->b_blocknr;
754 __le32 *p = (__le32 *)primary->b_data;
757 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
758 if (le32_to_cpu(*p++) !=
759 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
760 ext4_warning(sb, "reserved GDT %llu"
761 " missing grp %d (%llu)",
764 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
768 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
776 * Called when we need to bring a reserved group descriptor table block into
777 * use from the resize inode. The primary copy of the new GDT block currently
778 * is an indirect block (under the double indirect block in the resize inode).
779 * The new backup GDT blocks will be stored as leaf blocks in this indirect
780 * block, in group order. Even though we know all the block numbers we need,
781 * we check to ensure that the resize inode has actually reserved these blocks.
783 * Don't need to update the block bitmaps because the blocks are still in use.
785 * We get all of the error cases out of the way, so that we are sure to not
786 * fail once we start modifying the data on disk, because JBD has no rollback.
788 static int add_new_gdb(handle_t *handle, struct inode *inode,
791 struct super_block *sb = inode->i_sb;
792 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
793 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
794 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
795 struct buffer_head **o_group_desc, **n_group_desc;
796 struct buffer_head *dind;
797 struct buffer_head *gdb_bh;
799 struct ext4_iloc iloc;
803 if (test_opt(sb, DEBUG))
805 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
808 gdb_bh = sb_bread(sb, gdblock);
812 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
818 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
819 dind = sb_bread(sb, le32_to_cpu(*data));
825 data = (__le32 *)dind->b_data;
826 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
827 ext4_warning(sb, "new group %u GDT block %llu not reserved",
833 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
834 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
838 BUFFER_TRACE(gdb_bh, "get_write_access");
839 err = ext4_journal_get_write_access(handle, gdb_bh);
843 BUFFER_TRACE(dind, "get_write_access");
844 err = ext4_journal_get_write_access(handle, dind);
846 ext4_std_error(sb, err);
848 /* ext4_reserve_inode_write() gets a reference on the iloc */
849 err = ext4_reserve_inode_write(handle, inode, &iloc);
853 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
854 sizeof(struct buffer_head *),
858 ext4_warning(sb, "not enough memory for %lu groups",
864 * Finally, we have all of the possible failures behind us...
866 * Remove new GDT block from inode double-indirect block and clear out
867 * the new GDT block for use (which also "frees" the backup GDT blocks
868 * from the reserved inode). We don't need to change the bitmaps for
869 * these blocks, because they are marked as in-use from being in the
870 * reserved inode, and will become GDT blocks (primary and backup).
872 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
873 err = ext4_handle_dirty_metadata(handle, NULL, dind);
875 ext4_std_error(sb, err);
878 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
879 ext4_mark_iloc_dirty(handle, inode, &iloc);
880 memset(gdb_bh->b_data, 0, sb->s_blocksize);
881 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
883 ext4_std_error(sb, err);
890 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
891 memcpy(n_group_desc, o_group_desc,
892 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
894 n_group_desc[gdb_num] = gdb_bh;
895 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
896 EXT4_SB(sb)->s_gdb_count++;
897 ext4_kvfree_array_rcu(o_group_desc);
899 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
900 err = ext4_handle_dirty_super(handle, sb);
902 ext4_std_error(sb, err);
907 kvfree(n_group_desc);
914 ext4_debug("leaving with error %d\n", err);
919 * add_new_gdb_meta_bg is the sister of add_new_gdb.
