blob: 728e3efa84b5450be114fd080a6fed571100c908 [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
Mingming Caodab291a2006-10-11 01:21:01 -070014#include <linux/ext4_jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070015
16#include <linux/errno.h>
17#include <linux/slab.h>
18
Andreas Dilger717d50e2007-10-16 18:38:25 -040019#include "group.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070020
21#define outside(b, first, last) ((b) < (first) || (b) >= (last))
22#define inside(b, first, last) ((b) >= (first) && (b) < (last))
23
24static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070025 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070026{
Mingming Cao617ba132006-10-11 01:20:53 -070027 struct ext4_sb_info *sbi = EXT4_SB(sb);
28 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070029 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070030 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050031 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070032 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
33 unsigned overhead = ext4_bg_has_super(sb, group) ?
34 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070035 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070036 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070037 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070038 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070039 int err = -EINVAL;
40
41 input->free_blocks_count = free_blocks_count =
42 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
43
44 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070045 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070046 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070047 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048 "no-super", input->group, input->blocks_count,
49 free_blocks_count, input->reserved_blocks);
50
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070051 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070052 if (group != sbi->s_groups_count)
Mingming Cao617ba132006-10-11 01:20:53 -070053 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070054 "Cannot add at group %u (only %lu groups)",
55 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070056 else if (offset != 0)
57 ext4_warning(sb, __FUNCTION__, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070058 else if (input->reserved_blocks > input->blocks_count / 5)
Mingming Cao617ba132006-10-11 01:20:53 -070059 ext4_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070060 input->reserved_blocks);
61 else if (free_blocks_count < 0)
Mingming Cao617ba132006-10-11 01:20:53 -070062 ext4_warning(sb, __FUNCTION__, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070063 input->blocks_count);
64 else if (!(bh = sb_bread(sb, end - 1)))
Mingming Cao617ba132006-10-11 01:20:53 -070065 ext4_warning(sb, __FUNCTION__,
Mingming Cao2ae02102006-10-11 01:21:11 -070066 "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070067 end - 1);
68 else if (outside(input->block_bitmap, start, end))
Mingming Cao617ba132006-10-11 01:20:53 -070069 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070070 "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070071 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070072 else if (outside(input->inode_bitmap, start, end))
Mingming Cao617ba132006-10-11 01:20:53 -070073 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070074 "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070075 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076 else if (outside(input->inode_table, start, end) ||
77 outside(itend - 1, start, end))
Mingming Cao617ba132006-10-11 01:20:53 -070078 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070079 "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070080 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081 else if (input->inode_bitmap == input->block_bitmap)
Mingming Cao617ba132006-10-11 01:20:53 -070082 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070083 "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070084 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070085 else if (inside(input->block_bitmap, input->inode_table, itend))
Mingming Cao617ba132006-10-11 01:20:53 -070086 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070087 "Block bitmap (%llu) in inode table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070088 (unsigned long long)input->block_bitmap,
89 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070090 else if (inside(input->inode_bitmap, input->inode_table, itend))
Mingming Cao617ba132006-10-11 01:20:53 -070091 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070092 "Inode bitmap (%llu) in inode table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070093 (unsigned long long)input->inode_bitmap,
94 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095 else if (inside(input->block_bitmap, start, metaend))
Mingming Cao617ba132006-10-11 01:20:53 -070096 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070097 "Block bitmap (%llu) in GDT table"
Mingming Cao2ae02102006-10-11 01:21:11 -070098 " (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070099 (unsigned long long)input->block_bitmap,
100 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700101 else if (inside(input->inode_bitmap, start, metaend))
Mingming Cao617ba132006-10-11 01:20:53 -0700102 ext4_warning(sb, __FUNCTION__,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700103 "Inode bitmap (%llu) in GDT table"
Mingming Cao2ae02102006-10-11 01:21:11 -0700104 " (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700105 (unsigned long long)input->inode_bitmap,
106 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700107 else if (inside(input->inode_table, start, metaend) ||
108 inside(itend - 1, start, metaend))
Mingming Cao617ba132006-10-11 01:20:53 -0700109 ext4_warning(sb, __FUNCTION__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700110 "Inode table (%llu-%llu) overlaps"
111 "GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700112 (unsigned long long)input->inode_table,
113 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700114 else
115 err = 0;
116 brelse(bh);
117
118 return err;
119}
120
121static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700122 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700123{
124 struct buffer_head *bh;
125 int err;
126
127 bh = sb_getblk(sb, blk);
128 if (!bh)
129 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700130 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700131 brelse(bh);
132 bh = ERR_PTR(err);
133 } else {
134 lock_buffer(bh);
135 memset(bh->b_data, 0, sb->s_blocksize);
136 set_buffer_uptodate(bh);
137 unlock_buffer(bh);
138 }
139
140 return bh;
141}
142
143/*
Eric Sandeen14904102007-10-16 18:38:25 -0400144 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
145 * If that fails, restart the transaction & regain write access for the
146 * buffer head which is used for block_bitmap modifications.
