blob: dac23561f3eb8e171843150579bd82f2e397a780 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/resize.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
Mingming Cao617ba132006-10-11 01:20:53 -07004 * Support for resizing an ext4 filesystem while it is mounted.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005 *
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7 *
8 * This could probably be made into a module, because it is not often in use.
9 */
10
11
Mingming Cao617ba132006-10-11 01:20:53 -070012#define EXT4FS_DEBUG
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013
Dave Kleikampac27a0e2006-10-11 01:20:50 -070014#include <linux/errno.h>
15#include <linux/slab.h>
16
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040017#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018
Yongqiang Yang8f82f842011-07-26 21:35:44 -040019int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
Yongqiang Yangce723c32011-07-26 21:39:09 -040026 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
Yongqiang Yang8f82f842011-07-26 21:35:44 -040036 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
Dave Kleikampac27a0e2006-10-11 01:20:50 -070048#define outside(b, first, last) ((b) < (first) || (b) >= (last))
49#define inside(b, first, last) ((b) >= (first) && (b) < (last))
50
51static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070052 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053{
Mingming Cao617ba132006-10-11 01:20:53 -070054 struct ext4_sb_info *sbi = EXT4_SB(sb);
55 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070056 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070057 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050058 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070059 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
60 unsigned overhead = ext4_bg_has_super(sb, group) ?
61 (1 + ext4_bg_num_gdb(sb, group) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -070063 ext4_fsblk_t metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070065 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070066 int err = -EINVAL;
67
68 input->free_blocks_count = free_blocks_count =
69 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
70
71 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070072 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070073 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070074 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075 "no-super", input->group, input->blocks_count,
76 free_blocks_count, input->reserved_blocks);
77
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070078 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070079 if (group != sbi->s_groups_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -050080 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070082 else if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050083 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -050085 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086 input->reserved_blocks);
87 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -050088 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070089 input->blocks_count);
90 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -050091 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -070092 end - 1);
93 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050094 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070095 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -050097 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -070098 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400100 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500101 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700102 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500104 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700105 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500107 ext4_warning(sb, "Block bitmap (%llu) in inode table "
108 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700109 (unsigned long long)input->block_bitmap,
110 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700111 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500112 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
113 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700114 (unsigned long long)input->inode_bitmap,
115 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500117 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700118 (unsigned long long)input->block_bitmap,
119 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500121 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700122 (unsigned long long)input->inode_bitmap,
123 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700124 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400125 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500126 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
127 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700128 (unsigned long long)input->inode_table,
129 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130 else
131 err = 0;
132 brelse(bh);
133
134 return err;
135}
136
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500137/*
138 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
139 * group each time.
140 */
141struct ext4_new_flex_group_data {
142 struct ext4_new_group_data *groups; /* new_group_data for groups
143 in the flex group */
144 __u16 *bg_flags; /* block group flags of groups
145 in @groups */
146 ext4_group_t count; /* number of groups in @groups
147 */
148};
149
150/*
151 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
152 * @flexbg_size.
153 *
154 * Returns NULL on failure otherwise address of the allocated structure.
155 */
156static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
157{
158 struct ext4_new_flex_group_data *flex_gd;
159
160 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
161 if (flex_gd == NULL)
162 goto out3;
163
164 flex_gd->count = flexbg_size;
165
166 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
167 flexbg_size, GFP_NOFS);
168 if (flex_gd->groups == NULL)
169 goto out2;
170
171 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
172 if (flex_gd->bg_flags == NULL)
173 goto out1;
174
175 return flex_gd;
176
177out1:
178 kfree(flex_gd->groups);
179out2:
180 kfree(flex_gd);
181out3:
182 return NULL;
183}
184
185static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
186{
187 kfree(flex_gd->bg_flags);
188 kfree(flex_gd->groups);
189 kfree(flex_gd);
190}
191
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500192/*
193 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
194 * and inode tables for a flex group.
195 *
196 * This function is used by 64bit-resize. Note that this function allocates
197 * group tables from the 1st group of groups contained by @flexgd, which may
198 * be a partial of a flex group.
