blob: ca5c8aa00a2fe10a621348913a0c85908c6e7f6e [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
19#define outside(b, first, last) ((b) < (first) || (b) >= (last))
20#define inside(b, first, last) ((b) >= (first) && (b) < (last))
21
22static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070023 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070024{
Mingming Cao617ba132006-10-11 01:20:53 -070025 struct ext4_sb_info *sbi = EXT4_SB(sb);
26 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070027 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070028 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050029 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070030 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
31 unsigned overhead = ext4_bg_has_super(sb, group) ?
32 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070034 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070035 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070036 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070037 int err = -EINVAL;
38
39 input->free_blocks_count = free_blocks_count =
40 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
41
42 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070043 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070044 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070045 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070046 "no-super", input->group, input->blocks_count,
47 free_blocks_count, input->reserved_blocks);
48
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070049 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070050 if (group != sbi->s_groups_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -050051 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070052 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070053 else if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050054 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070055 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -050056 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070057 input->reserved_blocks);
58 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050059 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070060 input->blocks_count);
61 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -050062 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070063 end - 1);
64 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050065 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070066 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070067 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050068 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070069 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070070 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -040071 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050072 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070073 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070074 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -050075 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070076 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -050078 ext4_warning(sb, "Block bitmap (%llu) in inode table "
79 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070080 (unsigned long long)input->block_bitmap,
81 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070082 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -050083 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
84 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070085 (unsigned long long)input->inode_bitmap,
86 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070087 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -050088 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070089 (unsigned long long)input->block_bitmap,
90 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070091 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -050092 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070093 (unsigned long long)input->inode_bitmap,
94 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -040096 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -050097 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
98 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070099 (unsigned long long)input->inode_table,
100 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700101 else
102 err = 0;
103 brelse(bh);
104
105 return err;
106}
107
108static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700109 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700110{
111 struct buffer_head *bh;
112 int err;
113
114 bh = sb_getblk(sb, blk);
115 if (!bh)
116 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700117 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700118 brelse(bh);
119 bh = ERR_PTR(err);
120 } else {
121 lock_buffer(bh);
122 memset(bh->b_data, 0, sb->s_blocksize);
123 set_buffer_uptodate(bh);
124 unlock_buffer(bh);
125 }
126
127 return bh;
128}
129
130/*
Eric Sandeen14904102007-10-16 18:38:25 -0400131 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
132 * If that fails, restart the transaction & regain write access for the
133 * buffer head which is used for block_bitmap modifications.
134 */
135static int extend_or_restart_transaction(handle_t *handle, int thresh,
136 struct buffer_head *bh)
137{
138 int err;
139
Frank Mayhar03901312009-01-07 00:06:22 -0500140 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400141 return 0;
142
143 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
144 if (err < 0)
145 return err;
146 if (err) {
147 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
148 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400149 if ((err = ext4_journal_get_write_access(handle, bh)))
Eric Sandeen14904102007-10-16 18:38:25 -0400150 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400151 }
Eric Sandeen14904102007-10-16 18:38:25 -0400152
153 return 0;
154}
155
156/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700157 * Set up the block and inode bitmaps, and the inode table for the new group.
158 * This doesn't need to be part of the main transaction, since we are only
159 * changing blocks outside the actual filesystem. We still do journaling to
160 * ensure the recovery is correct in case of a failure just after resize.
161 * If any part of this fails, we simply abort the resize.