147 */
148static int extend_or_restart_transaction(handle_t *handle, int thresh,
149 struct buffer_head *bh)
150{
151 int err;
152
153 if (handle->h_buffer_credits >= thresh)
154 return 0;
155
156 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
157 if (err < 0)
158 return err;
159 if (err) {
160 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
161 return err;
162 if ((err = ext4_journal_get_write_access(handle, bh)))
163 return err;
164 }
165
166 return 0;
167}
168
169/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700170 * Set up the block and inode bitmaps, and the inode table for the new group.
171 * This doesn't need to be part of the main transaction, since we are only
172 * changing blocks outside the actual filesystem. We still do journaling to
173 * ensure the recovery is correct in case of a failure just after resize.
174 * If any part of this fails, we simply abort the resize.
175 */
176static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700177 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700178{
Mingming Cao617ba132006-10-11 01:20:53 -0700179 struct ext4_sb_info *sbi = EXT4_SB(sb);
180 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
181 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700183 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184 struct buffer_head *bh;
185 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700186 ext4_fsblk_t block;
187 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700188 int i;
189 int err = 0, err2;
190
Eric Sandeen14904102007-10-16 18:38:25 -0400191 /* This transaction may be extended/restarted along the way */
192 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
193
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194 if (IS_ERR(handle))
195 return PTR_ERR(handle);
196
197 lock_super(sb);
198 if (input->group != sbi->s_groups_count) {
199 err = -EBUSY;
200 goto exit_journal;
201 }
202
203 if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
204 err = PTR_ERR(bh);
205 goto exit_journal;
206 }
207
Mingming Cao617ba132006-10-11 01:20:53 -0700208 if (ext4_bg_has_super(sb, input->group)) {
Eric Sandeenc549a952008-01-28 23:58:27 -0500209 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
Mingming Cao617ba132006-10-11 01:20:53 -0700210 ext4_set_bit(0, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211 }
212
213 /* Copy all of the GDT blocks into the backup in this group */
214 for (i = 0, bit = 1, block = start + 1;
215 i < gdblocks; i++, block++, bit++) {
216 struct buffer_head *gdb;
217
Eric Sandeenc549a952008-01-28 23:58:27 -0500218 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700219
Eric Sandeen14904102007-10-16 18:38:25 -0400220 if ((err = extend_or_restart_transaction(handle, 1, bh)))
221 goto exit_bh;
222
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700223 gdb = sb_getblk(sb, block);
224 if (!gdb) {
225 err = -EIO;
226 goto exit_bh;
227 }
Mingming Cao617ba132006-10-11 01:20:53 -0700228 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229 brelse(gdb);
230 goto exit_bh;
231 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400232 lock_buffer(gdb);
233 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 set_buffer_uptodate(gdb);
Eric Sandeen5b615282007-10-16 18:38:25 -0400235 unlock_buffer(gdb);
Mingming Cao617ba132006-10-11 01:20:53 -0700236 ext4_journal_dirty_metadata(handle, gdb);
237 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700238 brelse(gdb);
239 }
240
241 /* Zero out all of the reserved backup group descriptor table blocks */
242 for (i = 0, bit = gdblocks + 1, block = start + bit;
243 i < reserved_gdb; i++, block++, bit++) {
244 struct buffer_head *gdb;
245
Eric Sandeenc549a952008-01-28 23:58:27 -0500246 ext4_debug("clear reserved block %#04llx (+%d)\n", block, bit);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700247
Eric Sandeen14904102007-10-16 18:38:25 -0400248 if ((err = extend_or_restart_transaction(handle, 1, bh)))
249 goto exit_bh;
250
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251 if (IS_ERR(gdb = bclean(handle, sb, block))) {
252 err = PTR_ERR(bh);
253 goto exit_bh;
254 }
Mingming Cao617ba132006-10-11 01:20:53 -0700255 ext4_journal_dirty_metadata(handle, gdb);
256 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700257 brelse(gdb);
258 }
Eric Sandeenc549a952008-01-28 23:58:27 -0500259 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700261 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500262 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700264 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700265
266 /* Zero out all of the inode table blocks */
