blob: b6ec1843a015d2f1d243d74a0818a5d31f06d536 [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"
Andreas Dilger717d50e2007-10-16 18:38:25 -040018#include "group.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070019
20#define outside(b, first, last) ((b) < (first) || (b) >= (last))
21#define inside(b, first, last) ((b) >= (first) && (b) < (last))
22
23static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070024 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070025{
Mingming Cao617ba132006-10-11 01:20:53 -070026 struct ext4_sb_info *sbi = EXT4_SB(sb);
27 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070028 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070029 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050030 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070031 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
32 unsigned overhead = ext4_bg_has_super(sb, group) ?
33 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070034 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070035 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070036 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070037 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070038 int err = -EINVAL;
39
40 input->free_blocks_count = free_blocks_count =
41 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
42
43 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070044 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070046 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070047 "no-super", input->group, input->blocks_count,
48 free_blocks_count, input->reserved_blocks);
49
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070050 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070051 if (group != sbi->s_groups_count)
Harvey Harrison46e665e2008-04-17 10:38:59 -040052 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053 "Cannot add at group %u (only %lu groups)",
54 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070055 else if (offset != 0)
Harvey Harrison46e665e2008-04-17 10:38:59 -040056 ext4_warning(sb, __func__, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070057 else if (input->reserved_blocks > input->blocks_count / 5)
Harvey Harrison46e665e2008-04-17 10:38:59 -040058 ext4_warning(sb, __func__, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070059 input->reserved_blocks);
60 else if (free_blocks_count < 0)
Harvey Harrison46e665e2008-04-17 10:38:59 -040061 ext4_warning(sb, __func__, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 input->blocks_count);
63 else if (!(bh = sb_bread(sb, end - 1)))
Harvey Harrison46e665e2008-04-17 10:38:59 -040064 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -070065 "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066 end - 1);
67 else if (outside(input->block_bitmap, start, end))
Harvey Harrison46e665e2008-04-17 10:38:59 -040068 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070069 "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070070 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070071 else if (outside(input->inode_bitmap, start, end))
Harvey Harrison46e665e2008-04-17 10:38:59 -040072 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070073 "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070074 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -040076 outside(itend - 1, start, end))
Harvey Harrison46e665e2008-04-17 10:38:59 -040077 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070078 "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070079 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070080 else if (input->inode_bitmap == input->block_bitmap)
Harvey Harrison46e665e2008-04-17 10:38:59 -040081 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070082 "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070083 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 else if (inside(input->block_bitmap, input->inode_table, itend))
Harvey Harrison46e665e2008-04-17 10:38:59 -040085 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070086 "Block bitmap (%llu) in inode table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070087 (unsigned long long)input->block_bitmap,
88 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070089 else if (inside(input->inode_bitmap, input->inode_table, itend))
Harvey Harrison46e665e2008-04-17 10:38:59 -040090 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070091 "Inode bitmap (%llu) in inode table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070092 (unsigned long long)input->inode_bitmap,
93 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070094 else if (inside(input->block_bitmap, start, metaend))
Harvey Harrison46e665e2008-04-17 10:38:59 -040095 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070096 "Block bitmap (%llu) in GDT table"
Mingming Cao2ae02102006-10-11 01:21:11 -070097 " (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070098 (unsigned long long)input->block_bitmap,
99 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700100 else if (inside(input->inode_bitmap, start, metaend))
Harvey Harrison46e665e2008-04-17 10:38:59 -0400101 ext4_warning(sb, __func__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700102 "Inode bitmap (%llu) in GDT table"
Mingming Cao2ae02102006-10-11 01:21:11 -0700103 " (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700104 (unsigned long long)input->inode_bitmap,
105 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400107 inside(itend - 1, start, metaend))
Harvey Harrison46e665e2008-04-17 10:38:59 -0400108 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700109 "Inode table (%llu-%llu) overlaps"
110 "GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700111 (unsigned long long)input->inode_table,
112 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700113 else
114 err = 0;
115 brelse(bh);
116
117 return err;
118}
119
120static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700121 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700122{
123 struct buffer_head *bh;
124 int err;
125
126 bh = sb_getblk(sb, blk);
127 if (!bh)
128 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700129 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130 brelse(bh);
131 bh = ERR_PTR(err);
132 } else {
133 lock_buffer(bh);
134 memset(bh->b_data, 0, sb->s_blocksize);
135 set_buffer_uptodate(bh);
136 unlock_buffer(bh);
137 }
138
139 return bh;
140}
141
142/*
Eric Sandeen14904102007-10-16 18:38:25 -0400143 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
144 * If that fails, restart the transaction & regain write access for the
145 * buffer head which is used for block_bitmap modifications.