199 *
200 * @sb: super block of fs to which the groups belongs
201 */
202static void ext4_alloc_group_tables(struct super_block *sb,
203 struct ext4_new_flex_group_data *flex_gd,
204 int flexbg_size)
205{
206 struct ext4_new_group_data *group_data = flex_gd->groups;
207 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
208 ext4_fsblk_t start_blk;
209 ext4_fsblk_t last_blk;
210 ext4_group_t src_group;
211 ext4_group_t bb_index = 0;
212 ext4_group_t ib_index = 0;
213 ext4_group_t it_index = 0;
214 ext4_group_t group;
215 ext4_group_t last_group;
216 unsigned overhead;
217
218 BUG_ON(flex_gd->count == 0 || group_data == NULL);
219
220 src_group = group_data[0].group;
221 last_group = src_group + flex_gd->count - 1;
222
223 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
224 (last_group & ~(flexbg_size - 1))));
225next_group:
226 group = group_data[0].group;
227 start_blk = ext4_group_first_block_no(sb, src_group);
228 last_blk = start_blk + group_data[src_group - group].blocks_count;
229
230 overhead = ext4_bg_has_super(sb, src_group) ?
231 (1 + ext4_bg_num_gdb(sb, src_group) +
232 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
233
234 start_blk += overhead;
235
236 BUG_ON(src_group >= group_data[0].group + flex_gd->count);
237 /* We collect contiguous blocks as much as possible. */
238 src_group++;
239 for (; src_group <= last_group; src_group++)
240 if (!ext4_bg_has_super(sb, src_group))
241 last_blk += group_data[src_group - group].blocks_count;
242 else
243 break;
244
245 /* Allocate block bitmaps */
246 for (; bb_index < flex_gd->count; bb_index++) {
247 if (start_blk >= last_blk)
248 goto next_group;
249 group_data[bb_index].block_bitmap = start_blk++;
250 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
251 group -= group_data[0].group;
252 group_data[group].free_blocks_count--;
253 if (flexbg_size > 1)
254 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
255 }
256
257 /* Allocate inode bitmaps */
258 for (; ib_index < flex_gd->count; ib_index++) {
259 if (start_blk >= last_blk)
260 goto next_group;
261 group_data[ib_index].inode_bitmap = start_blk++;
262 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
263 group -= group_data[0].group;
264 group_data[group].free_blocks_count--;
265 if (flexbg_size > 1)
266 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
267 }
268
269 /* Allocate inode tables */
270 for (; it_index < flex_gd->count; it_index++) {
271 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
272 goto next_group;
273 group_data[it_index].inode_table = start_blk;
274 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
275 group -= group_data[0].group;
276 group_data[group].free_blocks_count -=
277 EXT4_SB(sb)->s_itb_per_group;
278 if (flexbg_size > 1)
279 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
280
281 start_blk += EXT4_SB(sb)->s_itb_per_group;
282 }
283
284 if (test_opt(sb, DEBUG)) {
285 int i;
286 group = group_data[0].group;
287
288 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
289 "%d groups, flexbg size is %d:\n", flex_gd->count,
290 flexbg_size);
291
292 for (i = 0; i < flex_gd->count; i++) {
293 printk(KERN_DEBUG "adding %s group %u: %u "
294 "blocks (%d free)\n",
295 ext4_bg_has_super(sb, group + i) ? "normal" :
296 "no-super", group + i,
297 group_data[i].blocks_count,
298 group_data[i].free_blocks_count);
299 }
300 }
301}
302
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700304 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700305{
306 struct buffer_head *bh;
307 int err;
308
309 bh = sb_getblk(sb, blk);
310 if (!bh)
311 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700312 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700313 brelse(bh);
314 bh = ERR_PTR(err);
315 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316 memset(bh->b_data, 0, sb->s_blocksize);
317 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700318 }
319
320 return bh;
321}
322
323/*
Eric Sandeen14904102007-10-16 18:38:25 -0400324 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
325 * If that fails, restart the transaction & regain write access for the
326 * buffer head which is used for block_bitmap modifications.