267 for (i = 0, block = input->inode_table, bit = block - start;
268 i < sbi->s_itb_per_group; i++, bit++, block++) {
269 struct buffer_head *it;
270
Eric Sandeenc549a952008-01-28 23:58:27 -0500271 ext4_debug("clear inode block %#04llx (+%d)\n", block, bit);
Eric Sandeen14904102007-10-16 18:38:25 -0400272
273 if ((err = extend_or_restart_transaction(handle, 1, bh)))
274 goto exit_bh;
275
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700276 if (IS_ERR(it = bclean(handle, sb, block))) {
277 err = PTR_ERR(it);
278 goto exit_bh;
279 }
Mingming Cao617ba132006-10-11 01:20:53 -0700280 ext4_journal_dirty_metadata(handle, it);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281 brelse(it);
Mingming Cao617ba132006-10-11 01:20:53 -0700282 ext4_set_bit(bit, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700283 }
Eric Sandeen14904102007-10-16 18:38:25 -0400284
285 if ((err = extend_or_restart_transaction(handle, 2, bh)))
286 goto exit_bh;
287
Mingming Cao617ba132006-10-11 01:20:53 -0700288 mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289 bh->b_data);
Mingming Cao617ba132006-10-11 01:20:53 -0700290 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 brelse(bh);
292
293 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500294 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 input->inode_bitmap, input->inode_bitmap - start);
296 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
297 err = PTR_ERR(bh);
298 goto exit_journal;
299 }
300
Mingming Cao617ba132006-10-11 01:20:53 -0700301 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700302 bh->b_data);
Mingming Cao617ba132006-10-11 01:20:53 -0700303 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700304exit_bh:
305 brelse(bh);
306
307exit_journal:
308 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700309 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700310 err = err2;
311
312 return err;
313}
314
315/*
316 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700317 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700318 * calling this for the first time. In a sparse filesystem it will be the
319 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
320 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
321 */
Mingming Cao617ba132006-10-11 01:20:53 -0700322static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323 unsigned *five, unsigned *seven)
324{
325 unsigned *min = three;
326 int mult = 3;
327 unsigned ret;
328
Mingming Cao617ba132006-10-11 01:20:53 -0700329 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
330 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700331 ret = *min;
332 *min += 1;
333 return ret;
334 }
335
336 if (*five < *min) {
337 min = five;
338 mult = 5;
339 }
340 if (*seven < *min) {
341 min = seven;
342 mult = 7;
343 }
344
345 ret = *min;
346 *min *= mult;
347
348 return ret;
349}
350
351/*
352 * Check that all of the backup GDT blocks are held in the primary GDT block.
353 * It is assumed that they are stored in group order. Returns the number of
354 * groups in current filesystem that have BACKUPS, or -ve error code.
355 */
356static int verify_reserved_gdb(struct super_block *sb,
357 struct buffer_head *primary)
358{
Mingming Cao617ba132006-10-11 01:20:53 -0700359 const ext4_fsblk_t blk = primary->b_blocknr;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500360 const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700361 unsigned three = 1;
362 unsigned five = 5;
363 unsigned seven = 7;
364 unsigned grp;
365 __le32 *p = (__le32 *)primary->b_data;
366 int gdbackups = 0;
367
Mingming Cao617ba132006-10-11 01:20:53 -0700368 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700369 if (le32_to_cpu(*p++) !=
370 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Mingming Cao617ba132006-10-11 01:20:53 -0700371 ext4_warning(sb, __FUNCTION__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700372 "reserved GDT %llu"
373 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700374 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700375 grp *
376 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
377 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700378 return -EINVAL;
379 }
Mingming Cao617ba132006-10-11 01:20:53 -0700380 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700381 return -EFBIG;
382 }
383
384 return gdbackups;
385}
386
387/*
388 * Called when we need to bring a reserved group descriptor table block into
389 * use from the resize inode. The primary copy of the new GDT block currently
390 * is an indirect block (under the double indirect block in the resize inode).