162 */
163static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700164 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700165{
Mingming Cao617ba132006-10-11 01:20:53 -0700166 struct ext4_sb_info *sbi = EXT4_SB(sb);
167 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
168 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700169 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700170 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171 struct buffer_head *bh;
172 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700173 ext4_fsblk_t block;
174 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700175 int i;
176 int err = 0, err2;
177
Eric Sandeen14904102007-10-16 18:38:25 -0400178 /* This transaction may be extended/restarted along the way */
179 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
180
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700181 if (IS_ERR(handle))
182 return PTR_ERR(handle);
183
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400184 mutex_lock(&sbi->s_resize_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185 if (input->group != sbi->s_groups_count) {
186 err = -EBUSY;
187 goto exit_journal;
188 }
189
190 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
191 err = PTR_ERR(bh);
192 goto exit_journal;
193 }
194
Mingming Cao617ba132006-10-11 01:20:53 -0700195 if (ext4_bg_has_super(sb, input->group)) {
Eric Sandeenc549a952008-01-28 23:58:27 -0500196 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
Mingming Cao617ba132006-10-11 01:20:53 -0700197 ext4_set_bit(0, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700198 }
199
200 /* Copy all of the GDT blocks into the backup in this group */
201 for (i = 0, bit = 1, block = start + 1;
202 i < gdblocks; i++, block++, bit++) {
203 struct buffer_head *gdb;
204
Eric Sandeenc549a952008-01-28 23:58:27 -0500205 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700206
Eric Sandeen14904102007-10-16 18:38:25 -0400207 if ((err = extend_or_restart_transaction(handle, 1, bh)))
208 goto exit_bh;
209
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 gdb = sb_getblk(sb, block);
211 if (!gdb) {
212 err = -EIO;
213 goto exit_bh;
214 }
Mingming Cao617ba132006-10-11 01:20:53 -0700215 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700216 brelse(gdb);
217 goto exit_bh;
218 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400219 lock_buffer(gdb);
220 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700221 set_buffer_uptodate(gdb);
Eric Sandeen5b615282007-10-16 18:38:25 -0400222 unlock_buffer(gdb);
Frank Mayhar03901312009-01-07 00:06:22 -0500223 ext4_handle_dirty_metadata(handle, NULL, gdb);
Mingming Cao617ba132006-10-11 01:20:53 -0700224 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700225 brelse(gdb);
226 }
227
228 /* Zero out all of the reserved backup group descriptor table blocks */
229 for (i = 0, bit = gdblocks + 1, block = start + bit;
230 i < reserved_gdb; i++, block++, bit++) {
231 struct buffer_head *gdb;
232
Eric Sandeenc549a952008-01-28 23:58:27 -0500233 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234
Eric Sandeen14904102007-10-16 18:38:25 -0400235 if ((err = extend_or_restart_transaction(handle, 1, bh)))
236 goto exit_bh;
237
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700238 if (IS_ERR(gdb = bclean(handle, sb, block))) {
Roel Kluinc09eef32009-12-07 10:38:16 -0500239 err = PTR_ERR(gdb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700240 goto exit_bh;
241 }
Frank Mayhar03901312009-01-07 00:06:22 -0500242 ext4_handle_dirty_metadata(handle, NULL, gdb);
Mingming Cao617ba132006-10-11 01:20:53 -0700243 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244 brelse(gdb);
245 }
Eric Sandeenc549a952008-01-28 23:58:27 -0500246 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700247 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700248 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500249 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700251 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700252
253 /* Zero out all of the inode table blocks */
254 for (i = 0, block = input->inode_table, bit = block - start;
255 i < sbi->s_itb_per_group; i++, bit++, block++) {
256 struct buffer_head *it;
257
Eric Sandeenc549a952008-01-28 23:58:27 -0500258 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit);
Eric Sandeen14904102007-10-16 18:38:25 -0400259
260 if ((err = extend_or_restart_transaction(handle, 1, bh)))
261 goto exit_bh;
262
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263 if (IS_ERR(it = bclean(handle, sb, block))) {
264 err = PTR_ERR(it);
265 goto exit_bh;
266 }
Frank Mayhar03901312009-01-07 00:06:22 -0500267 ext4_handle_dirty_metadata(handle, NULL, it);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700268 brelse(it);
Mingming Cao617ba132006-10-11 01:20:53 -0700269 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700270 }
Eric Sandeen14904102007-10-16 18:38:25 -0400271
272 if ((err = extend_or_restart_transaction(handle, 2, bh)))
273 goto exit_bh;
274
Aneesh Kumar K.V648f5872009-01-05 21:46:04 -0500275 mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, bh->b_data);
Frank Mayhar03901312009-01-07 00:06:22 -0500276 ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700277 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700278 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500279 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700280 input->inode_bitmap, input->inode_bitmap - start);
281 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
282 err = PTR_ERR(bh);
283 goto exit_journal;
284 }
285
Aneesh Kumar K.V648f5872009-01-05 21:46:04 -0500286 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700287 bh->b_data);
Frank Mayhar03901312009-01-07 00:06:22 -0500288 ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289exit_bh:
290 brelse(bh);
291
292exit_journal:
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400293 mutex_unlock(&sbi->s_resize_lock);
Mingming Cao617ba132006-10-11 01:20:53 -0700294 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 err = err2;
296
297 return err;
298}
299
300/*
301 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700302 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303 * calling this for the first time. In a sparse filesystem it will be the
304 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
305 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
306 */
Mingming Cao617ba132006-10-11 01:20:53 -0700307static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700308 unsigned *five, unsigned *seven)
309{
310 unsigned *min = three;
311 int mult = 3;
312 unsigned ret;
313
Mingming Cao617ba132006-10-11 01:20:53 -0700314 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
315 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316 ret = *min;
317 *min += 1;
318 return ret;
319 }
320
321 if (*five < *min) {
322 min = five;
323 mult = 5;
324 }
325 if (*seven < *min) {
326 min = seven;
327 mult = 7;
328 }
329
330 ret = *min;
331 *min *= mult;
332
333 return ret;
334}
335
336/*
337 * Check that all of the backup GDT blocks are held in the primary GDT block.
