blob: 707d3f16f7ce63732222e78c247a8a87e668b433 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/resize.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
Mingming Cao617ba132006-10-11 01:20:53 -07004 * Support for resizing an ext4 filesystem while it is mounted.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005 *
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7 *
8 * This could probably be made into a module, because it is not often in use.
9 */
10
11
Mingming Cao617ba132006-10-11 01:20:53 -070012#define EXT4FS_DEBUG
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013
Dave Kleikampac27a0e2006-10-11 01:20:50 -070014#include <linux/errno.h>
15#include <linux/slab.h>
16
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040017#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018
Yongqiang Yang8f82f842011-07-26 21:35:44 -040019int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
Yongqiang Yangce723c32011-07-26 21:39:09 -040026 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
Yongqiang Yang8f82f842011-07-26 21:35:44 -040036 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048#define outside(b, first, last) ((b) < (first) || (b) >= (last))
49#define inside(b, first, last) ((b) >= (first) && (b) < (last))
50
51static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070052 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053{
Mingming Cao617ba132006-10-11 01:20:53 -070054 struct ext4_sb_info *sbi = EXT4_SB(sb);
55 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070056 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070057 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050058 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070059 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 unsigned overhead = ext4_bg_has_super(sb, group) ?
61 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070063 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070065 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066 int err = -EINVAL;
67
68 input->free_blocks_count = free_blocks_count =
69 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
70
71 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070072 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070073 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070074 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 "no-super", input->group, input->blocks_count,
76 free_blocks_count, input->reserved_blocks);
77
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070078 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070079 if (group != sbi->s_groups_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070082 else if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050083 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -050085 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086 input->reserved_blocks);
87 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050088 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070089 input->blocks_count);
90 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -050091 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070092 end - 1);
93 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050094 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070095 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050097 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070098 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400100 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500101 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700102 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500104 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700105 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500107 ext4_warning(sb, "Block bitmap (%llu) in inode table "
108 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700109 (unsigned long long)input->block_bitmap,
110 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700111 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500112 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
113 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700114 (unsigned long long)input->inode_bitmap,
115 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500117 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700118 (unsigned long long)input->block_bitmap,
119 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500121 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700122 (unsigned long long)input->inode_bitmap,
123 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700124 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400125 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500126 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
127 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700128 (unsigned long long)input->inode_table,
129 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130 else
131 err = 0;
132 brelse(bh);
133
134 return err;
135}
136
137static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700138 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700139{
140 struct buffer_head *bh;
141 int err;
142
143 bh = sb_getblk(sb, blk);
144 if (!bh)
145 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700146 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700147 brelse(bh);
148 bh = ERR_PTR(err);
149 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150 memset(bh->b_data, 0, sb->s_blocksize);
151 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700152 }
153
154 return bh;
155}
156
157/*
Eric Sandeen14904102007-10-16 18:38:25 -0400158 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
159 * If that fails, restart the transaction & regain write access for the
160 * buffer head which is used for block_bitmap modifications.
161 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400162static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400163{
164 int err;
165
Frank Mayhar03901312009-01-07 00:06:22 -0500166 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400167 return 0;
168
169 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
170 if (err < 0)
171 return err;
172 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400173 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
174 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400175 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400176 }
Eric Sandeen14904102007-10-16 18:38:25 -0400177
178 return 0;
179}
180
181/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182 * Set up the block and inode bitmaps, and the inode table for the new group.
183 * This doesn't need to be part of the main transaction, since we are only
184 * changing blocks outside the actual filesystem. We still do journaling to
185 * ensure the recovery is correct in case of a failure just after resize.
186 * If any part of this fails, we simply abort the resize.