391 * The new backup GDT blocks will be stored as leaf blocks in this indirect
392 * block, in group order. Even though we know all the block numbers we need,
393 * we check to ensure that the resize inode has actually reserved these blocks.
394 *
395 * Don't need to update the block bitmaps because the blocks are still in use.
396 *
397 * We get all of the error cases out of the way, so that we are sure to not
398 * fail once we start modifying the data on disk, because JBD has no rollback.
399 */
400static int add_new_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700401 struct ext4_new_group_data *input,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700402 struct buffer_head **primary)
403{
404 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700405 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
406 unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
407 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700408 struct buffer_head **o_group_desc, **n_group_desc;
409 struct buffer_head *dind;
410 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700411 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700412 __le32 *data;
413 int err;
414
415 if (test_opt(sb, DEBUG))
416 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700417 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700418 gdb_num);
419
420 /*
421 * If we are not using the primary superblock/GDT copy don't resize,
422 * because the user tools have no way of handling this. Probably a
423 * bad time to do it anyways.
424 */
Mingming Cao617ba132006-10-11 01:20:53 -0700425 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
426 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
427 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700428 "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700429 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700430 return -EPERM;
431 }
432
433 *primary = sb_bread(sb, gdblock);
434 if (!*primary)
435 return -EIO;
436
437 if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
438 err = gdbackups;
439 goto exit_bh;
440 }
441
Mingming Cao617ba132006-10-11 01:20:53 -0700442 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700443 dind = sb_bread(sb, le32_to_cpu(*data));
444 if (!dind) {
445 err = -EIO;
446 goto exit_bh;
447 }
448
449 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700450 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
451 ext4_warning(sb, __FUNCTION__,
Mingming Cao2ae02102006-10-11 01:21:11 -0700452 "new group %u GDT block %llu not reserved",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700453 input->group, gdblock);
454 err = -EINVAL;
455 goto exit_dind;
456 }
457
Mingming Cao617ba132006-10-11 01:20:53 -0700458 if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700459 goto exit_dind;
460
Mingming Cao617ba132006-10-11 01:20:53 -0700461 if ((err = ext4_journal_get_write_access(handle, *primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700462 goto exit_sbh;
463
Mingming Cao617ba132006-10-11 01:20:53 -0700464 if ((err = ext4_journal_get_write_access(handle, dind)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700465 goto exit_primary;
466
Mingming Cao617ba132006-10-11 01:20:53 -0700467 /* ext4_reserve_inode_write() gets a reference on the iloc */
468 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700469 goto exit_dindj;
470
471 n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
472 GFP_KERNEL);
473 if (!n_group_desc) {
474 err = -ENOMEM;
Mingming Cao617ba132006-10-11 01:20:53 -0700475 ext4_warning (sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700476 "not enough memory for %lu groups", gdb_num + 1);
477 goto exit_inode;
478 }
479
480 /*
481 * Finally, we have all of the possible failures behind us...
482 *
483 * Remove new GDT block from inode double-indirect block and clear out
484 * the new GDT block for use (which also "frees" the backup GDT blocks
485 * from the reserved inode). We don't need to change the bitmaps for
486 * these blocks, because they are marked as in-use from being in the
487 * reserved inode, and will become GDT blocks (primary and backup).
488 */
Mingming Cao617ba132006-10-11 01:20:53 -0700489 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
490 ext4_journal_dirty_metadata(handle, dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700491 brelse(dind);
492 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700493 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 memset((*primary)->b_data, 0, sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700495 ext4_journal_dirty_metadata(handle, *primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496
Mingming Cao617ba132006-10-11 01:20:53 -0700497 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700498 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700499 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 n_group_desc[gdb_num] = *primary;
Mingming Cao617ba132006-10-11 01:20:53 -0700501 EXT4_SB(sb)->s_group_desc = n_group_desc;
502 EXT4_SB(sb)->s_gdb_count++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503 kfree(o_group_desc);
504
Marcin Slusarze8546d02008-04-17 10:38:59 -0400505 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Mingming Cao617ba132006-10-11 01:20:53 -0700506 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700507
508 return 0;
509
510exit_inode:
Mingming Cao617ba132006-10-11 01:20:53 -0700511 //ext4_journal_release_buffer(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700512 brelse(iloc.bh);
513exit_dindj:
Mingming Cao617ba132006-10-11 01:20:53 -0700514 //ext4_journal_release_buffer(handle, dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700515exit_primary:
Mingming Cao617ba132006-10-11 01:20:53 -0700516 //ext4_journal_release_buffer(handle, *primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517exit_sbh:
Mingming Cao617ba132006-10-11 01:20:53 -0700518 //ext4_journal_release_buffer(handle, *primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519exit_dind:
520 brelse(dind);
521exit_bh:
522 brelse(*primary);
523
Mingming Cao617ba132006-10-11 01:20:53 -0700524 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700525 return err;
526}
527
528/*
529 * Called when we are adding a new group which has a backup copy of each of
530 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
531 * We need to add these reserved backup GDT blocks to the resize inode, so
532 * that they are kept for future resizing and not allocated to files.