338 * It is assumed that they are stored in group order. Returns the number of
339 * groups in current filesystem that have BACKUPS, or -ve error code.
340 */
341static int verify_reserved_gdb(struct super_block *sb,
342 struct buffer_head *primary)
343{
Mingming Cao617ba132006-10-11 01:20:53 -0700344 const ext4_fsblk_t blk = primary->b_blocknr;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500345 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700346 unsigned three = 1;
347 unsigned five = 5;
348 unsigned seven = 7;
349 unsigned grp;
350 __le32 *p = (__le32 *)primary->b_data;
351 int gdbackups = 0;
352
Mingming Cao617ba132006-10-11 01:20:53 -0700353 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700354 if (le32_to_cpu(*p++) !=
355 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500356 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700357 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700358 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700359 grp *
360 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
361 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362 return -EINVAL;
363 }
Mingming Cao617ba132006-10-11 01:20:53 -0700364 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700365 return -EFBIG;
366 }
367
368 return gdbackups;
369}
370
371/*
372 * Called when we need to bring a reserved group descriptor table block into
373 * use from the resize inode. The primary copy of the new GDT block currently
374 * is an indirect block (under the double indirect block in the resize inode).
375 * The new backup GDT blocks will be stored as leaf blocks in this indirect
376 * block, in group order. Even though we know all the block numbers we need,
377 * we check to ensure that the resize inode has actually reserved these blocks.
378 *
379 * Don't need to update the block bitmaps because the blocks are still in use.
380 *
381 * We get all of the error cases out of the way, so that we are sure to not
382 * fail once we start modifying the data on disk, because JBD has no rollback.
383 */
384static int add_new_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700385 struct ext4_new_group_data *input,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700386 struct buffer_head **primary)
387{
388 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700389 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
390 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
391 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392 struct buffer_head **o_group_desc, **n_group_desc;
393 struct buffer_head *dind;
394 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700395 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700396 __le32 *data;
397 int err;
398
399 if (test_opt(sb, DEBUG))
400 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700401 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700402 gdb_num);
403
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400404 /*
405 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400406 * because the user tools have no way of handling this. Probably a
407 * bad time to do it anyways.
408 */
Mingming Cao617ba132006-10-11 01:20:53 -0700409 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
410 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500411 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700412 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700413 return -EPERM;
414 }
415
416 *primary = sb_bread(sb, gdblock);
417 if (!*primary)
418 return -EIO;
419
420 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
421 err = gdbackups;
422 goto exit_bh;
423 }
424
Mingming Cao617ba132006-10-11 01:20:53 -0700425 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700426 dind = sb_bread(sb, le32_to_cpu(*data));
427 if (!dind) {
428 err = -EIO;
429 goto exit_bh;
430 }
431
432 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700433 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500434 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700435 input->group, gdblock);
436 err = -EINVAL;
437 goto exit_dind;
438 }
439
Mingming Cao617ba132006-10-11 01:20:53 -0700440 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 goto exit_dind;
442
Mingming Cao617ba132006-10-11 01:20:53 -0700443 if ((err = ext4_journal_get_write_access(handle, *primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700444 goto exit_sbh;
445
Mingming Cao617ba132006-10-11 01:20:53 -0700446 if ((err = ext4_journal_get_write_access(handle, dind)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700447 goto exit_primary;
448
Mingming Cao617ba132006-10-11 01:20:53 -0700449 /* ext4_reserve_inode_write() gets a reference on the iloc */
450 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451 goto exit_dindj;
452
453 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
Josef Bacik216553c2008-04-29 22:02:02 -0400454 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700455 if (!n_group_desc) {
456 err = -ENOMEM;
Eric Sandeen12062dd2010-02-15 14:19:27 -0500457 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700458 "not enough memory for %lu groups", gdb_num + 1);
459 goto exit_inode;
460 }
461
462 /*
463 * Finally, we have all of the possible failures behind us...