187 */
188static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700189 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700190{
Mingming Cao617ba132006-10-11 01:20:53 -0700191 struct ext4_sb_info *sbi = EXT4_SB(sb);
192 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
193 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700195 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700196 struct buffer_head *bh;
197 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700198 ext4_fsblk_t block;
199 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700200 int i;
201 int err = 0, err2;
202
Eric Sandeen14904102007-10-16 18:38:25 -0400203 /* This transaction may be extended/restarted along the way */
204 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
205
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700206 if (IS_ERR(handle))
207 return PTR_ERR(handle);
208
Yongqiang Yang8f82f842011-07-26 21:35:44 -0400209 BUG_ON(input->group != sbi->s_groups_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211 /* Copy all of the GDT blocks into the backup in this group */
212 for (i = 0, bit = 1, block = start + 1;
213 i < gdblocks; i++, block++, bit++) {
214 struct buffer_head *gdb;
215
Eric Sandeenc549a952008-01-28 23:58:27 -0500216 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400217 err = extend_or_restart_transaction(handle, 1);
218 if (err)
219 goto exit_journal;
Eric Sandeen14904102007-10-16 18:38:25 -0400220
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700221 gdb = sb_getblk(sb, block);
222 if (!gdb) {
223 err = -EIO;
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400224 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700225 }
Mingming Cao617ba132006-10-11 01:20:53 -0700226 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700227 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400228 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400230 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231 set_buffer_uptodate(gdb);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500232 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
233 if (unlikely(err)) {
234 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400235 goto exit_journal;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500236 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237 brelse(gdb);
238 }
239
240 /* Zero out all of the reserved backup group descriptor table blocks */
Theodore Ts'oda4889452011-02-21 20:39:58 -0500241 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400242 block, sbi->s_itb_per_group);
243 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400244 GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400245 if (err)
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400246 goto exit_journal;
247
248 err = extend_or_restart_transaction(handle, 2);
249 if (err)
250 goto exit_journal;
251
252 bh = bclean(handle, sb, input->block_bitmap);
253 if (IS_ERR(bh)) {
254 err = PTR_ERR(bh);
255 goto exit_journal;
256 }
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400257
258 if (ext4_bg_has_super(sb, input->group)) {
259 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
260 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
261 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262
Eric Sandeenc549a952008-01-28 23:58:27 -0500263 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700265 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500266 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700267 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700268 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700269
270 /* Zero out all of the inode table blocks */
Lukas Czernera31437b2010-10-27 21:30:05 -0400271 block = input->inode_table;
Theodore Ts'oda4889452011-02-21 20:39:58 -0500272 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400273 block, sbi->s_itb_per_group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400274 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400275 if (err)
276 goto exit_bh;
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400277 ext4_set_bits(bh->b_data, input->inode_table - start,
278 sbi->s_itb_per_group);
Eric Sandeen14904102007-10-16 18:38:25 -0400279
Eric Sandeen14904102007-10-16 18:38:25 -0400280
Theodore Ts'o61d08672010-10-27 21:30:15 -0400281 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
282 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500283 err = ext4_handle_dirty_metadata(handle, NULL, bh);
284 if (unlikely(err)) {
285 ext4_std_error(sb, err);
286 goto exit_bh;
287 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700288 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500290 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 input->inode_bitmap, input->inode_bitmap - start);
292 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
293 err = PTR_ERR(bh);
294 goto exit_journal;
295 }
296
Theodore Ts'o61d08672010-10-27 21:30:15 -0400297 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
298 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500299 err = ext4_handle_dirty_metadata(handle, NULL, bh);
300 if (unlikely(err))
301 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700302exit_bh:
303 brelse(bh);
304
305exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -0700306 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307 err = err2;
308
309 return err;
310}
311
312/*
313 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700314 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315 * calling this for the first time. In a sparse filesystem it will be the
316 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
317 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
318 */
Mingming Cao617ba132006-10-11 01:20:53 -0700319static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700320 unsigned *five, unsigned *seven)
321{
322 unsigned *min = three;
323 int mult = 3;
324 unsigned ret;
325
Mingming Cao617ba132006-10-11 01:20:53 -0700326 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
327 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700328 ret = *min;
329 *min += 1;
330 return ret;
331 }
332
333 if (*five < *min) {
334 min = five;
335 mult = 5;
336 }
337 if (*seven < *min) {
338 min = seven;
339 mult = 7;
340 }
341
342 ret = *min;
343 *min *= mult;
344
345 return ret;
346}
347
348/*
349 * Check that all of the backup GDT blocks are held in the primary GDT block.