533 *
534 * Each reserved backup GDT block will go into a different indirect block.
535 * The indirect blocks are actually the primary reserved GDT blocks,
536 * so we know in advance what their block numbers are. We only get the
537 * double-indirect block to verify it is pointing to the primary reserved
538 * GDT blocks so we don't overwrite a data block by accident. The reserved
539 * backup GDT blocks are stored in their reserved primary GDT block.
540 */
541static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Mingming Cao617ba132006-10-11 01:20:53 -0700542 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700543{
544 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700545 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700546 struct buffer_head **primary;
547 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700548 struct ext4_iloc iloc;
549 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700550 __le32 *data, *end;
551 int gdbackups = 0;
552 int res, i;
553 int err;
554
555 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL);
556 if (!primary)
557 return -ENOMEM;
558
Mingming Cao617ba132006-10-11 01:20:53 -0700559 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700560 dind = sb_bread(sb, le32_to_cpu(*data));
561 if (!dind) {
562 err = -EIO;
563 goto exit_free;
564 }
565
Mingming Cao617ba132006-10-11 01:20:53 -0700566 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
567 data = (__le32 *)dind->b_data + EXT4_SB(sb)->s_gdb_count;
568 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) {
Mingming Cao617ba132006-10-11 01:20:53 -0700573 ext4_warning(sb, __FUNCTION__,
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) {
Mingming Cao617ba132006-10-11 01:20:53 -0700717 ext4_warning(sb, __FUNCTION__,
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)) {
757 ext4_warning(sb, __FUNCTION__,
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)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700764 ext4_warning(sb, __FUNCTION__, "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)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700770 ext4_warning(sb, __FUNCTION__, "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,
776 EXT4_FEATURE_COMPAT_RESIZE_INODE)){
777 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778 "No reserved GDT blocks, can't resize");
779 return -EPERM;
780 }
David Howells1d1fe1e2008-02-07 00:15:37 -0800781 inode = ext4_iget(sb, EXT4_RESIZE_INO);
782 if (IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700783 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700784 "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -0800785 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 }
787 }
788
789 if ((err = verify_group_input(sb, input)))
790 goto exit_put;
791
792 if ((err = setup_new_group_blocks(sb, input)))
793 goto exit_put;
794
795 /*
796 * We will always be modifying at least the superblock and a GDT
797 * block. If we are adding a group past the last current GDT block,
798 * we will also modify the inode and the dindirect block. If we
799 * are adding a group with superblock/GDT backups we will also
800 * modify each of the reserved GDT dindirect blocks.
801 */
Mingming Cao617ba132006-10-11 01:20:53 -0700802 handle = ext4_journal_start_sb(sb,
803 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804 3 + reserved_gdb : 4);
805 if (IS_ERR(handle)) {
806 err = PTR_ERR(handle);
807 goto exit_put;
808 }
809
810 lock_super(sb);
811 if (input->group != sbi->s_groups_count) {
Mingming Cao617ba132006-10-11 01:20:53 -0700812 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813 "multiple resizers run on filesystem!");
814 err = -EBUSY;
815 goto exit_journal;
816 }
817
Mingming Cao617ba132006-10-11 01:20:53 -0700818 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819 goto exit_journal;
820
821 /*
822 * We will only either add reserved group blocks to a backup group
823 * or remove reserved blocks for the first group in a new group block.
824 * Doing both would be mean more complex code, and sane people don't
825 * use non-sparse filesystems anymore. This is already checked above.