327 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400328static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400329{
330 int err;
331
Frank Mayhar03901312009-01-07 00:06:22 -0500332 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400333 return 0;
334
335 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
336 if (err < 0)
337 return err;
338 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400339 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
340 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400341 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400342 }
Eric Sandeen14904102007-10-16 18:38:25 -0400343
344 return 0;
345}
346
347/*
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500348 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
349 *
350 * Helper function for ext4_setup_new_group_blocks() which set .
351 *
352 * @sb: super block
353 * @handle: journal handle
354 * @flex_gd: flex group data
355 */
356static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
357 struct ext4_new_flex_group_data *flex_gd,
358 ext4_fsblk_t block, ext4_group_t count)
359{
360 ext4_group_t count2;
361
362 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
363 for (count2 = count; count > 0; count -= count2, block += count2) {
364 ext4_fsblk_t start;
365 struct buffer_head *bh;
366 ext4_group_t group;
367 int err;
368
369 ext4_get_group_no_and_offset(sb, block, &group, NULL);
370 start = ext4_group_first_block_no(sb, group);
371 group -= flex_gd->groups[0].group;
372
373 count2 = sb->s_blocksize * 8 - (block - start);
374 if (count2 > count)
375 count2 = count;
376
377 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
378 BUG_ON(flex_gd->count > 1);
379 continue;
380 }
381
382 err = extend_or_restart_transaction(handle, 1);
383 if (err)
384 return err;
385
386 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
387 if (!bh)
388 return -EIO;
389
390 err = ext4_journal_get_write_access(handle, bh);
391 if (err)
392 return err;
393 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
394 block - start, count2);
395 ext4_set_bits(bh->b_data, block - start, count2);
396
397 err = ext4_handle_dirty_metadata(handle, NULL, bh);
398 if (unlikely(err))
399 return err;
400 brelse(bh);
401 }
402
403 return 0;
404}
405
406/*
407 * Set up the block and inode bitmaps, and the inode table for the new groups.
408 * This doesn't need to be part of the main transaction, since we are only
409 * changing blocks outside the actual filesystem. We still do journaling to
410 * ensure the recovery is correct in case of a failure just after resize.
411 * If any part of this fails, we simply abort the resize.
412 *
413 * setup_new_flex_group_blocks handles a flex group as follow:
414 * 1. copy super block and GDT, and initialize group tables if necessary.
415 * In this step, we only set bits in blocks bitmaps for blocks taken by
416 * super block and GDT.
417 * 2. allocate group tables in block bitmaps, that is, set bits in block
418 * bitmap for blocks taken by group tables.
419 */
420static int setup_new_flex_group_blocks(struct super_block *sb,
421 struct ext4_new_flex_group_data *flex_gd)
422{
423 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
424 ext4_fsblk_t start;
425 ext4_fsblk_t block;
426 struct ext4_sb_info *sbi = EXT4_SB(sb);
427 struct ext4_super_block *es = sbi->s_es;
428 struct ext4_new_group_data *group_data = flex_gd->groups;
429 __u16 *bg_flags = flex_gd->bg_flags;
430 handle_t *handle;
431 ext4_group_t group, count;
432 struct buffer_head *bh = NULL;
433 int reserved_gdb, i, j, err = 0, err2;
434
435 BUG_ON(!flex_gd->count || !group_data ||
436 group_data[0].group != sbi->s_groups_count);
437
438 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
439
440 /* This transaction may be extended/restarted along the way */
441 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
442 if (IS_ERR(handle))
443 return PTR_ERR(handle);
444
445 group = group_data[0].group;
446 for (i = 0; i < flex_gd->count; i++, group++) {
447 unsigned long gdblocks;
448
449 gdblocks = ext4_bg_num_gdb(sb, group);