350 * It is assumed that they are stored in group order. Returns the number of
351 * groups in current filesystem that have BACKUPS, or -ve error code.
352 */
353static int verify_reserved_gdb(struct super_block *sb,
354 struct buffer_head *primary)
355{
Mingming Cao617ba132006-10-11 01:20:53 -0700356 const ext4_fsblk_t blk = primary->b_blocknr;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500357 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700358 unsigned three = 1;
359 unsigned five = 5;
360 unsigned seven = 7;
361 unsigned grp;
362 __le32 *p = (__le32 *)primary->b_data;
363 int gdbackups = 0;
364
Mingming Cao617ba132006-10-11 01:20:53 -0700365 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700366 if (le32_to_cpu(*p++) !=
367 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500368 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700369 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700370 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700371 grp *
372 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
373 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700374 return -EINVAL;
375 }
Mingming Cao617ba132006-10-11 01:20:53 -0700376 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700377 return -EFBIG;
378 }
379
380 return gdbackups;
381}
382
383/*
384 * Called when we need to bring a reserved group descriptor table block into
385 * use from the resize inode. The primary copy of the new GDT block currently
386 * is an indirect block (under the double indirect block in the resize inode).
387 * The new backup GDT blocks will be stored as leaf blocks in this indirect
388 * block, in group order. Even though we know all the block numbers we need,
389 * we check to ensure that the resize inode has actually reserved these blocks.
390 *
391 * Don't need to update the block bitmaps because the blocks are still in use.
392 *
393 * We get all of the error cases out of the way, so that we are sure to not
394 * fail once we start modifying the data on disk, because JBD has no rollback.
395 */
396static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400397 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700398{
399 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700400 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400401 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700402 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700403 struct buffer_head **o_group_desc, **n_group_desc;
404 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400405 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700406 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700407 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700408 __le32 *data;
409 int err;
410
411 if (test_opt(sb, DEBUG))
412 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700413 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700414 gdb_num);
415
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400416 /*
417 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400418 * because the user tools have no way of handling this. Probably a
419 * bad time to do it anyways.
420 */
Mingming Cao617ba132006-10-11 01:20:53 -0700421 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
422 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500423 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700424 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700425 return -EPERM;
426 }
427
Yongqiang Yang2f919712011-07-27 21:16:33 -0400428 gdb_bh = sb_bread(sb, gdblock);
429 if (!gdb_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700430 return -EIO;
431
Yongqiang Yang2f919712011-07-27 21:16:33 -0400432 gdbackups = verify_reserved_gdb(sb, gdb_bh);
433 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700434 err = gdbackups;
435 goto exit_bh;
436 }
437
Mingming Cao617ba132006-10-11 01:20:53 -0700438 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700439 dind = sb_bread(sb, le32_to_cpu(*data));
440 if (!dind) {
441 err = -EIO;
442 goto exit_bh;
443 }
444
445 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700446 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500447 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Yongqiang Yang2f919712011-07-27 21:16:33 -0400448 group, gdblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700449 err = -EINVAL;
450 goto exit_dind;
451 }
452
Theodore Ts'ob4097142011-01-10 12:46:59 -0500453 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
454 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700455 goto exit_dind;
456
Yongqiang Yang2f919712011-07-27 21:16:33 -0400457 err = ext4_journal_get_write_access(handle, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500458 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700459 goto exit_sbh;
460
Theodore Ts'ob4097142011-01-10 12:46:59 -0500461 err = ext4_journal_get_write_access(handle, dind);
462 if (unlikely(err))
463 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700464
Mingming Cao617ba132006-10-11 01:20:53 -0700465 /* ext4_reserve_inode_write() gets a reference on the iloc */
Theodore Ts'ob4097142011-01-10 12:46:59 -0500466 err = ext4_reserve_inode_write(handle, inode, &iloc);
467 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468 goto exit_dindj;
469
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400470 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
471 sizeof(struct buffer_head *),
472 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700473 if (!n_group_desc) {
474 err = -ENOMEM;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400475 ext4_warning(sb, "not enough memory for %lu groups",
476 gdb_num + 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700477 goto exit_inode;
478 }
479
480 /*
481 * Finally, we have all of the possible failures behind us...