921 static int add_new_gdb_meta_bg(struct super_block *sb,
922 handle_t *handle, ext4_group_t group) {
923 ext4_fsblk_t gdblock;
924 struct buffer_head *gdb_bh;
925 struct buffer_head **o_group_desc, **n_group_desc;
926 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
929 gdblock = ext4_meta_bg_first_block_no(sb, group) +
930 ext4_bg_has_super(sb, group);
931 gdb_bh = sb_bread(sb, gdblock);
934 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
935 sizeof(struct buffer_head *),
940 ext4_warning(sb, "not enough memory for %lu groups",
946 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
947 memcpy(n_group_desc, o_group_desc,
948 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
950 n_group_desc[gdb_num] = gdb_bh;
952 BUFFER_TRACE(gdb_bh, "get_write_access");
953 err = ext4_journal_get_write_access(handle, gdb_bh);
955 kvfree(n_group_desc);
960 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
961 EXT4_SB(sb)->s_gdb_count++;
962 ext4_kvfree_array_rcu(o_group_desc);
967 * Called when we are adding a new group which has a backup copy of each of
968 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
969 * We need to add these reserved backup GDT blocks to the resize inode, so
970 * that they are kept for future resizing and not allocated to files.
972 * Each reserved backup GDT block will go into a different indirect block.
973 * The indirect blocks are actually the primary reserved GDT blocks,
974 * so we know in advance what their block numbers are. We only get the
975 * double-indirect block to verify it is pointing to the primary reserved
976 * GDT blocks so we don't overwrite a data block by accident. The reserved
977 * backup GDT blocks are stored in their reserved primary GDT block.
979 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
982 struct super_block *sb = inode->i_sb;
983 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
984 struct buffer_head **primary;
985 struct buffer_head *dind;
986 struct ext4_iloc iloc;
993 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
997 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
998 dind = sb_bread(sb, le32_to_cpu(*data));
1004 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1005 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1006 EXT4_ADDR_PER_BLOCK(sb));
1007 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1009 /* Get each reserved primary GDT block and verify it holds backups */
1010 for (res = 0; res < reserved_gdb; res++, blk++) {
1011 if (le32_to_cpu(*data) != blk) {
1012 ext4_warning(sb, "reserved block %llu"
1013 " not at offset %ld",
1015 (long)(data - (__le32 *)dind->b_data));
1019 primary[res] = sb_bread(sb, blk);
1020 if (!primary[res]) {
1024 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1025 if (gdbackups < 0) {
1026 brelse(primary[res]);
1031 data = (__le32 *)dind->b_data;
1034 for (i = 0; i < reserved_gdb; i++) {
1035 BUFFER_TRACE(primary[i], "get_write_access");
1036 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1040 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1044 * Finally we can add each of the reserved backup GDT blocks from
1045 * the new group to its reserved primary GDT block.
1047 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1048 for (i = 0; i < reserved_gdb; i++) {
1050 data = (__le32 *)primary[i]->b_data;
1051 /* printk("reserving backup %lu[%u] = %lu\n",
1052 primary[i]->b_blocknr, gdbackups,
1053 blk + primary[i]->b_blocknr); */
1054 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1055 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1059 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1060 ext4_mark_iloc_dirty(handle, inode, &iloc);
1064 brelse(primary[res]);
1074 * Update the backup copies of the ext4 metadata. These don't need to be part
1075 * of the main resize transaction, because e2fsck will re-write them if there
1076 * is a problem (basically only OOM will cause a problem). However, we
1077 * _should_ update the backups if possible, in case the primary gets trashed
1078 * for some reason and we need to run e2fsck from a backup superblock. The
1079 * important part is that the new block and inode counts are in the backup
1080 * superblocks, and the location of the new group metadata in the GDT backups.
1082 * We do not need take the s_resize_lock for this, because these
1083 * blocks are not otherwise touched by the filesystem code when it is
1084 * mounted. We don't need to worry about last changing from
1085 * sbi->s_groups_count, because the worst that can happen is that we
1086 * do not copy the full number of backups at this time. The resize
1087 * which changed s_groups_count will backup again.