146 */
147static int extend_or_restart_transaction(handle_t *handle, int thresh,
148 struct buffer_head *bh)
149{
150 int err;
151
152 if (handle->h_buffer_credits >= thresh)
153 return 0;
154
155 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
156 if (err < 0)
157 return err;
158 if (err) {
159 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
160 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400161 if ((err = ext4_journal_get_write_access(handle, bh)))
Eric Sandeen14904102007-10-16 18:38:25 -0400162 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400163 }
Eric Sandeen14904102007-10-16 18:38:25 -0400164
165 return 0;
166}
167
168/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700169 * Set up the block and inode bitmaps, and the inode table for the new group.
170 * This doesn't need to be part of the main transaction, since we are only
171 * changing blocks outside the actual filesystem. We still do journaling to
172 * ensure the recovery is correct in case of a failure just after resize.
173 * If any part of this fails, we simply abort the resize.
174 */
175static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700176 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700177{
Mingming Cao617ba132006-10-11 01:20:53 -0700178 struct ext4_sb_info *sbi = EXT4_SB(sb);
179 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
180 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700181 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700182 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700183 struct buffer_head *bh;
184 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700185 ext4_fsblk_t block;
186 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700187 int i;
188 int err = 0, err2;
189
Eric Sandeen14904102007-10-16 18:38:25 -0400190 /* This transaction may be extended/restarted along the way */
191 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
192
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700193 if (IS_ERR(handle))
194 return PTR_ERR(handle);
195
196 lock_super(sb);
197 if (input->group != sbi->s_groups_count) {
198 err = -EBUSY;
199 goto exit_journal;
200 }
201
202 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
203 err = PTR_ERR(bh);
204 goto exit_journal;
205 }
206
Mingming Cao617ba132006-10-11 01:20:53 -0700207 if (ext4_bg_has_super(sb, input->group)) {
Eric Sandeenc549a952008-01-28 23:58:27 -0500208 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
Mingming Cao617ba132006-10-11 01:20:53 -0700209 ext4_set_bit(0, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700210 }
211
212 /* Copy all of the GDT blocks into the backup in this group */
213 for (i = 0, bit = 1, block = start + 1;
214 i < gdblocks; i++, block++, bit++) {
215 struct buffer_head *gdb;
216
Eric Sandeenc549a952008-01-28 23:58:27 -0500217 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700218
Eric Sandeen14904102007-10-16 18:38:25 -0400219 if ((err = extend_or_restart_transaction(handle, 1, bh)))
220 goto exit_bh;
221
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700222 gdb = sb_getblk(sb, block);
223 if (!gdb) {
224 err = -EIO;
225 goto exit_bh;
226 }
Mingming Cao617ba132006-10-11 01:20:53 -0700227 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700228 brelse(gdb);
229 goto exit_bh;
230 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400231 lock_buffer(gdb);
232 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700233 set_buffer_uptodate(gdb);
Eric Sandeen5b615282007-10-16 18:38:25 -0400234 unlock_buffer(gdb);
Mingming Cao617ba132006-10-11 01:20:53 -0700235 ext4_journal_dirty_metadata(handle, gdb);
236 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237 brelse(gdb);
238 }
239
240 /* Zero out all of the reserved backup group descriptor table blocks */
241 for (i = 0, bit = gdblocks + 1, block = start + bit;
242 i < reserved_gdb; i++, block++, bit++) {
243 struct buffer_head *gdb;
244
Eric Sandeenc549a952008-01-28 23:58:27 -0500245 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246
Eric Sandeen14904102007-10-16 18:38:25 -0400247 if ((err = extend_or_restart_transaction(handle, 1, bh)))
248 goto exit_bh;
249
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250 if (IS_ERR(gdb = bclean(handle, sb, block))) {
251 err = PTR_ERR(bh);
252 goto exit_bh;
253 }
Mingming Cao617ba132006-10-11 01:20:53 -0700254 ext4_journal_dirty_metadata(handle, gdb);
255 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700256 brelse(gdb);
257 }
Eric Sandeenc549a952008-01-28 23:58:27 -0500258 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700259 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700260 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500261 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700263 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264
265 /* Zero out all of the inode table blocks */
266 for (i = 0, block = input->inode_table, bit = block - start;
267 i < sbi->s_itb_per_group; i++, bit++, block++) {
268 struct buffer_head *it;
269
Eric Sandeenc549a952008-01-28 23:58:27 -0500270 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit);
Eric Sandeen14904102007-10-16 18:38:25 -0400271
272 if ((err = extend_or_restart_transaction(handle, 1, bh)))
273 goto exit_bh;
274
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700275 if (IS_ERR(it = bclean(handle, sb, block))) {
276 err = PTR_ERR(it);
277 goto exit_bh;
278 }
Mingming Cao617ba132006-10-11 01:20:53 -0700279 ext4_journal_dirty_metadata(handle, it);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700280 brelse(it);