482 *
483 * Remove new GDT block from inode double-indirect block and clear out
484 * the new GDT block for use (which also "frees" the backup GDT blocks
485 * from the reserved inode). We don't need to change the bitmaps for
486 * these blocks, because they are marked as in-use from being in the
487 * reserved inode, and will become GDT blocks (primary and backup).
488 */
Mingming Cao617ba132006-10-11 01:20:53 -0700489 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500490 err = ext4_handle_dirty_metadata(handle, NULL, dind);
491 if (unlikely(err)) {
492 ext4_std_error(sb, err);
493 goto exit_inode;
494 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700496 ext4_mark_iloc_dirty(handle, inode, &iloc);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400497 memset(gdb_bh->b_data, 0, sb->s_blocksize);
498 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500499 if (unlikely(err)) {
500 ext4_std_error(sb, err);
501 goto exit_inode;
502 }
503 brelse(dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700504
Mingming Cao617ba132006-10-11 01:20:53 -0700505 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700506 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700507 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Yongqiang Yang2f919712011-07-27 21:16:33 -0400508 n_group_desc[gdb_num] = gdb_bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700509 EXT4_SB(sb)->s_group_desc = n_group_desc;
510 EXT4_SB(sb)->s_gdb_count++;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400511 ext4_kvfree(o_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700512
Marcin Slusarze8546d02008-04-17 10:38:59 -0400513 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500514 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
515 if (err)
516 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517
Theodore Ts'ob4097142011-01-10 12:46:59 -0500518 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519
520exit_inode:
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400521 ext4_kvfree(n_group_desc);
Amir Goldstein537a0312011-03-20 22:57:02 -0400522 /* ext4_handle_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700523 brelse(iloc.bh);
524exit_dindj:
Amir Goldstein537a0312011-03-20 22:57:02 -0400525 /* ext4_handle_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700526exit_sbh:
Amir Goldstein537a0312011-03-20 22:57:02 -0400527 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700528exit_dind:
529 brelse(dind);
530exit_bh:
Yongqiang Yang2f919712011-07-27 21:16:33 -0400531 brelse(gdb_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532
Mingming Cao617ba132006-10-11 01:20:53 -0700533 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700534 return err;
535}
536
537/*
538 * Called when we are adding a new group which has a backup copy of each of
539 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
540 * We need to add these reserved backup GDT blocks to the resize inode, so
541 * that they are kept for future resizing and not allocated to files.
542 *
543 * Each reserved backup GDT block will go into a different indirect block.
544 * The indirect blocks are actually the primary reserved GDT blocks,
545 * so we know in advance what their block numbers are. We only get the
546 * double-indirect block to verify it is pointing to the primary reserved
547 * GDT blocks so we don't overwrite a data block by accident. The reserved
548 * backup GDT blocks are stored in their reserved primary GDT block.