1089 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1090 int size, int meta_bg)
1092 struct ext4_sb_info *sbi = EXT4_SB(sb);
1094 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1098 ext4_group_t group = 0;
1099 int rest = sb->s_blocksize - size;
1103 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1104 if (IS_ERR(handle)) {
1106 err = PTR_ERR(handle);
1111 group = ext4_list_backups(sb, &three, &five, &seven);
1112 last = sbi->s_groups_count;
1114 group = ext4_get_group_number(sb, blk_off) + 1;
1115 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1118 while (group < sbi->s_groups_count) {
1119 struct buffer_head *bh;
1120 ext4_fsblk_t backup_block;
1122 /* Out of journal space, and can't get more - abort - so sad */
1123 if (ext4_handle_valid(handle) &&
1124 handle->h_buffer_credits == 0 &&
1125 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1126 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1130 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1132 backup_block = (ext4_group_first_block_no(sb, group) +
1133 ext4_bg_has_super(sb, group));
1135 bh = sb_getblk(sb, backup_block);
1136 if (unlikely(!bh)) {
1140 ext4_debug("update metadata backup %llu(+%llu)\n",
1141 backup_block, backup_block -
1142 ext4_group_first_block_no(sb, group));
1143 BUFFER_TRACE(bh, "get_write_access");
1144 if ((err = ext4_journal_get_write_access(handle, bh))) {
1149 memcpy(bh->b_data, data, size);
1151 memset(bh->b_data + size, 0, rest);
1152 set_buffer_uptodate(bh);
1154 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1156 ext4_std_error(sb, err);
1160 group = ext4_list_backups(sb, &three, &five, &seven);
1161 else if (group == last)
1166 if ((err2 = ext4_journal_stop(handle)) && !err)
1170 * Ugh! Need to have e2fsck write the backup copies. It is too
1171 * late to revert the resize, we shouldn't fail just because of
1172 * the backup copies (they are only needed in case of corruption).
1174 * However, if we got here we have a journal problem too, so we
1175 * can't really start a transaction to mark the superblock.
1176 * Chicken out and just set the flag on the hope it will be written
1177 * to disk, and if not - we will simply wait until next fsck.
1181 ext4_warning(sb, "can't update backup for group %u (err %d), "
1182 "forcing fsck on next reboot", group, err);
1183 sbi->s_mount_state &= ~EXT4_VALID_FS;
1184 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1185 mark_buffer_dirty(sbi->s_sbh);
1190 * ext4_add_new_descs() adds @count group descriptor of groups
1191 * starting at @group
1193 * @handle: journal handle
1195 * @group: the group no. of the first group desc to be added
1196 * @resize_inode: the resize inode
1197 * @count: number of group descriptors to be added
1199 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1200 ext4_group_t group, struct inode *resize_inode,
1203 struct ext4_sb_info *sbi = EXT4_SB(sb);
1204 struct ext4_super_block *es = sbi->s_es;
1205 struct buffer_head *gdb_bh;
1206 int i, gdb_off, gdb_num, err = 0;
1209 meta_bg = ext4_has_feature_meta_bg(sb);
1210 for (i = 0; i < count; i++, group++) {
1211 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1212 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1214 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1215 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1218 * We will only either add reserved group blocks to a backup group
1219 * or remove reserved blocks for the first group in a new group block.
1220 * Doing both would be mean more complex code, and sane people don't
1221 * use non-sparse filesystems anymore. This is already checked above.