Mingming Cao617ba132006-10-11 01:20:53 -0700281 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700282 }
Eric Sandeen14904102007-10-16 18:38:25 -0400283
284 if ((err = extend_or_restart_transaction(handle, 2, bh)))
285 goto exit_bh;
286
Mingming Cao617ba132006-10-11 01:20:53 -0700287 mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700288 bh->b_data);
Mingming Cao617ba132006-10-11 01:20:53 -0700289 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290 brelse(bh);
291
292 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500293 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294 input->inode_bitmap, input->inode_bitmap - start);
295 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
296 err = PTR_ERR(bh);
297 goto exit_journal;
298 }
299
Mingming Cao617ba132006-10-11 01:20:53 -0700300 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700301 bh->b_data);
Mingming Cao617ba132006-10-11 01:20:53 -0700302 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303exit_bh:
304 brelse(bh);
305
306exit_journal:
307 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700308 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700309 err = err2;
310
311 return err;
312}
313
314/*
315 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700316 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700317 * calling this for the first time. In a sparse filesystem it will be the
318 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
319 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
320 */
Mingming Cao617ba132006-10-11 01:20:53 -0700321static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700322 unsigned *five, unsigned *seven)
323{
324 unsigned *min = three;
325 int mult = 3;
326 unsigned ret;
327
Mingming Cao617ba132006-10-11 01:20:53 -0700328 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
329 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700330 ret = *min;
331 *min += 1;
332 return ret;
333 }
334
335 if (*five < *min) {
336 min = five;
337 mult = 5;
338 }
339 if (*seven < *min) {
340 min = seven;
341 mult = 7;
342 }
343
344 ret = *min;
345 *min *= mult;
346
347 return ret;
348}
349
350/*
351 * Check that all of the backup GDT blocks are held in the primary GDT block.
352 * It is assumed that they are stored in group order. Returns the number of
353 * groups in current filesystem that have BACKUPS, or -ve error code.
354 */
355static int verify_reserved_gdb(struct super_block *sb,
356 struct buffer_head *primary)
357{
Mingming Cao617ba132006-10-11 01:20:53 -0700358 const ext4_fsblk_t blk = primary->b_blocknr;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500359 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700360 unsigned three = 1;
361 unsigned five = 5;
362 unsigned seven = 7;
363 unsigned grp;
364 __le32 *p = (__le32 *)primary->b_data;
365 int gdbackups = 0;
366
Mingming Cao617ba132006-10-11 01:20:53 -0700367 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700368 if (le32_to_cpu(*p++) !=
369 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Harvey Harrison46e665e2008-04-17 10:38:59 -0400370 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700371 "reserved GDT %llu"
372 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700373 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700374 grp *
375 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
376 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700377 return -EINVAL;
378 }
Mingming Cao617ba132006-10-11 01:20:53 -0700379 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700380 return -EFBIG;
381 }
382
383 return gdbackups;
384}
385
386/*
387 * Called when we need to bring a reserved group descriptor table block into
388 * use from the resize inode. The primary copy of the new GDT block currently
389 * is an indirect block (under the double indirect block in the resize inode).
390 * The new backup GDT blocks will be stored as leaf blocks in this indirect
391 * block, in group order. Even though we know all the block numbers we need,
392 * we check to ensure that the resize inode has actually reserved these blocks.
393 *
394 * Don't need to update the block bitmaps because the blocks are still in use.
395 *
396 * We get all of the error cases out of the way, so that we are sure to not
397 * fail once we start modifying the data on disk, because JBD has no rollback.
398 */
399static int add_new_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700400 struct ext4_new_group_data *input,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700401 struct buffer_head **primary)
402{
403 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700404 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
405 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
406 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700407 struct buffer_head **o_group_desc, **n_group_desc;
408 struct buffer_head *dind;
409 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700410 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411 __le32 *data;
412 int err;
413
414 if (test_opt(sb, DEBUG))
415 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700416 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417 gdb_num);
418
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400419 /*
420 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400421 * because the user tools have no way of handling this. Probably a
422 * bad time to do it anyways.