549 */
550static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400551 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700552{
553 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700554 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 struct buffer_head **primary;
556 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700557 struct ext4_iloc iloc;
558 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700559 __le32 *data, *end;
560 int gdbackups = 0;
561 int res, i;
562 int err;
563
Josef Bacik216553c2008-04-29 22:02:02 -0400564 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700565 if (!primary)
566 return -ENOMEM;
567
Mingming Cao617ba132006-10-11 01:20:53 -0700568 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569 dind = sb_bread(sb, le32_to_cpu(*data));
570 if (!dind) {
571 err = -EIO;
572 goto exit_free;
573 }
574
Mingming Cao617ba132006-10-11 01:20:53 -0700575 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400576 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
577 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700578 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579
580 /* Get each reserved primary GDT block and verify it holds backups */
581 for (res = 0; res < reserved_gdb; res++, blk++) {
582 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500583 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700584 " not at offset %ld",
585 blk,
586 (long)(data - (__le32 *)dind->b_data));
587 err = -EINVAL;
588 goto exit_bh;
589 }
590 primary[res] = sb_bread(sb, blk);
591 if (!primary[res]) {
592 err = -EIO;
593 goto exit_bh;
594 }
595 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
596 brelse(primary[res]);
597 err = gdbackups;
598 goto exit_bh;
599 }
600 if (++data >= end)
601 data = (__le32 *)dind->b_data;
602 }
603
604 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700605 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700606 /*
607 int j;
608 for (j = 0; j < i; j++)
Amir Goldstein537a0312011-03-20 22:57:02 -0400609 ext4_handle_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700610 */
611 goto exit_bh;
612 }
613 }
614
Mingming Cao617ba132006-10-11 01:20:53 -0700615 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700616 goto exit_bh;
617
618 /*
619 * Finally we can add each of the reserved backup GDT blocks from
620 * the new group to its reserved primary GDT block.
621 */
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400622 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623 for (i = 0; i < reserved_gdb; i++) {
624 int err2;
625 data = (__le32 *)primary[i]->b_data;
626 /* printk("reserving backup %lu[%u] = %lu\n",
627 primary[i]->b_blocknr, gdbackups,
628 blk + primary[i]->b_blocknr); */
629 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -0500630 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700631 if (!err)
632 err = err2;
633 }
634 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700635 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700636
637exit_bh:
638 while (--res >= 0)
639 brelse(primary[res]);
640 brelse(dind);
641
642exit_free:
643 kfree(primary);
644
645 return err;
646}
647
648/*
Mingming Cao617ba132006-10-11 01:20:53 -0700649 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700650 * of the main resize transaction, because e2fsck will re-write them if there
651 * is a problem (basically only OOM will cause a problem). However, we
652 * _should_ update the backups if possible, in case the primary gets trashed
653 * for some reason and we need to run e2fsck from a backup superblock. The
654 * important part is that the new block and inode counts are in the backup
655 * superblocks, and the location of the new group metadata in the GDT backups.
656 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400657 * We do not need take the s_resize_lock for this, because these
658 * blocks are not otherwise touched by the filesystem code when it is
659 * mounted. We don't need to worry about last changing from
660 * sbi->s_groups_count, because the worst that can happen is that we
661 * do not copy the full number of backups at this time. The resize
662 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663 */
664static void update_backups(struct super_block *sb,
665 int blk_off, char *data, int size)
666{
Mingming Cao617ba132006-10-11 01:20:53 -0700667 struct ext4_sb_info *sbi = EXT4_SB(sb);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500668 const ext4_group_t last = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700669 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700670 unsigned three = 1;
671 unsigned five = 5;
672 unsigned seven = 7;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500673 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674 int rest = sb->s_blocksize - size;
675 handle_t *handle;
676 int err = 0, err2;
677
Mingming Cao617ba132006-10-11 01:20:53 -0700678 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700679 if (IS_ERR(handle)) {
680 group = 1;
681 err = PTR_ERR(handle);
682 goto exit_err;
683 }
684
Mingming Cao617ba132006-10-11 01:20:53 -0700685 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686 struct buffer_head *bh;
687
688 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -0500689 if (ext4_handle_valid(handle) &&
690 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -0700691 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
692 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700693 break;
694
695 bh = sb_getblk(sb, group * bpg + blk_off);
696 if (!bh) {
697 err = -EIO;
698 break;
699 }
Mingming Cao617ba132006-10-11 01:20:53 -0700700 ext4_debug("update metadata backup %#04lx\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700701 (unsigned long)bh->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -0700702 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700703 break;
704 lock_buffer(bh);
705 memcpy(bh->b_data, data, size);
706 if (rest)
707 memset(bh->b_data + size, 0, rest);
708 set_buffer_uptodate(bh);
709 unlock_buffer(bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500710 err = ext4_handle_dirty_metadata(handle, NULL, bh);
711 if (unlikely(err))
712 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713 brelse(bh);
714 }
Mingming Cao617ba132006-10-11 01:20:53 -0700715 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700716 err = err2;
717
718 /*
719 * Ugh! Need to have e2fsck write the backup copies. It is too
720 * late to revert the resize, we shouldn't fail just because of
721 * the backup copies (they are only needed in case of corruption).