464 *
465 * Remove new GDT block from inode double-indirect block and clear out
466 * the new GDT block for use (which also "frees" the backup GDT blocks
467 * from the reserved inode). We don't need to change the bitmaps for
468 * these blocks, because they are marked as in-use from being in the
469 * reserved inode, and will become GDT blocks (primary and backup).
470 */
Mingming Cao617ba132006-10-11 01:20:53 -0700471 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Frank Mayhar03901312009-01-07 00:06:22 -0500472 ext4_handle_dirty_metadata(handle, NULL, dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700473 brelse(dind);
474 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700475 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700476 memset((*primary)->b_data, 0, sb->s_blocksize);
Frank Mayhar03901312009-01-07 00:06:22 -0500477 ext4_handle_dirty_metadata(handle, NULL, *primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700478
Mingming Cao617ba132006-10-11 01:20:53 -0700479 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700480 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700481 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482 n_group_desc[gdb_num] = *primary;
Mingming Cao617ba132006-10-11 01:20:53 -0700483 EXT4_SB(sb)->s_group_desc = n_group_desc;
484 EXT4_SB(sb)->s_gdb_count++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485 kfree(o_group_desc);
486
Marcin Slusarze8546d02008-04-17 10:38:59 -0400487 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Frank Mayhar03901312009-01-07 00:06:22 -0500488 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489
490 return 0;
491
492exit_inode:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400493 /* ext4_journal_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 brelse(iloc.bh);
495exit_dindj:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400496 /* ext4_journal_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700497exit_primary:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400498 /* ext4_journal_release_buffer(handle, *primary); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700499exit_sbh:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400500 /* ext4_journal_release_buffer(handle, *primary); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700501exit_dind:
502 brelse(dind);
503exit_bh:
504 brelse(*primary);
505
Mingming Cao617ba132006-10-11 01:20:53 -0700506 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700507 return err;
508}
509
510/*
511 * Called when we are adding a new group which has a backup copy of each of
512 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
513 * We need to add these reserved backup GDT blocks to the resize inode, so
514 * that they are kept for future resizing and not allocated to files.
515 *
516 * Each reserved backup GDT block will go into a different indirect block.
517 * The indirect blocks are actually the primary reserved GDT blocks,
518 * so we know in advance what their block numbers are. We only get the
519 * double-indirect block to verify it is pointing to the primary reserved
520 * GDT blocks so we don't overwrite a data block by accident. The reserved
521 * backup GDT blocks are stored in their reserved primary GDT block.
522 */
523static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700524 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525{
526 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700527 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700528 struct buffer_head **primary;
529 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700530 struct ext4_iloc iloc;
531 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700532 __le32 *data, *end;
533 int gdbackups = 0;
534 int res, i;
535 int err;
536
Josef Bacik216553c2008-04-29 22:02:02 -0400537 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700538 if (!primary)
539 return -ENOMEM;
540
Mingming Cao617ba132006-10-11 01:20:53 -0700541 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700542 dind = sb_bread(sb, le32_to_cpu(*data));
543 if (!dind) {
544 err = -EIO;
545 goto exit_free;
546 }
547
Mingming Cao617ba132006-10-11 01:20:53 -0700548 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400549 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
550 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700551 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700552
553 /* Get each reserved primary GDT block and verify it holds backups */
554 for (res = 0; res < reserved_gdb; res++, blk++) {
555 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500556 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700557 " not at offset %ld",
558 blk,
559 (long)(data - (__le32 *)dind->b_data));
560 err = -EINVAL;
561 goto exit_bh;
562 }
563 primary[res] = sb_bread(sb, blk);
564 if (!primary[res]) {
565 err = -EIO;
566 goto exit_bh;
567 }
568 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
569 brelse(primary[res]);
570 err = gdbackups;
571 goto exit_bh;
572 }
573 if (++data >= end)
574 data = (__le32 *)dind->b_data;
575 }
576
577 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700578 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579 /*
580 int j;
581 for (j = 0; j < i; j++)
Mingming Cao617ba132006-10-11 01:20:53 -0700582 ext4_journal_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700583 */
584 goto exit_bh;
585 }
586 }
587
Mingming Cao617ba132006-10-11 01:20:53 -0700588 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700589 goto exit_bh;
590
591 /*
592 * Finally we can add each of the reserved backup GDT blocks from
593 * the new group to its reserved primary GDT block.