423 */
Mingming Cao617ba132006-10-11 01:20:53 -0700424 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
425 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400426 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700427 "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700428 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700429 return -EPERM;
430 }
431
432 *primary = sb_bread(sb, gdblock);
433 if (!*primary)
434 return -EIO;
435
436 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
437 err = gdbackups;
438 goto exit_bh;
439 }
440
Mingming Cao617ba132006-10-11 01:20:53 -0700441 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700442 dind = sb_bread(sb, le32_to_cpu(*data));
443 if (!dind) {
444 err = -EIO;
445 goto exit_bh;
446 }
447
448 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700449 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400450 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700451 "new group %u GDT block %llu not reserved",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700452 input->group, gdblock);
453 err = -EINVAL;
454 goto exit_dind;
455 }
456
Mingming Cao617ba132006-10-11 01:20:53 -0700457 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700458 goto exit_dind;
459
Mingming Cao617ba132006-10-11 01:20:53 -0700460 if ((err = ext4_journal_get_write_access(handle, *primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700461 goto exit_sbh;
462
Mingming Cao617ba132006-10-11 01:20:53 -0700463 if ((err = ext4_journal_get_write_access(handle, dind)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700464 goto exit_primary;
465
Mingming Cao617ba132006-10-11 01:20:53 -0700466 /* ext4_reserve_inode_write() gets a reference on the iloc */
467 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700468 goto exit_dindj;
469
470 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
Josef Bacik216553c2008-04-29 22:02:02 -0400471 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700472 if (!n_group_desc) {
473 err = -ENOMEM;
Harvey Harrison46e665e2008-04-17 10:38:59 -0400474 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700475 "not enough memory for %lu groups", gdb_num + 1);
476 goto exit_inode;
477 }
478
479 /*
480 * Finally, we have all of the possible failures behind us...
481 *
482 * Remove new GDT block from inode double-indirect block and clear out
483 * the new GDT block for use (which also "frees" the backup GDT blocks
484 * from the reserved inode). We don't need to change the bitmaps for
485 * these blocks, because they are marked as in-use from being in the
486 * reserved inode, and will become GDT blocks (primary and backup).
487 */
Mingming Cao617ba132006-10-11 01:20:53 -0700488 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
489 ext4_journal_dirty_metadata(handle, dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700490 brelse(dind);
491 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700492 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700493 memset((*primary)->b_data, 0, sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700494 ext4_journal_dirty_metadata(handle, *primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495
Mingming Cao617ba132006-10-11 01:20:53 -0700496 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700497 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700498 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700499 n_group_desc[gdb_num] = *primary;
Mingming Cao617ba132006-10-11 01:20:53 -0700500 EXT4_SB(sb)->s_group_desc = n_group_desc;
501 EXT4_SB(sb)->s_gdb_count++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700502 kfree(o_group_desc);
503
Marcin Slusarze8546d02008-04-17 10:38:59 -0400504 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Mingming Cao617ba132006-10-11 01:20:53 -0700505 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700506
507 return 0;
508
509exit_inode:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400510 /* ext4_journal_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700511 brelse(iloc.bh);
512exit_dindj:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400513 /* ext4_journal_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700514exit_primary:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400515 /* ext4_journal_release_buffer(handle, *primary); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700516exit_sbh:
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400517 /* ext4_journal_release_buffer(handle, *primary); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700518exit_dind:
519 brelse(dind);
520exit_bh:
521 brelse(*primary);
522
Mingming Cao617ba132006-10-11 01:20:53 -0700523 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700524 return err;
525}
526
527/*
528 * Called when we are adding a new group which has a backup copy of each of
529 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
530 * We need to add these reserved backup GDT blocks to the resize inode, so
531 * that they are kept for future resizing and not allocated to files.
532 *
533 * Each reserved backup GDT block will go into a different indirect block.
534 * The indirect blocks are actually the primary reserved GDT blocks,
535 * so we know in advance what their block numbers are. We only get the
536 * double-indirect block to verify it is pointing to the primary reserved
537 * GDT blocks so we don't overwrite a data block by accident. The reserved
538 * backup GDT blocks are stored in their reserved primary GDT block.