1224 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1226 BUFFER_TRACE(gdb_bh, "get_write_access");
1227 err = ext4_journal_get_write_access(handle, gdb_bh);
1229 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1230 err = reserve_backup_gdb(handle, resize_inode, group);
1231 } else if (meta_bg != 0) {
1232 err = add_new_gdb_meta_bg(sb, handle, group);
1234 err = add_new_gdb(handle, resize_inode, group);
1242 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1244 struct buffer_head *bh = sb_getblk(sb, block);
1247 if (!bh_uptodate_or_lock(bh)) {
1248 if (bh_submit_read(bh) < 0) {
1257 static int ext4_set_bitmap_checksums(struct super_block *sb,
1259 struct ext4_group_desc *gdp,
1260 struct ext4_new_group_data *group_data)
1262 struct buffer_head *bh;
1264 if (!ext4_has_metadata_csum(sb))
1267 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1270 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1271 EXT4_INODES_PER_GROUP(sb) / 8);
1274 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1277 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1284 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1286 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1287 struct ext4_new_flex_group_data *flex_gd)
1289 struct ext4_new_group_data *group_data = flex_gd->groups;
1290 struct ext4_group_desc *gdp;
1291 struct ext4_sb_info *sbi = EXT4_SB(sb);
1292 struct buffer_head *gdb_bh;
1294 __u16 *bg_flags = flex_gd->bg_flags;
1295 int i, gdb_off, gdb_num, err = 0;
1298 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1299 group = group_data->group;
1301 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1302 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1305 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1307 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1308 /* Update group descriptor block for new group */
1309 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1310 gdb_off * EXT4_DESC_SIZE(sb));
1312 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1313 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1314 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1315 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1317 ext4_std_error(sb, err);
1321 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1322 ext4_free_group_clusters_set(sb, gdp,
1323 EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1324 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1325 if (ext4_has_group_desc_csum(sb))
1326 ext4_itable_unused_set(sb, gdp,
1327 EXT4_INODES_PER_GROUP(sb));
1328 gdp->bg_flags = cpu_to_le16(*bg_flags);
1329 ext4_group_desc_csum_set(sb, group, gdp);
1331 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1332 if (unlikely(err)) {
1333 ext4_std_error(sb, err);
1338 * We can allocate memory for mb_alloc based on the new group
1341 err = ext4_mb_add_groupinfo(sb, group, gdp);
1349 * ext4_update_super() updates the super block so that the newly added
1350 * groups can be seen by the filesystem.
1353 * @flex_gd: new added groups
1355 static void ext4_update_super(struct super_block *sb,
1356 struct ext4_new_flex_group_data *flex_gd)
1358 ext4_fsblk_t blocks_count = 0;
1359 ext4_fsblk_t free_blocks = 0;
1360 ext4_fsblk_t reserved_blocks = 0;
1361 struct ext4_new_group_data *group_data = flex_gd->groups;
1362 struct ext4_sb_info *sbi = EXT4_SB(sb);
1363 struct ext4_super_block *es = sbi->s_es;
1366 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1368 * Make the new blocks and inodes valid next. We do this before
1369 * increasing the group count so that once the group is enabled,
1370 * all of its blocks and inodes are already valid.
1372 * We always allocate group-by-group, then block-by-block or
1373 * inode-by-inode within a group, so enabling these
1374 * blocks/inodes before the group is live won't actually let us
1375 * allocate the new space yet.
1377 for (i = 0; i < flex_gd->count; i++) {
1378 blocks_count += group_data[i].blocks_count;
1379 free_blocks += group_data[i].free_blocks_count;
1382 reserved_blocks = ext4_r_blocks_count(es) * 100;
1383 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1384 reserved_blocks *= blocks_count;
1385 do_div(reserved_blocks, 100);
1387 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1388 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1389 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1391 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1394 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1396 * We need to protect s_groups_count against other CPUs seeing
1397 * inconsistent state in the superblock.
1399 * The precise rules we use are:
1401 * * Writers must perform a smp_wmb() after updating all
1402 * dependent data and before modifying the groups count
1404 * * Readers must perform an smp_rmb() after reading the groups
1405 * count and before reading any dependent data.