594 */
Mingming Cao617ba132006-10-11 01:20:53 -0700595 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596 for (i = 0; i < reserved_gdb; i++) {
597 int err2;
598 data = (__le32 *)primary[i]->b_data;
599 /* printk("reserving backup %lu[%u] = %lu\n",
600 primary[i]->b_blocknr, gdbackups,
601 blk + primary[i]->b_blocknr); */
602 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -0500603 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604 if (!err)
605 err = err2;
606 }
607 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700608 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700609
610exit_bh:
611 while (--res >= 0)
612 brelse(primary[res]);
613 brelse(dind);
614
615exit_free:
616 kfree(primary);
617
618 return err;
619}
620
621/*
Mingming Cao617ba132006-10-11 01:20:53 -0700622 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700623 * of the main resize transaction, because e2fsck will re-write them if there
624 * is a problem (basically only OOM will cause a problem). However, we
625 * _should_ update the backups if possible, in case the primary gets trashed
626 * for some reason and we need to run e2fsck from a backup superblock. The
627 * important part is that the new block and inode counts are in the backup
628 * superblocks, and the location of the new group metadata in the GDT backups.
629 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400630 * We do not need take the s_resize_lock for this, because these
631 * blocks are not otherwise touched by the filesystem code when it is
632 * mounted. We don't need to worry about last changing from
633 * sbi->s_groups_count, because the worst that can happen is that we
634 * do not copy the full number of backups at this time. The resize
635 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700636 */
637static void update_backups(struct super_block *sb,
638 int blk_off, char *data, int size)
639{
Mingming Cao617ba132006-10-11 01:20:53 -0700640 struct ext4_sb_info *sbi = EXT4_SB(sb);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500641 const ext4_group_t last = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700642 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700643 unsigned three = 1;
644 unsigned five = 5;
645 unsigned seven = 7;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500646 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700647 int rest = sb->s_blocksize - size;
648 handle_t *handle;
649 int err = 0, err2;
650
Mingming Cao617ba132006-10-11 01:20:53 -0700651 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700652 if (IS_ERR(handle)) {
653 group = 1;
654 err = PTR_ERR(handle);
655 goto exit_err;
656 }
657
Mingming Cao617ba132006-10-11 01:20:53 -0700658 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700659 struct buffer_head *bh;
660
661 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -0500662 if (ext4_handle_valid(handle) &&
663 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -0700664 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
665 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700666 break;
667
668 bh = sb_getblk(sb, group * bpg + blk_off);
669 if (!bh) {
670 err = -EIO;
671 break;
672 }
Mingming Cao617ba132006-10-11 01:20:53 -0700673 ext4_debug("update metadata backup %#04lx\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700674 (unsigned long)bh->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -0700675 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700676 break;
677 lock_buffer(bh);
678 memcpy(bh->b_data, data, size);
679 if (rest)
680 memset(bh->b_data + size, 0, rest);
681 set_buffer_uptodate(bh);
682 unlock_buffer(bh);
Frank Mayhar03901312009-01-07 00:06:22 -0500683 ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700684 brelse(bh);
685 }
Mingming Cao617ba132006-10-11 01:20:53 -0700686 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700687 err = err2;
688
689 /*
690 * Ugh! Need to have e2fsck write the backup copies. It is too
691 * late to revert the resize, we shouldn't fail just because of
692 * the backup copies (they are only needed in case of corruption).