539 */
540static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700541 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700542{
543 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700544 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700545 struct buffer_head **primary;
546 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700547 struct ext4_iloc iloc;
548 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700549 __le32 *data, *end;
550 int gdbackups = 0;
551 int res, i;
552 int err;
553
Josef Bacik216553c2008-04-29 22:02:02 -0400554 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 if (!primary)
556 return -ENOMEM;
557
Mingming Cao617ba132006-10-11 01:20:53 -0700558 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700559 dind = sb_bread(sb, le32_to_cpu(*data));
560 if (!dind) {
561 err = -EIO;
562 goto exit_free;
563 }
564
Mingming Cao617ba132006-10-11 01:20:53 -0700565 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400566 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
567 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700568 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569
570 /* Get each reserved primary GDT block and verify it holds backups */
571 for (res = 0; res < reserved_gdb; res++, blk++) {
572 if (le32_to_cpu(*data) != blk) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400573 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700574 "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700575 " not at offset %ld",
576 blk,
577 (long)(data - (__le32 *)dind->b_data));
578 err = -EINVAL;
579 goto exit_bh;
580 }
581 primary[res] = sb_bread(sb, blk);
582 if (!primary[res]) {
583 err = -EIO;
584 goto exit_bh;
585 }
586 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
587 brelse(primary[res]);
588 err = gdbackups;
589 goto exit_bh;
590 }
591 if (++data >= end)
592 data = (__le32 *)dind->b_data;
593 }
594
595 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700596 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700597 /*
598 int j;
599 for (j = 0; j < i; j++)
Mingming Cao617ba132006-10-11 01:20:53 -0700600 ext4_journal_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601 */
602 goto exit_bh;
603 }
604 }
605
Mingming Cao617ba132006-10-11 01:20:53 -0700606 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700607 goto exit_bh;
608
609 /*
610 * Finally we can add each of the reserved backup GDT blocks from
611 * the new group to its reserved primary GDT block.
612 */
Mingming Cao617ba132006-10-11 01:20:53 -0700613 blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700614 for (i = 0; i < reserved_gdb; i++) {
615 int err2;
616 data = (__le32 *)primary[i]->b_data;
617 /* printk("reserving backup %lu[%u] = %lu\n",
618 primary[i]->b_blocknr, gdbackups,
619 blk + primary[i]->b_blocknr); */
620 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -0700621 err2 = ext4_journal_dirty_metadata(handle, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 if (!err)
623 err = err2;
624 }
625 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700626 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700627
628exit_bh:
629 while (--res >= 0)
630 brelse(primary[res]);
631 brelse(dind);
632
633exit_free:
634 kfree(primary);
635
636 return err;
637}
638
639/*
Mingming Cao617ba132006-10-11 01:20:53 -0700640 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700641 * of the main resize transaction, because e2fsck will re-write them if there
642 * is a problem (basically only OOM will cause a problem). However, we
643 * _should_ update the backups if possible, in case the primary gets trashed
644 * for some reason and we need to run e2fsck from a backup superblock. The
645 * important part is that the new block and inode counts are in the backup
646 * superblocks, and the location of the new group metadata in the GDT backups.
647 *
648 * We do not need lock_super() for this, because these blocks are not
649 * otherwise touched by the filesystem code when it is mounted. We don't
650 * need to worry about last changing from sbi->s_groups_count, because the
651 * worst that can happen is that we do not copy the full number of backups
652 * at this time. The resize which changed s_groups_count will backup again.
653 */
654static void update_backups(struct super_block *sb,
655 int blk_off, char *data, int size)
656{
Mingming Cao617ba132006-10-11 01:20:53 -0700657 struct ext4_sb_info *sbi = EXT4_SB(sb);
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500658 const ext4_group_t last = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700659 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 unsigned three = 1;
661 unsigned five = 5;
662 unsigned seven = 7;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500663 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664 int rest = sb->s_blocksize - size;
665 handle_t *handle;
666 int err = 0, err2;
667
Mingming Cao617ba132006-10-11 01:20:53 -0700668 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700669 if (IS_ERR(handle)) {
670 group = 1;
671 err = PTR_ERR(handle);
672 goto exit_err;
673 }
674
Mingming Cao617ba132006-10-11 01:20:53 -0700675 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700676 struct buffer_head *bh;
677
678 /* Out of journal space, and can't get more - abort - so sad */
679 if (handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -0700680 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
681 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700682 break;
683
684 bh = sb_getblk(sb, group * bpg + blk_off);
685 if (!bh) {
686 err = -EIO;
687 break;
688 }
Mingming Cao617ba132006-10-11 01:20:53 -0700689 ext4_debug("update metadata backup %#04lx\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700690 (unsigned long)bh->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -0700691 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700692 break;
693 lock_buffer(bh);
694 memcpy(bh->b_data, data, size);
695 if (rest)
696 memset(bh->b_data + size, 0, rest);
697 set_buffer_uptodate(bh);
698 unlock_buffer(bh);
Mingming Cao617ba132006-10-11 01:20:53 -0700699 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 brelse(bh);
701 }
Mingming Cao617ba132006-10-11 01:20:53 -0700702 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700703 err = err2;
704
705 /*
706 * Ugh! Need to have e2fsck write the backup copies. It is too
707 * late to revert the resize, we shouldn't fail just because of
708 * the backup copies (they are only needed in case of corruption).
709 *
710 * However, if we got here we have a journal problem too, so we
711 * can't really start a transaction to mark the superblock.
712 * Chicken out and just set the flag on the hope it will be written
713 * to disk, and if not - we will simply wait until next fsck.