826 */
827 if (gdb_off) {
828 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -0700829 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700830 goto exit_journal;
831
Mingming Cao617ba132006-10-11 01:20:53 -0700832 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833 (err = reserve_backup_gdb(handle, inode, input)))
834 goto exit_journal;
835 } else if ((err = add_new_gdb(handle, inode, input, &primary)))
836 goto exit_journal;
837
838 /*
839 * OK, now we've set up the new group. Time to make it active.
840 *
841 * Current kernels don't lock all allocations via lock_super(),
842 * so we have to be safe wrt. concurrent accesses the group
843 * data. So we need to be careful to set all of the relevant
844 * group descriptor data etc. *before* we enable the group.
845 *
846 * The key field here is sbi->s_groups_count: as long as
847 * that retains its old value, nobody is going to access the new
848 * group.
849 *
850 * So first we update all the descriptor metadata for the new
851 * group; then we update the total disk blocks count; then we
852 * update the groups count to enable the group; then finally we
853 * update the free space counts so that the system can start
854 * using the new disk blocks.
855 */
856
857 /* Update group descriptor block for new group */
Mingming Cao617ba132006-10-11 01:20:53 -0700858 gdp = (struct ext4_group_desc *)primary->b_data + gdb_off;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859
Alexandre Ratchov8fadc142006-10-11 01:21:15 -0700860 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
861 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
862 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700863 gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
Mingming Cao617ba132006-10-11 01:20:53 -0700864 gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
Andreas Dilger717d50e2007-10-16 18:38:25 -0400865 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700866
867 /*
868 * Make the new blocks and inodes valid next. We do this before
869 * increasing the group count so that once the group is enabled,
870 * all of its blocks and inodes are already valid.
871 *
872 * We always allocate group-by-group, then block-by-block or
873 * inode-by-inode within a group, so enabling these
874 * blocks/inodes before the group is live won't actually let us
875 * allocate the new space yet.
876 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700877 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700878 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400879 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880
881 /*
882 * We need to protect s_groups_count against other CPUs seeing
883 * inconsistent state in the superblock.
884 *
885 * The precise rules we use are:
886 *
887 * * Writers of s_groups_count *must* hold lock_super
888 * AND
889 * * Writers must perform a smp_wmb() after updating all dependent
890 * data and before modifying the groups count
891 *
892 * * Readers must hold lock_super() over the access
893 * OR
894 * * Readers must perform an smp_rmb() after reading the groups count
895 * and before reading any dependent data.
896 *
897 * NB. These rules can be relaxed when checking the group count
898 * while freeing data, as we can only allocate from a block
899 * group after serialising against the group count, and we can
900 * only then free after serialising in turn against that
901 * allocation.
902 */
903 smp_wmb();
904
905 /* Update the global fs size fields */
906 sbi->s_groups_count++;
907
Mingming Cao617ba132006-10-11 01:20:53 -0700908 ext4_journal_dirty_metadata(handle, primary);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909
910 /* Update the reserved block counts only once the new group is
911 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700912 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700913 input->reserved_blocks);
914
915 /* Update the free space counts */
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700916 percpu_counter_add(&sbi->s_freeblocks_counter,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700917 input->free_blocks_count);
Peter Zijlstraaa0dff22007-10-16 23:25:42 -0700918 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -0700919 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920
Mingming Cao617ba132006-10-11 01:20:53 -0700921 ext4_journal_dirty_metadata(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700922 sb->s_dirt = 1;
923
924exit_journal:
925 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700926 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700927 err = err2;
928 if (!err) {
929 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -0700930 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 update_backups(sb, primary->b_blocknr, primary->b_data,
932 primary->b_size);
933 }
934exit_put:
935 iput(inode);
936 return err;
Mingming Cao617ba132006-10-11 01:20:53 -0700937} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700938
939/* Extend the filesystem to the new number of blocks specified. This entry
940 * point is only used to extend the current filesystem to the end of the last
941 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
942 * for emergencies (because it has no dependencies on reserved blocks).
943 *
Mingming Cao617ba132006-10-11 01:20:53 -0700944 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700945 * allow the "remount" trick to work for arbitrary resizing, assuming enough
946 * GDT blocks are reserved to grow to the desired size.