1407 * NB. These rules can be relaxed when checking the group count
1408 * while freeing data, as we can only allocate from a block
1409 * group after serialising against the group count, and we can
1410 * only then free after serialising in turn against that
1415 /* Update the global fs size fields */
1416 sbi->s_groups_count += flex_gd->count;
1417 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1418 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1420 /* Update the reserved block counts only once the new group is
1422 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1425 /* Update the free space counts */
1426 percpu_counter_add(&sbi->s_freeclusters_counter,
1427 EXT4_NUM_B2C(sbi, free_blocks));
1428 percpu_counter_add(&sbi->s_freeinodes_counter,
1429 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1431 ext4_debug("free blocks count %llu",
1432 percpu_counter_read(&sbi->s_freeclusters_counter));
1433 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1434 ext4_group_t flex_group;
1435 struct flex_groups *fg;
1437 flex_group = ext4_flex_group(sbi, group_data[0].group);
1438 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1439 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1440 &fg->free_clusters);
1441 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1446 * Update the fs overhead information
1448 ext4_calculate_overhead(sb);
1450 if (test_opt(sb, DEBUG))
1451 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1452 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1453 blocks_count, free_blocks, reserved_blocks);
1456 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1457 * _before_ we start modifying the filesystem, because we cannot abort the
1458 * transaction and not have it write the data to disk.
1460 static int ext4_flex_group_add(struct super_block *sb,
1461 struct inode *resize_inode,
1462 struct ext4_new_flex_group_data *flex_gd)
1464 struct ext4_sb_info *sbi = EXT4_SB(sb);
1465 struct ext4_super_block *es = sbi->s_es;
1466 ext4_fsblk_t o_blocks_count;
1470 unsigned reserved_gdb;
1471 int err = 0, err2 = 0, credit;
1473 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1475 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1476 o_blocks_count = ext4_blocks_count(es);
1477 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1480 err = setup_new_flex_group_blocks(sb, flex_gd);
1484 * We will always be modifying at least the superblock and GDT
1485 * blocks. If we are adding a group past the last current GDT block,
1486 * we will also modify the inode and the dindirect block. If we
1487 * are adding a group with superblock/GDT backups we will also
1488 * modify each of the reserved GDT dindirect blocks.
1490 credit = 3; /* sb, resize inode, resize inode dindirect */
1492 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1493 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1494 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1495 if (IS_ERR(handle)) {
1496 err = PTR_ERR(handle);
1500 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1501 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1505 group = flex_gd->groups[0].group;
1506 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1507 err = ext4_add_new_descs(handle, sb, group,
1508 resize_inode, flex_gd->count);
1512 err = ext4_setup_new_descs(handle, sb, flex_gd);
1516 ext4_update_super(sb, flex_gd);
1518 err = ext4_handle_dirty_super(handle, sb);
1521 err2 = ext4_journal_stop(handle);
1526 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1527 int gdb_num_end = ((group + flex_gd->count - 1) /
1528 EXT4_DESC_PER_BLOCK(sb));
1529 int meta_bg = ext4_has_feature_meta_bg(sb);
1530 sector_t old_gdb = 0;
1532 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1533 sizeof(struct ext4_super_block), 0);
1534 for (; gdb_num <= gdb_num_end; gdb_num++) {
1535 struct buffer_head *gdb_bh;
1537 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1539 if (old_gdb == gdb_bh->b_blocknr)
1541 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1542 gdb_bh->b_size, meta_bg);
1543 old_gdb = gdb_bh->b_blocknr;
1550 static int ext4_setup_next_flex_gd(struct super_block *sb,
1551 struct ext4_new_flex_group_data *flex_gd,
1552 ext4_fsblk_t n_blocks_count,
1553 unsigned long flexbg_size)
1555 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1556 struct ext4_new_group_data *group_data = flex_gd->groups;
1557 ext4_fsblk_t o_blocks_count;
1558 ext4_group_t n_group;
1560 ext4_group_t last_group;
1562 ext4_grpblk_t blocks_per_group;
1565 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1567 o_blocks_count = ext4_blocks_count(es);
1569 if (o_blocks_count == n_blocks_count)
1572 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1574 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1576 last_group = group | (flexbg_size - 1);
1577 if (last_group > n_group)
1578 last_group = n_group;
1580 flex_gd->count = last_group - group + 1;
1582 for (i = 0; i < flex_gd->count; i++) {
1585 group_data[i].group = group + i;
1586 group_data[i].blocks_count = blocks_per_group;
1587 overhead = ext4_group_overhead_blocks(sb, group + i);
1588 group_data[i].free_blocks_count = blocks_per_group - overhead;
1589 if (ext4_has_group_desc_csum(sb)) {
1590 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1591 EXT4_BG_INODE_UNINIT;
1592 if (!test_opt(sb, INIT_INODE_TABLE))
1593 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1595 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1598 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1599 /* We need to initialize block bitmap of last group. */
1600 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1602 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1603 group_data[i - 1].blocks_count = last + 1;
1604 group_data[i - 1].free_blocks_count -= blocks_per_group-
1611 /* Add group descriptor data to an existing or new group descriptor block.