722 *
723 * However, if we got here we have a journal problem too, so we
724 * can't really start a transaction to mark the superblock.
725 * Chicken out and just set the flag on the hope it will be written
726 * to disk, and if not - we will simply wait until next fsck.
727 */
728exit_err:
729 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500730 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -0700732 sbi->s_mount_state &= ~EXT4_VALID_FS;
733 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700734 mark_buffer_dirty(sbi->s_sbh);
735 }
736}
737
738/* Add group descriptor data to an existing or new group descriptor block.
739 * Ensure we handle all possible error conditions _before_ we start modifying
740 * the filesystem, because we cannot abort the transaction and not have it
741 * write the data to disk.
742 *
743 * If we are on a GDT block boundary, we need to get the reserved GDT block.
744 * Otherwise, we may need to add backup GDT blocks for a sparse group.
745 *
746 * We only need to hold the superblock lock while we are actually adding
747 * in the new group's counts to the superblock. Prior to that we have
748 * not really "added" the group at all. We re-check that we are still
749 * adding in the last group in case things have changed since verifying.
750 */
Mingming Cao617ba132006-10-11 01:20:53 -0700751int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752{
Mingming Cao617ba132006-10-11 01:20:53 -0700753 struct ext4_sb_info *sbi = EXT4_SB(sb);
754 struct ext4_super_block *es = sbi->s_es;
755 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
757 struct buffer_head *primary = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700758 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759 struct inode *inode = NULL;
760 handle_t *handle;
761 int gdb_off, gdb_num;
762 int err, err2;
763
Mingming Cao617ba132006-10-11 01:20:53 -0700764 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
765 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700766
Mingming Cao617ba132006-10-11 01:20:53 -0700767 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
768 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500769 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700770 return -EPERM;
771 }
772
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700773 if (ext4_blocks_count(es) + input->blocks_count <
774 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500775 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776 return -EINVAL;
777 }
778
Mingming Cao617ba132006-10-11 01:20:53 -0700779 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700780 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500781 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 return -EINVAL;
783 }
784
785 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700786 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -0400787 EXT4_FEATURE_COMPAT_RESIZE_INODE)
788 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500789 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 "No reserved GDT blocks, can't resize");
791 return -EPERM;
792 }
David Howells1d1fe1e2008-02-07 00:15:37 -0800793 inode = ext4_iget(sb, EXT4_RESIZE_INO);
794 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500795 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -0800796 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700797 }
798 }
799
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500800
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700801 if ((err = verify_group_input(sb, input)))
802 goto exit_put;
803
804 if ((err = setup_new_group_blocks(sb, input)))
805 goto exit_put;
806
807 /*
808 * We will always be modifying at least the superblock and a GDT
809 * block. If we are adding a group past the last current GDT block,
810 * we will also modify the inode and the dindirect block. If we
811 * are adding a group with superblock/GDT backups we will also
812 * modify each of the reserved GDT dindirect blocks.
813 */
Mingming Cao617ba132006-10-11 01:20:53 -0700814 handle = ext4_journal_start_sb(sb,
815 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700816 3 + reserved_gdb : 4);
817 if (IS_ERR(handle)) {
818 err = PTR_ERR(handle);
819 goto exit_put;
820 }
821
Mingming Cao617ba132006-10-11 01:20:53 -0700822 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700823 goto exit_journal;
824
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400825 /*
826 * We will only either add reserved group blocks to a backup group
827 * or remove reserved blocks for the first group in a new group block.
828 * Doing both would be mean more complex code, and sane people don't
829 * use non-sparse filesystems anymore. This is already checked above.