450 start = ext4_group_first_block_no(sb, group);
451
452 /* Copy all of the GDT blocks into the backup in this group */
453 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
454 struct buffer_head *gdb;
455
456 ext4_debug("update backup group %#04llx\n", block);
457 err = extend_or_restart_transaction(handle, 1);
458 if (err)
459 goto out;
460
461 gdb = sb_getblk(sb, block);
462 if (!gdb) {
463 err = -EIO;
464 goto out;
465 }
466
467 err = ext4_journal_get_write_access(handle, gdb);
468 if (err) {
469 brelse(gdb);
470 goto out;
471 }
472 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
473 gdb->b_size);
474 set_buffer_uptodate(gdb);
475
476 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
477 if (unlikely(err)) {
478 brelse(gdb);
479 goto out;
480 }
481 brelse(gdb);
482 }
483
484 /* Zero out all of the reserved backup group descriptor
485 * table blocks
486 */
487 if (ext4_bg_has_super(sb, group)) {
488 err = sb_issue_zeroout(sb, gdblocks + start + 1,
489 reserved_gdb, GFP_NOFS);
490 if (err)
491 goto out;
492 }
493
494 /* Initialize group tables of the grop @group */
495 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
496 goto handle_bb;
497
498 /* Zero out all of the inode table blocks */
499 block = group_data[i].inode_table;
500 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
501 block, sbi->s_itb_per_group);
502 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
503 GFP_NOFS);
504 if (err)
505 goto out;
506
507handle_bb:
508 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
509 goto handle_ib;
510
511 /* Initialize block bitmap of the @group */
512 block = group_data[i].block_bitmap;
513 err = extend_or_restart_transaction(handle, 1);
514 if (err)
515 goto out;
516
517 bh = bclean(handle, sb, block);
518 if (IS_ERR(bh)) {
519 err = PTR_ERR(bh);
520 goto out;
521 }
522 if (ext4_bg_has_super(sb, group)) {
523 ext4_debug("mark backup superblock %#04llx (+0)\n",
524 start);
525 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
526 1);
527 }
528 ext4_mark_bitmap_end(group_data[i].blocks_count,
529 sb->s_blocksize * 8, bh->b_data);
530 err = ext4_handle_dirty_metadata(handle, NULL, bh);
531 if (err)
532 goto out;
533 brelse(bh);
534
535handle_ib:
536 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
537 continue;
538
539 /* Initialize inode bitmap of the @group */
540 block = group_data[i].inode_bitmap;
541 err = extend_or_restart_transaction(handle, 1);
542 if (err)
543 goto out;
544 /* Mark unused entries in inode bitmap used */
545 bh = bclean(handle, sb, block);
546 if (IS_ERR(bh)) {
547 err = PTR_ERR(bh);
548 goto out;
549 }
550
551 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
552 sb->s_blocksize * 8, bh->b_data);
553 err = ext4_handle_dirty_metadata(handle, NULL, bh);
554 if (err)
555 goto out;
556 brelse(bh);
557 }
558 bh = NULL;
559
560 /* Mark group tables in block bitmap */
561 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
562 count = group_table_count[j];
563 start = (&group_data[0].block_bitmap)[j];
564 block = start;
565 for (i = 1; i < flex_gd->count; i++) {
566 block += group_table_count[j];
567 if (block == (&group_data[i].block_bitmap)[j]) {
568 count += group_table_count[j];
569 continue;
570 }
571 err = set_flexbg_block_bitmap(sb, handle,
572 flex_gd, start, count);
573 if (err)
574 goto out;
575 count = group_table_count[j];
576 start = group_data[i].block_bitmap;
577 block = start;
578 }
579
580 if (count) {
581 err = set_flexbg_block_bitmap(sb, handle,
582 flex_gd, start, count);
583 if (err)
584 goto out;
585 }
586 }
587
588out:
589 brelse(bh);
590 err2 = ext4_journal_stop(handle);
591 if (err2 && !err)
592 err = err2;
593
594 return err;
595}
596
597/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700598 * Set up the block and inode bitmaps, and the inode table for the new group.
599 * This doesn't need to be part of the main transaction, since we are only
600 * changing blocks outside the actual filesystem. We still do journaling to
601 * ensure the recovery is correct in case of a failure just after resize.
602 * If any part of this fails, we simply abort the resize.