714 */
715exit_err:
716 if (err) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400717 ext4_warning(sb, __func__,
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500718 "can't update backup for group %lu (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -0700720 sbi->s_mount_state &= ~EXT4_VALID_FS;
721 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700722 mark_buffer_dirty(sbi->s_sbh);
723 }
724}
725
726/* Add group descriptor data to an existing or new group descriptor block.
727 * Ensure we handle all possible error conditions _before_ we start modifying
728 * the filesystem, because we cannot abort the transaction and not have it
729 * write the data to disk.
730 *
731 * If we are on a GDT block boundary, we need to get the reserved GDT block.
732 * Otherwise, we may need to add backup GDT blocks for a sparse group.
733 *
734 * We only need to hold the superblock lock while we are actually adding
735 * in the new group's counts to the superblock. Prior to that we have
736 * not really "added" the group at all. We re-check that we are still
737 * adding in the last group in case things have changed since verifying.
738 */
Mingming Cao617ba132006-10-11 01:20:53 -0700739int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740{
Mingming Cao617ba132006-10-11 01:20:53 -0700741 struct ext4_sb_info *sbi = EXT4_SB(sb);
742 struct ext4_super_block *es = sbi->s_es;
743 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700744 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
745 struct buffer_head *primary = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700746 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700747 struct inode *inode = NULL;
748 handle_t *handle;
749 int gdb_off, gdb_num;
750 int err, err2;
751
Mingming Cao617ba132006-10-11 01:20:53 -0700752 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
753 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700754
Mingming Cao617ba132006-10-11 01:20:53 -0700755 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
756 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400757 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700758 "Can't resize non-sparse filesystem further");
759 return -EPERM;
760 }
761
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700762 if (ext4_blocks_count(es) + input->blocks_count <
763 ext4_blocks_count(es)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400764 ext4_warning(sb, __func__, "blocks_count overflow\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765 return -EINVAL;
766 }
767
Mingming Cao617ba132006-10-11 01:20:53 -0700768 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700769 le32_to_cpu(es->s_inodes_count)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400770 ext4_warning(sb, __func__, "inodes_count overflow\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700771 return -EINVAL;
772 }
773
774 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700775 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -0400776 EXT4_FEATURE_COMPAT_RESIZE_INODE)
777 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400778 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700779 "No reserved GDT blocks, can't resize");
780 return -EPERM;
781 }
David Howells1d1fe1e2008-02-07 00:15:37 -0800782 inode = ext4_iget(sb, EXT4_RESIZE_INO);
783 if (IS_ERR(inode)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400784 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700785 "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -0800786 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787 }
788 }
789
790 if ((err = verify_group_input(sb, input)))
791 goto exit_put;
792
793 if ((err = setup_new_group_blocks(sb, input)))
794 goto exit_put;
795
796 /*
797 * We will always be modifying at least the superblock and a GDT
798 * block. If we are adding a group past the last current GDT block,
799 * we will also modify the inode and the dindirect block. If we
800 * are adding a group with superblock/GDT backups we will also
801 * modify each of the reserved GDT dindirect blocks.
802 */
Mingming Cao617ba132006-10-11 01:20:53 -0700803 handle = ext4_journal_start_sb(sb,
804 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805 3 + reserved_gdb : 4);
806 if (IS_ERR(handle)) {
807 err = PTR_ERR(handle);
808 goto exit_put;
809 }
810
811 lock_super(sb);
812 if (input->group != sbi->s_groups_count) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400813 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700814 "multiple resizers run on filesystem!");
815 err = -EBUSY;
816 goto exit_journal;
817 }
818
Mingming Cao617ba132006-10-11 01:20:53 -0700819 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700820 goto exit_journal;
821
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400822 /*
823 * We will only either add reserved group blocks to a backup group
824 * or remove reserved blocks for the first group in a new group block.
825 * Doing both would be mean more complex code, and sane people don't
826 * use non-sparse filesystems anymore. This is already checked above.
827 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700828 if (gdb_off) {
829 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -0700830 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 goto exit_journal;
832
Mingming Cao617ba132006-10-11 01:20:53 -0700833 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700834 (err = reserve_backup_gdb(handle, inode, input)))
835 goto exit_journal;
836 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
837 goto exit_journal;
838
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400839 /*
840 * OK, now we've set up the new group. Time to make it active.
841 *
842 * Current kernels don't lock all allocations via lock_super(),
843 * so we have to be safe wrt. concurrent accesses the group
844 * data. So we need to be careful to set all of the relevant
845 * group descriptor data etc. *before* we enable the group.