830 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 if (gdb_off) {
832 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -0700833 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700834 goto exit_journal;
835
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400836 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
837 err = reserve_backup_gdb(handle, inode, input->group);
838 if (err)
839 goto exit_journal;
840 }
Yongqiang Yang2f919712011-07-27 21:16:33 -0400841 } else {
842 /*
843 * Note that we can access new group descriptor block safely
844 * only if add_new_gdb() succeeds.
845 */
846 err = add_new_gdb(handle, inode, input->group);
847 if (err)
848 goto exit_journal;
849 primary = sbi->s_group_desc[gdb_num];
850 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700851
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400852 /*
853 * OK, now we've set up the new group. Time to make it active.
854 *
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400855 * so we have to be safe wrt. concurrent accesses the group
856 * data. So we need to be careful to set all of the relevant
857 * group descriptor data etc. *before* we enable the group.
858 *
859 * The key field here is sbi->s_groups_count: as long as
860 * that retains its old value, nobody is going to access the new
861 * group.
862 *
863 * So first we update all the descriptor metadata for the new
864 * group; then we update the total disk blocks count; then we
865 * update the groups count to enable the group; then finally we
866 * update the free space counts so that the system can start
867 * using the new disk blocks.
868 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869
870 /* Update group descriptor block for new group */
Frederic Bohe28569222008-06-20 11:48:48 -0400871 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
872 gdb_off * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700873
Theodore Ts'ofdff73f2009-01-26 19:06:41 -0500874 memset(gdp, 0, EXT4_DESC_SIZE(sb));
Alexandre Ratchov8fadc142006-10-11 01:21:15 -0700875 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
876 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
877 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500878 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
879 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'ofdff73f2009-01-26 19:06:41 -0500880 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400881 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700882
883 /*
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400884 * We can allocate memory for mb_alloc based on the new group
885 * descriptor
886 */
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500887 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -0400888 if (err)
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400889 goto exit_journal;
890
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400891 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700892 * Make the new blocks and inodes valid next. We do this before
893 * increasing the group count so that once the group is enabled,
894 * all of its blocks and inodes are already valid.
895 *
896 * We always allocate group-by-group, then block-by-block or
897 * inode-by-inode within a group, so enabling these
898 * blocks/inodes before the group is live won't actually let us
899 * allocate the new space yet.
900 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700901 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700902 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400903 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700904
905 /*
906 * We need to protect s_groups_count against other CPUs seeing
907 * inconsistent state in the superblock.
908 *
909 * The precise rules we use are:
910 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 * * Writers must perform a smp_wmb() after updating all dependent
912 * data and before modifying the groups count
913 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700914 * * Readers must perform an smp_rmb() after reading the groups count
915 * and before reading any dependent data.
916 *
917 * NB. These rules can be relaxed when checking the group count
918 * while freeing data, as we can only allocate from a block
919 * group after serialising against the group count, and we can
920 * only then free after serialising in turn against that
921 * allocation.
922 */
923 smp_wmb();
924
925 /* Update the global fs size fields */
926 sbi->s_groups_count++;
927
Theodore Ts'ob4097142011-01-10 12:46:59 -0500928 err = ext4_handle_dirty_metadata(handle, NULL, primary);
929 if (unlikely(err)) {
930 ext4_std_error(sb, err);
931 goto exit_journal;
932 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700933
934 /* Update the reserved block counts only once the new group is
935 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700936 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700937 input->reserved_blocks);
938
939 /* Update the free space counts */
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700940 percpu_counter_add(&sbi->s_freeblocks_counter,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700941 input->free_blocks_count);
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700942 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -0700943 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700944
Eric Sandeen42007ef2010-05-16 01:00:00 -0400945 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
946 sbi->s_log_groups_per_flex) {
Frederic Bohec62a11f2008-09-08 10:20:24 -0400947 ext4_group_t flex_group;
948 flex_group = ext4_flex_group(sbi, input->group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500949 atomic_add(input->free_blocks_count,
950 &sbi->s_flex_groups[flex_group].free_blocks);
951 atomic_add(EXT4_INODES_PER_GROUP(sb),
952 &sbi->s_flex_groups[flex_group].free_inodes);
Frederic Bohec62a11f2008-09-08 10:20:24 -0400953 }
954
Theodore Ts'oa0375152010-06-11 23:14:04 -0400955 ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700956
957exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -0700958 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700959 err = err2;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400960 if (!err && primary) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700961 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -0700962 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963 update_backups(sb, primary->b_blocknr, primary->b_data,
964 primary->b_size);
965 }
966exit_put:
967 iput(inode);
968 return err;
Mingming Cao617ba132006-10-11 01:20:53 -0700969} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700970
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400971/*
972 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973 * point is only used to extend the current filesystem to the end of the last
974 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
975 * for emergencies (because it has no dependencies on reserved blocks).