947 */
Mingming Cao617ba132006-10-11 01:20:53 -0700948int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
949 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950{
Mingming Cao617ba132006-10-11 01:20:53 -0700951 ext4_fsblk_t o_blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -0500952 ext4_group_t o_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -0700953 ext4_grpblk_t last;
954 ext4_grpblk_t add;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700955 struct buffer_head * bh;
956 handle_t *handle;
957 int err;
958 unsigned long freed_blocks;
959
960 /* We don't need to worry about locking wrt other resizers just
961 * yet: we're going to revalidate es->s_blocks_count after
962 * taking lock_super() below. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700963 o_blocks_count = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -0700964 o_groups_count = EXT4_SB(sb)->s_groups_count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700965
966 if (test_opt(sb, DEBUG))
Mingming Cao2ae02102006-10-11 01:21:11 -0700967 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968 o_blocks_count, n_blocks_count);
969
970 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
971 return 0;
972
973 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700974 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -0700975 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976 sb->s_id, n_blocks_count);
977 if (sizeof(sector_t) < 8)
Mingming Cao617ba132006-10-11 01:20:53 -0700978 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700979 "CONFIG_LBD not enabled\n");
980 return -EINVAL;
981 }
982
983 if (n_blocks_count < o_blocks_count) {
Mingming Cao617ba132006-10-11 01:20:53 -0700984 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 "can't shrink FS - resize aborted");
986 return -EBUSY;
987 }
988
989 /* Handle the remaining blocks in the last group only. */
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700990 ext4_get_group_no_and_offset(sb, o_blocks_count, NULL, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700991
992 if (last == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -0700993 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 "need to use ext2online to resize further");
995 return -EPERM;
996 }
997
Mingming Cao617ba132006-10-11 01:20:53 -0700998 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999
1000 if (o_blocks_count + add < o_blocks_count) {
Mingming Cao617ba132006-10-11 01:20:53 -07001001 ext4_warning(sb, __FUNCTION__, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001002 return -EINVAL;
1003 }
1004
1005 if (o_blocks_count + add > n_blocks_count)
1006 add = n_blocks_count - o_blocks_count;
1007
1008 if (o_blocks_count + add < n_blocks_count)
Mingming Cao617ba132006-10-11 01:20:53 -07001009 ext4_warning(sb, __FUNCTION__,
Mingming Cao2ae02102006-10-11 01:21:11 -07001010 "will only finish group (%llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011 " blocks, %u new)",
1012 o_blocks_count + add, add);
1013
1014 /* See if the device is actually as big as what was requested */
1015 bh = sb_bread(sb, o_blocks_count + add -1);
1016 if (!bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07001017 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018 "can't read last block, resize aborted");
1019 return -ENOSPC;
1020 }
1021 brelse(bh);
1022
1023 /* We will update the superblock, one block bitmap, and
Mingming Cao617ba132006-10-11 01:20:53 -07001024 * one group descriptor via ext4_free_blocks().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025 */
Mingming Cao617ba132006-10-11 01:20:53 -07001026 handle = ext4_journal_start_sb(sb, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 if (IS_ERR(handle)) {
1028 err = PTR_ERR(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07001029 ext4_warning(sb, __FUNCTION__, "error %d on journal start",err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 goto exit_put;
1031 }
1032
1033 lock_super(sb);
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001034 if (o_blocks_count != ext4_blocks_count(es)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001035 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001036 "multiple resizers run on filesystem!");
1037 unlock_super(sb);
Akinobu Mita5606bf52008-02-25 15:37:42 -05001038 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 err = -EBUSY;
1040 goto exit_put;
1041 }
1042
Mingming Cao617ba132006-10-11 01:20:53 -07001043 if ((err = ext4_journal_get_write_access(handle,
1044 EXT4_SB(sb)->s_sbh))) {
1045 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 "error %d on journal write access", err);
1047 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001048 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001049 goto exit_put;
1050 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001051 ext4_blocks_count_set(es, o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001052 ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053 sb->s_dirt = 1;
1054 unlock_super(sb);
Eric Sandeenc549a952008-01-28 23:58:27 -05001055 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056 o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001057 ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
Mingming Cao2ae02102006-10-11 01:21:11 -07001058 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059 o_blocks_count + add);
Mingming Cao617ba132006-10-11 01:20:53 -07001060 if ((err = ext4_journal_stop(handle)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001061 goto exit_put;
1062 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001063 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1064 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001065 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1066 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001067exit_put:
1068 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001069} /* ext4_group_extend */