846 *
847 * The key field here is sbi->s_groups_count: as long as
848 * that retains its old value, nobody is going to access the new
849 * group.
850 *
851 * So first we update all the descriptor metadata for the new
852 * group; then we update the total disk blocks count; then we
853 * update the groups count to enable the group; then finally we
854 * update the free space counts so that the system can start
855 * using the new disk blocks.
856 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700857
858 /* Update group descriptor block for new group */
Frederic Bohe28569222008-06-20 11:48:48 -0400859 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
860 gdb_off * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861
Alexandre Ratchov8fadc142006-10-11 01:21:15 -0700862 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
863 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
864 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865 gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
Mingming Cao617ba132006-10-11 01:20:53 -0700866 gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
Andreas Dilger717d50e2007-10-16 18:38:25 -0400867 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700868
869 /*
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400870 * We can allocate memory for mb_alloc based on the new group
871 * descriptor
872 */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -0400873 err = ext4_mb_add_more_groupinfo(sb, input->group, gdp);
874 if (err)
875 goto exit_journal;
876
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400877 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700878 * Make the new blocks and inodes valid next. We do this before
879 * increasing the group count so that once the group is enabled,
880 * all of its blocks and inodes are already valid.
881 *
882 * We always allocate group-by-group, then block-by-block or
883 * inode-by-inode within a group, so enabling these
884 * blocks/inodes before the group is live won't actually let us
885 * allocate the new space yet.
886 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700887 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400889 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700890
891 /*
892 * We need to protect s_groups_count against other CPUs seeing
893 * inconsistent state in the superblock.
894 *
895 * The precise rules we use are:
896 *
897 * * Writers of s_groups_count *must* hold lock_super
898 * AND
899 * * Writers must perform a smp_wmb() after updating all dependent
900 * data and before modifying the groups count
901 *
902 * * Readers must hold lock_super() over the access
903 * OR
904 * * Readers must perform an smp_rmb() after reading the groups count
905 * and before reading any dependent data.
906 *
907 * NB. These rules can be relaxed when checking the group count
908 * while freeing data, as we can only allocate from a block
909 * group after serialising against the group count, and we can
910 * only then free after serialising in turn against that
911 * allocation.
912 */
913 smp_wmb();
914
915 /* Update the global fs size fields */
916 sbi->s_groups_count++;
917
Mingming Cao617ba132006-10-11 01:20:53 -0700918 ext4_journal_dirty_metadata(handle, primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700919
920 /* Update the reserved block counts only once the new group is
921 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700922 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700923 input->reserved_blocks);
924
925 /* Update the free space counts */
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700926 percpu_counter_add(&sbi->s_freeblocks_counter,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700927 input->free_blocks_count);
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700928 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -0700929 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700930
Frederic Bohec62a11f2008-09-08 10:20:24 -0400931 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
932 ext4_group_t flex_group;
933 flex_group = ext4_flex_group(sbi, input->group);
934 sbi->s_flex_groups[flex_group].free_blocks +=
935 input->free_blocks_count;
936 sbi->s_flex_groups[flex_group].free_inodes +=
937 EXT4_INODES_PER_GROUP(sb);
938 }
939
Mingming Cao617ba132006-10-11 01:20:53 -0700940 ext4_journal_dirty_metadata(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700941 sb->s_dirt = 1;
942
943exit_journal:
944 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700945 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700946 err = err2;
947 if (!err) {
948 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -0700949 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 update_backups(sb, primary->b_blocknr, primary->b_data,
951 primary->b_size);
952 }
953exit_put:
954 iput(inode);
955 return err;
Mingming Cao617ba132006-10-11 01:20:53 -0700956} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700957
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400958/*
959 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960 * point is only used to extend the current filesystem to the end of the last
961 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
962 * for emergencies (because it has no dependencies on reserved blocks).
963 *
Mingming Cao617ba132006-10-11 01:20:53 -0700964 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700965 * allow the "remount" trick to work for arbitrary resizing, assuming enough
966 * GDT blocks are reserved to grow to the desired size.