976 *
Mingming Cao617ba132006-10-11 01:20:53 -0700977 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 * allow the "remount" trick to work for arbitrary resizing, assuming enough
979 * GDT blocks are reserved to grow to the desired size.
980 */
Mingming Cao617ba132006-10-11 01:20:53 -0700981int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
982 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700983{
Mingming Cao617ba132006-10-11 01:20:53 -0700984 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700985 ext4_grpblk_t last;
986 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400987 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700988 handle_t *handle;
Yongqiang Yangcc7365d2011-07-26 21:46:07 -0400989 int err, err2;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400990 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700991
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700992 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993
994 if (test_opt(sb, DEBUG))
Yongqiang Yang2b79b092011-07-26 21:53:35 -0400995 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700996 o_blocks_count, n_blocks_count);
997
998 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
999 return 0;
1000
1001 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001002 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -07001003 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001004 sb->s_id, n_blocks_count);
1005 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001006 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007 return -EINVAL;
1008 }
1009
1010 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001011 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001012 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013 }
1014
1015 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001016 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017
1018 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001019 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020 return -EPERM;
1021 }
1022
Mingming Cao617ba132006-10-11 01:20:53 -07001023 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024
1025 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001026 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 return -EINVAL;
1028 }
1029
1030 if (o_blocks_count + add > n_blocks_count)
1031 add = n_blocks_count - o_blocks_count;
1032
1033 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001034 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001035 o_blocks_count + add, add);
1036
1037 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001038 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001040 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001041 return -ENOSPC;
1042 }
1043 brelse(bh);
1044
1045 /* We will update the superblock, one block bitmap, and
Mingming Cao617ba132006-10-11 01:20:53 -07001046 * one group descriptor via ext4_free_blocks().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 */
Mingming Cao617ba132006-10-11 01:20:53 -07001048 handle = ext4_journal_start_sb(sb, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001049 if (IS_ERR(handle)) {
1050 err = PTR_ERR(handle);
Eric Sandeen12062dd2010-02-15 14:19:27 -05001051 ext4_warning(sb, "error %d on journal start", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052 goto exit_put;
1053 }
1054
Mingming Cao617ba132006-10-11 01:20:53 -07001055 if ((err = ext4_journal_get_write_access(handle,
1056 EXT4_SB(sb)->s_sbh))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001057 ext4_warning(sb, "error %d on journal write access", err);
Mingming Cao617ba132006-10-11 01:20:53 -07001058 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059 goto exit_put;
1060 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001061 ext4_blocks_count_set(es, o_blocks_count + add);
Eric Sandeenc549a952008-01-28 23:58:27 -05001062 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063 o_blocks_count + add);
Aneesh Kumar K.Ve21675d2009-01-05 21:36:02 -05001064 /* We add the blocks to the bitmap and set the group need init bit */
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001065 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
Theodore Ts'oa0375152010-06-11 23:14:04 -04001066 ext4_handle_dirty_super(handle, sb);
Mingming Cao2ae02102006-10-11 01:21:11 -07001067 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001068 o_blocks_count + add);
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001069 err2 = ext4_journal_stop(handle);
1070 if (!err && err2)
1071 err = err2;
1072
1073 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 goto exit_put;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001075
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001076 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001077 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1078 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001079 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1080 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001081exit_put:
1082 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001083} /* ext4_group_extend */