693 *
694 * However, if we got here we have a journal problem too, so we
695 * can't really start a transaction to mark the superblock.
696 * Chicken out and just set the flag on the hope it will be written
697 * to disk, and if not - we will simply wait until next fsck.
698 */
699exit_err:
700 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500701 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700702 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -0700703 sbi->s_mount_state &= ~EXT4_VALID_FS;
704 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700705 mark_buffer_dirty(sbi->s_sbh);
706 }
707}
708
709/* Add group descriptor data to an existing or new group descriptor block.
710 * Ensure we handle all possible error conditions _before_ we start modifying
711 * the filesystem, because we cannot abort the transaction and not have it
712 * write the data to disk.
713 *
714 * If we are on a GDT block boundary, we need to get the reserved GDT block.
715 * Otherwise, we may need to add backup GDT blocks for a sparse group.
716 *
717 * We only need to hold the superblock lock while we are actually adding
718 * in the new group's counts to the superblock. Prior to that we have
719 * not really "added" the group at all. We re-check that we are still
720 * adding in the last group in case things have changed since verifying.
721 */
Mingming Cao617ba132006-10-11 01:20:53 -0700722int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723{
Mingming Cao617ba132006-10-11 01:20:53 -0700724 struct ext4_sb_info *sbi = EXT4_SB(sb);
725 struct ext4_super_block *es = sbi->s_es;
726 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700727 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
728 struct buffer_head *primary = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700729 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730 struct inode *inode = NULL;
731 handle_t *handle;
732 int gdb_off, gdb_num;
733 int err, err2;
734
Mingming Cao617ba132006-10-11 01:20:53 -0700735 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
736 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700737
Mingming Cao617ba132006-10-11 01:20:53 -0700738 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
739 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500740 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700741 return -EPERM;
742 }
743
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700744 if (ext4_blocks_count(es) + input->blocks_count <
745 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500746 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700747 return -EINVAL;
748 }
749
Mingming Cao617ba132006-10-11 01:20:53 -0700750 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500752 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700753 return -EINVAL;
754 }
755
756 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700757 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -0400758 EXT4_FEATURE_COMPAT_RESIZE_INODE)
759 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500760 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700761 "No reserved GDT blocks, can't resize");
762 return -EPERM;
763 }
David Howells1d1fe1e2008-02-07 00:15:37 -0800764 inode = ext4_iget(sb, EXT4_RESIZE_INO);
765 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500766 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -0800767 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700768 }
769 }
770
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500771
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700772 if ((err = verify_group_input(sb, input)))
773 goto exit_put;
774
775 if ((err = setup_new_group_blocks(sb, input)))
776 goto exit_put;
777
778 /*
779 * We will always be modifying at least the superblock and a GDT
780 * block. If we are adding a group past the last current GDT block,
781 * we will also modify the inode and the dindirect block. If we
782 * are adding a group with superblock/GDT backups we will also
783 * modify each of the reserved GDT dindirect blocks.
784 */
Mingming Cao617ba132006-10-11 01:20:53 -0700785 handle = ext4_journal_start_sb(sb,
786 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787 3 + reserved_gdb : 4);
788 if (IS_ERR(handle)) {
789 err = PTR_ERR(handle);
790 goto exit_put;
791 }
792
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400793 mutex_lock(&sbi->s_resize_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700794 if (input->group != sbi->s_groups_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500795 ext4_warning(sb, "multiple resizers run on filesystem!");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700796 err = -EBUSY;
797 goto exit_journal;
798 }
799
Mingming Cao617ba132006-10-11 01:20:53 -0700800 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700801 goto exit_journal;
802
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400803 /*
804 * We will only either add reserved group blocks to a backup group
805 * or remove reserved blocks for the first group in a new group block.
806 * Doing both would be mean more complex code, and sane people don't
807 * use non-sparse filesystems anymore. This is already checked above.
808 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809 if (gdb_off) {
810 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -0700811 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812 goto exit_journal;
813
Mingming Cao617ba132006-10-11 01:20:53 -0700814 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815 (err = reserve_backup_gdb(handle, inode, input)))
816 goto exit_journal;
817 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
818 goto exit_journal;
819
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400820 /*
821 * OK, now we've set up the new group. Time to make it active.