967 */
Mingming Cao617ba132006-10-11 01:20:53 -0700968int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
969 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700970{
Mingming Cao617ba132006-10-11 01:20:53 -0700971 ext4_fsblk_t o_blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500972 ext4_group_t o_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700973 ext4_grpblk_t last;
974 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400975 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976 handle_t *handle;
977 int err;
978 unsigned long freed_blocks;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -0400979 ext4_group_t group;
980 struct ext4_group_info *grp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700981
982 /* We don't need to worry about locking wrt other resizers just
983 * yet: we're going to revalidate es->s_blocks_count after
984 * taking lock_super() below. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700985 o_blocks_count = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -0700986 o_groups_count = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987
988 if (test_opt(sb, DEBUG))
Mingming Cao2ae02102006-10-11 01:21:11 -0700989 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990 o_blocks_count, n_blocks_count);
991
992 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
993 return 0;
994
995 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700996 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -0700997 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998 sb->s_id, n_blocks_count);
999 if (sizeof(sector_t) < 8)
Harvey Harrison46e665e2008-04-17 10:38:59 -04001000 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001001 "CONFIG_LBD not enabled\n");
1002 return -EINVAL;
1003 }
1004
1005 if (n_blocks_count < o_blocks_count) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001006 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007 "can't shrink FS - resize aborted");
1008 return -EBUSY;
1009 }
1010
1011 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001012 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013
1014 if (last == 0) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001015 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001016 "need to use ext2online to resize further");
1017 return -EPERM;
1018 }
1019
Mingming Cao617ba132006-10-11 01:20:53 -07001020 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001021
1022 if (o_blocks_count + add < o_blocks_count) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001023 ext4_warning(sb, __func__, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024 return -EINVAL;
1025 }
1026
1027 if (o_blocks_count + add > n_blocks_count)
1028 add = n_blocks_count - o_blocks_count;
1029
1030 if (o_blocks_count + add < n_blocks_count)
Harvey Harrison46e665e2008-04-17 10:38:59 -04001031 ext4_warning(sb, __func__,
Mingming Cao2ae02102006-10-11 01:21:11 -07001032 "will only finish group (%llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 " blocks, %u new)",
1034 o_blocks_count + add, add);
1035
1036 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001037 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001038 if (!bh) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001039 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 "can't read last block, resize aborted");
1041 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);
Harvey Harrison46e665e2008-04-17 10:38:59 -04001051 ext4_warning(sb, __func__, "error %d on journal start", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052 goto exit_put;
1053 }
1054
1055 lock_super(sb);
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001056 if (o_blocks_count != ext4_blocks_count(es)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001057 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001058 "multiple resizers run on filesystem!");
1059 unlock_super(sb);
Akinobu Mita5606bf52008-02-25 15:37:42 -05001060 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001061 err = -EBUSY;
1062 goto exit_put;
1063 }
1064
Mingming Cao617ba132006-10-11 01:20:53 -07001065 if ((err = ext4_journal_get_write_access(handle,
1066 EXT4_SB(sb)->s_sbh))) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001067 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001068 "error %d on journal write access", err);
1069 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001070 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001071 goto exit_put;
1072 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001073 ext4_blocks_count_set(es, o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001074 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001075 sb->s_dirt = 1;
1076 unlock_super(sb);
Eric Sandeenc549a952008-01-28 23:58:27 -05001077 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078 o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001079 ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
Mingming Cao2ae02102006-10-11 01:21:11 -07001080 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001081 o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001082 if ((err = ext4_journal_stop(handle)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001083 goto exit_put;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001084
1085 /*
1086 * Mark mballoc pages as not up to date so that they will be updated
1087 * next time they are loaded by ext4_mb_load_buddy.
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001088 *
1089 * XXX Bad, Bad, BAD!!! We should not be overloading the
1090 * Uptodate flag, particularly on thte bitmap bh, as way of
1091 * hinting to ext4_mb_load_buddy() that it needs to be
1092 * overloaded. A user could take a LVM snapshot, then do an
1093 * on-line fsck, and clear the uptodate flag, and this would
1094 * not be a bug in userspace, but a bug in the kernel. FIXME!!!
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001095 */
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001096 {
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001097 struct ext4_sb_info *sbi = EXT4_SB(sb);
1098 struct inode *inode = sbi->s_buddy_cache;
1099 int blocks_per_page;
1100 int block;
1101 int pnum;
1102 struct page *page;
1103
1104 /* Set buddy page as not up to date */
1105 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1106 block = group * 2;
1107 pnum = block / blocks_per_page;
1108 page = find_get_page(inode->i_mapping, pnum);
1109 if (page != NULL) {
1110 ClearPageUptodate(page);
1111 page_cache_release(page);
1112 }
1113
1114 /* Set bitmap page as not up to date */
1115 block++;
1116 pnum = block / blocks_per_page;
1117 page = find_get_page(inode->i_mapping, pnum);
1118 if (page != NULL) {
1119 ClearPageUptodate(page);
1120 page_cache_release(page);
1121 }
1122
1123 /* Get the info on the last group */
1124 grp = ext4_get_group_info(sb, group);
1125
1126 /* Update free blocks in group info */
1127 ext4_mb_update_group_info(grp, add);
1128 }
1129
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001130 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001131 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1132 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001133 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1134 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135exit_put:
1136 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001137} /* ext4_group_extend */