1612 * Ensure we handle all possible error conditions _before_ we start modifying
1613 * the filesystem, because we cannot abort the transaction and not have it
1614 * write the data to disk.
1616 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1617 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1619 * We only need to hold the superblock lock while we are actually adding
1620 * in the new group's counts to the superblock. Prior to that we have
1621 * not really "added" the group at all. We re-check that we are still
1622 * adding in the last group in case things have changed since verifying.
1624 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1626 struct ext4_new_flex_group_data flex_gd;
1627 struct ext4_sb_info *sbi = EXT4_SB(sb);
1628 struct ext4_super_block *es = sbi->s_es;
1629 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1630 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1631 struct inode *inode = NULL;
1636 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1638 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1639 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1643 if (ext4_blocks_count(es) + input->blocks_count <
1644 ext4_blocks_count(es)) {
1645 ext4_warning(sb, "blocks_count overflow");
1649 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1650 le32_to_cpu(es->s_inodes_count)) {
1651 ext4_warning(sb, "inodes_count overflow");
1655 if (reserved_gdb || gdb_off == 0) {
1656 if (!ext4_has_feature_resize_inode(sb) ||
1657 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1659 "No reserved GDT blocks, can't resize");
1662 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1663 if (IS_ERR(inode)) {
1664 ext4_warning(sb, "Error opening resize inode");
1665 return PTR_ERR(inode);
1670 err = verify_group_input(sb, input);
1674 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1678 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1683 flex_gd.groups = input;
1684 flex_gd.bg_flags = &bg_flags;
1685 err = ext4_flex_group_add(sb, inode, &flex_gd);
1689 } /* ext4_group_add */
1692 * extend a group without checking assuming that checking has been done.
1694 static int ext4_group_extend_no_check(struct super_block *sb,
1695 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1697 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1701 /* We will update the superblock, one block bitmap, and
1702 * one group descriptor via ext4_group_add_blocks().
1704 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1705 if (IS_ERR(handle)) {
1706 err = PTR_ERR(handle);
1707 ext4_warning(sb, "error %d on journal start", err);
1711 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1712 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1714 ext4_warning(sb, "error %d on journal write access", err);
1718 ext4_blocks_count_set(es, o_blocks_count + add);
1719 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1720 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1721 o_blocks_count + add);
1722 /* We add the blocks to the bitmap and set the group need init bit */
1723 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1726 ext4_handle_dirty_super(handle, sb);
1727 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1728 o_blocks_count + add);
1730 err2 = ext4_journal_stop(handle);
1735 if (test_opt(sb, DEBUG))
1736 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1737 "blocks\n", ext4_blocks_count(es));
1738 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1739 (char *)es, sizeof(struct ext4_super_block), 0);
1745 * Extend the filesystem to the new number of blocks specified. This entry
1746 * point is only used to extend the current filesystem to the end of the last
1747 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1748 * for emergencies (because it has no dependencies on reserved blocks).
1750 * If we _really_ wanted, we could use default values to call ext4_group_add()
1751 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1752 * GDT blocks are reserved to grow to the desired size.