822 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400823 * We do not lock all allocations via s_resize_lock
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400824 * so we have to be safe wrt. concurrent accesses the group
825 * data. So we need to be careful to set all of the relevant
826 * group descriptor data etc. *before* we enable the group.
827 *
828 * The key field here is sbi->s_groups_count: as long as
829 * that retains its old value, nobody is going to access the new
830 * group.
831 *
832 * So first we update all the descriptor metadata for the new
833 * group; then we update the total disk blocks count; then we
834 * update the groups count to enable the group; then finally we
835 * update the free space counts so that the system can start
836 * using the new disk blocks.
837 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700838
839 /* Update group descriptor block for new group */
Frederic Bohe28569222008-06-20 11:48:48 -0400840 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
841 gdb_off * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842
Theodore Ts'ofdff73f2009-01-26 19:06:41 -0500843 memset(gdp, 0, EXT4_DESC_SIZE(sb));
Alexandre Ratchov8fadc142006-10-11 01:21:15 -0700844 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
845 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
846 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -0500847 ext4_free_blks_set(sb, gdp, input->free_blocks_count);
848 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'ofdff73f2009-01-26 19:06:41 -0500849 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
Andreas Dilger717d50e2007-10-16 18:38:25 -0400850 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700851
852 /*
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400853 * We can allocate memory for mb_alloc based on the new group
854 * descriptor
855 */
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -0500856 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -0400857 if (err)
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400858 goto exit_journal;
859
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400860 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 * Make the new blocks and inodes valid next. We do this before
862 * increasing the group count so that once the group is enabled,
863 * all of its blocks and inodes are already valid.
864 *
865 * We always allocate group-by-group, then block-by-block or
866 * inode-by-inode within a group, so enabling these
867 * blocks/inodes before the group is live won't actually let us
868 * allocate the new space yet.
869 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700870 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400872 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700873
874 /*
875 * We need to protect s_groups_count against other CPUs seeing
876 * inconsistent state in the superblock.
877 *
878 * The precise rules we use are:
879 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400880 * * Writers of s_groups_count *must* hold s_resize_lock
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700881 * AND
882 * * Writers must perform a smp_wmb() after updating all dependent
883 * data and before modifying the groups count
884 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400885 * * Readers must hold s_resize_lock over the access
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700886 * OR
887 * * Readers must perform an smp_rmb() after reading the groups count
888 * and before reading any dependent data.
889 *
890 * NB. These rules can be relaxed when checking the group count
891 * while freeing data, as we can only allocate from a block
892 * group after serialising against the group count, and we can
893 * only then free after serialising in turn against that
894 * allocation.
895 */
896 smp_wmb();
897
898 /* Update the global fs size fields */
899 sbi->s_groups_count++;
900
Frank Mayhar03901312009-01-07 00:06:22 -0500901 ext4_handle_dirty_metadata(handle, NULL, primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700902
903 /* Update the reserved block counts only once the new group is
904 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700905 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700906 input->reserved_blocks);
907
908 /* Update the free space counts */
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700909 percpu_counter_add(&sbi->s_freeblocks_counter,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700910 input->free_blocks_count);
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700911 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -0700912 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700913
Eric Sandeen42007ef2010-05-16 01:00:00 -0400914 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
915 sbi->s_log_groups_per_flex) {
Frederic Bohec62a11f2008-09-08 10:20:24 -0400916 ext4_group_t flex_group;
917 flex_group = ext4_flex_group(sbi, input->group);
Theodore Ts'o9f24e422009-03-04 19:09:10 -0500918 atomic_add(input->free_blocks_count,
919 &sbi->s_flex_groups[flex_group].free_blocks);
920 atomic_add(EXT4_INODES_PER_GROUP(sb),
921 &sbi->s_flex_groups[flex_group].free_inodes);
Frederic Bohec62a11f2008-09-08 10:20:24 -0400922 }
923
Theodore Ts'oa0375152010-06-11 23:14:04 -0400924 ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925
926exit_journal:
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400927 mutex_unlock(&sbi->s_resize_lock);
Mingming Cao617ba132006-10-11 01:20:53 -0700928 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700929 err = err2;
930 if (!err) {
931 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -0700932 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700933 update_backups(sb, primary->b_blocknr, primary->b_data,
934 primary->b_size);
935 }
936exit_put:
937 iput(inode);
938 return err;
Mingming Cao617ba132006-10-11 01:20:53 -0700939} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700940
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400941/*
942 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700943 * point is only used to extend the current filesystem to the end of the last
944 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
945 * for emergencies (because it has no dependencies on reserved blocks).