603 */
604static int setup_new_group_blocks(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700605 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700606{
Mingming Cao617ba132006-10-11 01:20:53 -0700607 struct ext4_sb_info *sbi = EXT4_SB(sb);
608 ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
609 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700610 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
Mingming Cao617ba132006-10-11 01:20:53 -0700611 unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700612 struct buffer_head *bh;
613 handle_t *handle;
Mingming Cao617ba132006-10-11 01:20:53 -0700614 ext4_fsblk_t block;
615 ext4_grpblk_t bit;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700616 int i;
617 int err = 0, err2;
618
Eric Sandeen14904102007-10-16 18:38:25 -0400619 /* This transaction may be extended/restarted along the way */
620 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
621
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 if (IS_ERR(handle))
623 return PTR_ERR(handle);
624
Yongqiang Yang8f82f842011-07-26 21:35:44 -0400625 BUG_ON(input->group != sbi->s_groups_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700626
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700627 /* Copy all of the GDT blocks into the backup in this group */
628 for (i = 0, bit = 1, block = start + 1;
629 i < gdblocks; i++, block++, bit++) {
630 struct buffer_head *gdb;
631
Eric Sandeenc549a952008-01-28 23:58:27 -0500632 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400633 err = extend_or_restart_transaction(handle, 1);
634 if (err)
635 goto exit_journal;
Eric Sandeen14904102007-10-16 18:38:25 -0400636
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700637 gdb = sb_getblk(sb, block);
638 if (!gdb) {
639 err = -EIO;
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400640 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700641 }
Mingming Cao617ba132006-10-11 01:20:53 -0700642 if ((err = ext4_journal_get_write_access(handle, gdb))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700643 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400644 goto exit_journal;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700645 }
Eric Sandeen5b615282007-10-16 18:38:25 -0400646 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700647 set_buffer_uptodate(gdb);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500648 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
649 if (unlikely(err)) {
650 brelse(gdb);
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400651 goto exit_journal;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500652 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700653 brelse(gdb);
654 }
655
656 /* Zero out all of the reserved backup group descriptor table blocks */
Theodore Ts'oda488942011-02-21 20:39:58 -0500657 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400658 block, sbi->s_itb_per_group);
659 err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400660 GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400661 if (err)
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400662 goto exit_journal;
663
664 err = extend_or_restart_transaction(handle, 2);
665 if (err)
666 goto exit_journal;
667
668 bh = bclean(handle, sb, input->block_bitmap);
669 if (IS_ERR(bh)) {
670 err = PTR_ERR(bh);
671 goto exit_journal;
672 }
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400673
674 if (ext4_bg_has_super(sb, input->group)) {
675 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
676 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
677 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700678
Eric Sandeenc549a952008-01-28 23:58:27 -0500679 ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700680 input->block_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700681 ext4_set_bit(input->block_bitmap - start, bh->b_data);
Eric Sandeenc549a952008-01-28 23:58:27 -0500682 ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700683 input->inode_bitmap - start);
Mingming Cao617ba132006-10-11 01:20:53 -0700684 ext4_set_bit(input->inode_bitmap - start, bh->b_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700685
686 /* Zero out all of the inode table blocks */
Lukas Czernera31437b2010-10-27 21:30:05 -0400687 block = input->inode_table;
Theodore Ts'oda488942011-02-21 20:39:58 -0500688 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
Lukas Czernera31437b2010-10-27 21:30:05 -0400689 block, sbi->s_itb_per_group);
Theodore Ts'oa107e5a2010-10-27 23:44:47 -0400690 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
Lukas Czernera31437b2010-10-27 21:30:05 -0400691 if (err)
692 goto exit_bh;
Yongqiang Yangc3e94d12011-07-26 22:05:53 -0400693 ext4_set_bits(bh->b_data, input->inode_table - start,
694 sbi->s_itb_per_group);
Eric Sandeen14904102007-10-16 18:38:25 -0400695
Eric Sandeen14904102007-10-16 18:38:25 -0400696
Theodore Ts'o61d08672010-10-27 21:30:15 -0400697 ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
698 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500699 err = ext4_handle_dirty_metadata(handle, NULL, bh);
700 if (unlikely(err)) {
701 ext4_std_error(sb, err);
702 goto exit_bh;
703 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700704 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700705 /* Mark unused entries in inode bitmap used */