1754 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1755 ext4_fsblk_t n_blocks_count)
1757 ext4_fsblk_t o_blocks_count;
1760 struct buffer_head *bh;
1764 o_blocks_count = ext4_blocks_count(es);
1766 if (test_opt(sb, DEBUG))
1767 ext4_msg(sb, KERN_DEBUG,
1768 "extending last group from %llu to %llu blocks",
1769 o_blocks_count, n_blocks_count);
1771 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1774 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1775 ext4_msg(sb, KERN_ERR,
1776 "filesystem too large to resize to %llu blocks safely",
1778 if (sizeof(sector_t) < 8)
1779 ext4_warning(sb, "CONFIG_LBDAF not enabled");
1783 if (n_blocks_count < o_blocks_count) {
1784 ext4_warning(sb, "can't shrink FS - resize aborted");
1788 /* Handle the remaining blocks in the last group only. */
1789 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1792 ext4_warning(sb, "need to use ext2online to resize further");
1796 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1798 if (o_blocks_count + add < o_blocks_count) {
1799 ext4_warning(sb, "blocks_count overflow");
1803 if (o_blocks_count + add > n_blocks_count)
1804 add = n_blocks_count - o_blocks_count;
1806 if (o_blocks_count + add < n_blocks_count)
1807 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1808 o_blocks_count + add, add);
1810 /* See if the device is actually as big as what was requested */
1811 bh = sb_bread(sb, o_blocks_count + add - 1);
1813 ext4_warning(sb, "can't read last block, resize aborted");
1818 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1820 } /* ext4_group_extend */
1823 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1825 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1829 * Release the resize inode and drop the resize_inode feature if there
1830 * are no more reserved gdt blocks, and then convert the file system
1833 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1836 struct ext4_sb_info *sbi = EXT4_SB(sb);
1837 struct ext4_super_block *es = sbi->s_es;
1838 struct ext4_inode_info *ei = EXT4_I(inode);
1840 int i, ret, err = 0;
1843 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1845 if (es->s_reserved_gdt_blocks) {
1846 ext4_error(sb, "Unexpected non-zero "
1847 "s_reserved_gdt_blocks");
1851 /* Do a quick sanity check of the resize inode */
1852 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1853 goto invalid_resize_inode;
1854 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1855 if (i == EXT4_DIND_BLOCK) {
1859 goto invalid_resize_inode;
1862 goto invalid_resize_inode;
1864 credits += 3; /* block bitmap, bg descriptor, resize inode */
1867 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1869 return PTR_ERR(handle);
1871 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1872 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1876 ext4_clear_feature_resize_inode(sb);
1877 ext4_set_feature_meta_bg(sb);
1878 sbi->s_es->s_first_meta_bg =
1879 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1881 err = ext4_handle_dirty_super(handle, sb);
1883 ext4_std_error(sb, err);
1888 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1889 ext4_free_blocks(handle, inode, NULL, nr, 1,
1890 EXT4_FREE_BLOCKS_METADATA |
1891 EXT4_FREE_BLOCKS_FORGET);
1892 ei->i_data[EXT4_DIND_BLOCK] = 0;
1893 inode->i_blocks = 0;
1895 err = ext4_mark_inode_dirty(handle, inode);
1897 ext4_std_error(sb, err);
1901 ret = ext4_journal_stop(handle);
1906 invalid_resize_inode:
1907 ext4_error(sb, "corrupted/inconsistent resize inode");
1912 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1914 * @sb: super block of the fs to be resized
1915 * @n_blocks_count: the number of blocks resides in the resized fs
1917 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1919 struct ext4_new_flex_group_data *flex_gd = NULL;
1920 struct ext4_sb_info *sbi = EXT4_SB(sb);
1921 struct ext4_super_block *es = sbi->s_es;
1922 struct buffer_head *bh;
1923 struct inode *resize_inode = NULL;
1924 ext4_grpblk_t add, offset;
1925 unsigned long n_desc_blocks;
1926 unsigned long o_desc_blocks;