946 *
Mingming Cao617ba132006-10-11 01:20:53 -0700947 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948 * allow the "remount" trick to work for arbitrary resizing, assuming enough
949 * GDT blocks are reserved to grow to the desired size.
950 */
Mingming Cao617ba132006-10-11 01:20:53 -0700951int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
952 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953{
Mingming Cao617ba132006-10-11 01:20:53 -0700954 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700955 ext4_grpblk_t last;
956 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400957 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700958 handle_t *handle;
959 int err;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400960 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700961
962 /* We don't need to worry about locking wrt other resizers just
963 * yet: we're going to revalidate es->s_blocks_count after
Theodore Ts'o32ed5052009-04-25 22:53:39 -0400964 * taking the s_resize_lock below. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700965 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700966
967 if (test_opt(sb, DEBUG))
Mingming Cao2ae02102006-10-11 01:21:11 -0700968 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 o_blocks_count, n_blocks_count);
970
971 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
972 return 0;
973
974 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700975 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -0700976 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700977 sb->s_id, n_blocks_count);
978 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500979 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980 return -EINVAL;
981 }
982
983 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500984 ext4_warning(sb, "can't shrink FS - resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 return -EBUSY;
986 }
987
988 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400989 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990
991 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500992 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993 return -EPERM;
994 }
995
Mingming Cao617ba132006-10-11 01:20:53 -0700996 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997
998 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500999 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 return -EINVAL;
1001 }
1002
1003 if (o_blocks_count + add > n_blocks_count)
1004 add = n_blocks_count - o_blocks_count;
1005
1006 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001007 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008 o_blocks_count + add, add);
1009
1010 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001011 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001012 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001013 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001014 return -ENOSPC;
1015 }
1016 brelse(bh);
1017
1018 /* We will update the superblock, one block bitmap, and
Mingming Cao617ba132006-10-11 01:20:53 -07001019 * one group descriptor via ext4_free_blocks().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020 */
Mingming Cao617ba132006-10-11 01:20:53 -07001021 handle = ext4_journal_start_sb(sb, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022 if (IS_ERR(handle)) {
1023 err = PTR_ERR(handle);
Eric Sandeen12062dd2010-02-15 14:19:27 -05001024 ext4_warning(sb, "error %d on journal start", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025 goto exit_put;
1026 }
1027
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001028 mutex_lock(&EXT4_SB(sb)->s_resize_lock);
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001029 if (o_blocks_count != ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001030 ext4_warning(sb, "multiple resizers run on filesystem!");
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001031 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
Akinobu Mita5606bf52008-02-25 15:37:42 -05001032 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 err = -EBUSY;
1034 goto exit_put;
1035 }
1036
Mingming Cao617ba132006-10-11 01:20:53 -07001037 if ((err = ext4_journal_get_write_access(handle,
1038 EXT4_SB(sb)->s_sbh))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001039 ext4_warning(sb, "error %d on journal write access", err);
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001040 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001041 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042 goto exit_put;
1043 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001044 ext4_blocks_count_set(es, o_blocks_count + add);
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001045 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
Eric Sandeenc549a952008-01-28 23:58:27 -05001046 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 o_blocks_count + add);
Aneesh Kumar K.Ve21675d2009-01-05 21:36:02 -05001048 /* We add the blocks to the bitmap and set the group need init bit */
1049 ext4_add_groupblocks(handle, sb, o_blocks_count, add);
Theodore Ts'oa0375152010-06-11 23:14:04 -04001050 ext4_handle_dirty_super(handle, sb);
Mingming Cao2ae02102006-10-11 01:21:11 -07001051 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052 o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001053 if ((err = ext4_journal_stop(handle)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001054 goto exit_put;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001055
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001057 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1058 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001059 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1060 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001061exit_put:
1062 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001063} /* ext4_group_extend */