1927 ext4_group_t o_group;
1928 ext4_group_t n_group;
1929 ext4_fsblk_t o_blocks_count;
1930 ext4_fsblk_t n_blocks_count_retry = 0;
1931 unsigned long last_update_time = 0;
1932 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1935 /* See if the device is actually as big as what was requested */
1936 bh = sb_bread(sb, n_blocks_count - 1);
1938 ext4_warning(sb, "can't read last block, resize aborted");
1944 o_blocks_count = ext4_blocks_count(es);
1946 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1947 "to %llu blocks", o_blocks_count, n_blocks_count);
1949 if (n_blocks_count < o_blocks_count) {
1950 /* On-line shrinking not supported */
1951 ext4_warning(sb, "can't shrink FS - resize aborted");
1955 if (n_blocks_count == o_blocks_count)
1956 /* Nothing need to do */
1959 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1960 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1961 ext4_warning(sb, "resize would cause inodes_count overflow");
1964 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1966 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1967 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1969 meta_bg = ext4_has_feature_meta_bg(sb);
1971 if (ext4_has_feature_resize_inode(sb)) {
1973 ext4_error(sb, "resize_inode and meta_bg enabled "
1977 if (n_desc_blocks > o_desc_blocks +
1978 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1979 n_blocks_count_retry = n_blocks_count;
1980 n_desc_blocks = o_desc_blocks +
1981 le16_to_cpu(es->s_reserved_gdt_blocks);
1982 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1983 n_blocks_count = (ext4_fsblk_t)n_group *
1984 EXT4_BLOCKS_PER_GROUP(sb) +
1985 le32_to_cpu(es->s_first_data_block);
1986 n_group--; /* set to last group number */
1990 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1992 if (IS_ERR(resize_inode)) {
1993 ext4_warning(sb, "Error opening resize inode");
1994 return PTR_ERR(resize_inode);
1998 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1999 err = ext4_convert_meta_bg(sb, resize_inode);
2004 resize_inode = NULL;
2006 if (n_blocks_count_retry) {
2007 n_blocks_count = n_blocks_count_retry;
2008 n_blocks_count_retry = 0;
2014 * Make sure the last group has enough space so that it's
2015 * guaranteed to have enough space for all metadata blocks
2016 * that it might need to hold. (We might not need to store
2017 * the inode table blocks in the last block group, but there
2018 * will be cases where this might be needed.)
2020 if ((ext4_group_first_block_no(sb, n_group) +
2021 ext4_group_overhead_blocks(sb, n_group) + 2 +
2022 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2023 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2025 n_blocks_count_retry = 0;
2028 resize_inode = NULL;
2033 /* extend the last group */
2034 if (n_group == o_group)
2035 add = n_blocks_count - o_blocks_count;
2037 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
2039 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2044 if (ext4_blocks_count(es) == n_blocks_count)
2047 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2051 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2055 flex_gd = alloc_flex_gd(flexbg_size);
2056 if (flex_gd == NULL) {
2061 /* Add flex groups. Note that a regular group is a
2062 * flex group with 1 group.
2064 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2066 if (jiffies - last_update_time > HZ * 10) {
2067 if (last_update_time)
2068 ext4_msg(sb, KERN_INFO,
2069 "resized to %llu blocks",
2070 ext4_blocks_count(es));
2071 last_update_time = jiffies;
2073 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2075 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2080 if (!err && n_blocks_count_retry) {
2081 n_blocks_count = n_blocks_count_retry;
2082 n_blocks_count_retry = 0;
2083 free_flex_gd(flex_gd);
2087 resize_inode = NULL;
2094 free_flex_gd(flex_gd);
2095 if (resize_inode != NULL)
2098 ext4_warning(sb, "error (%d) occurred during "
2099 "file system resize", err);
2100 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2101 ext4_blocks_count(es));