Eric Sandeenc549a952008-01-28 23:58:27 -0500706 ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700707 input->inode_bitmap, input->inode_bitmap - start);
708 if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
709 err = PTR_ERR(bh);
710 goto exit_journal;
711 }
712
Theodore Ts'o61d08672010-10-27 21:30:15 -0400713 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
714 bh->b_data);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500715 err = ext4_handle_dirty_metadata(handle, NULL, bh);
716 if (unlikely(err))
717 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700718exit_bh:
719 brelse(bh);
720
721exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -0700722 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 err = err2;
724
725 return err;
726}
727
728/*
729 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700730 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731 * calling this for the first time. In a sparse filesystem it will be the
732 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
733 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
734 */
Mingming Cao617ba132006-10-11 01:20:53 -0700735static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700736 unsigned *five, unsigned *seven)
737{
738 unsigned *min = three;
739 int mult = 3;
740 unsigned ret;
741
Mingming Cao617ba132006-10-11 01:20:53 -0700742 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
743 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700744 ret = *min;
745 *min += 1;
746 return ret;
747 }
748
749 if (*five < *min) {
750 min = five;
751 mult = 5;
752 }
753 if (*seven < *min) {
754 min = seven;
755 mult = 7;
756 }
757
758 ret = *min;
759 *min *= mult;
760
761 return ret;
762}
763
764/*
765 * Check that all of the backup GDT blocks are held in the primary GDT block.
766 * It is assumed that they are stored in group order. Returns the number of
767 * groups in current filesystem that have BACKUPS, or -ve error code.
768 */
769static int verify_reserved_gdb(struct super_block *sb,
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500770 ext4_group_t end,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700771 struct buffer_head *primary)
772{
Mingming Cao617ba132006-10-11 01:20:53 -0700773 const ext4_fsblk_t blk = primary->b_blocknr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700774 unsigned three = 1;
775 unsigned five = 5;
776 unsigned seven = 7;
777 unsigned grp;
778 __le32 *p = (__le32 *)primary->b_data;
779 int gdbackups = 0;
780
Mingming Cao617ba132006-10-11 01:20:53 -0700781 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700782 if (le32_to_cpu(*p++) !=
783 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500784 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700785 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700787 grp *
788 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
789 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 return -EINVAL;
791 }
Mingming Cao617ba132006-10-11 01:20:53 -0700792 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700793 return -EFBIG;
794 }
795
796 return gdbackups;
797}
798
799/*
800 * Called when we need to bring a reserved group descriptor table block into
801 * use from the resize inode. The primary copy of the new GDT block currently
802 * is an indirect block (under the double indirect block in the resize inode).
803 * The new backup GDT blocks will be stored as leaf blocks in this indirect
804 * block, in group order. Even though we know all the block numbers we need,
805 * we check to ensure that the resize inode has actually reserved these blocks.
806 *
807 * Don't need to update the block bitmaps because the blocks are still in use.
808 *
809 * We get all of the error cases out of the way, so that we are sure to not
810 * fail once we start modifying the data on disk, because JBD has no rollback.
811 */
812static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400813 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700814{
815 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700816 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400817 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700818 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819 struct buffer_head **o_group_desc, **n_group_desc;
820 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400821 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700823 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700824 __le32 *data;
825 int err;
826
827 if (test_opt(sb, DEBUG))
828 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700829 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700830 gdb_num);
831
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400832 /*
833 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400834 * because the user tools have no way of handling this. Probably a
835 * bad time to do it anyways.
836 */
Mingming Cao617ba132006-10-11 01:20:53 -0700837 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
838 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500839 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700840 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700841 return -EPERM;
842 }
843
Yongqiang Yang2f919712011-07-27 21:16:33 -0400844 gdb_bh = sb_bread(sb, gdblock);
845 if (!gdb_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700846 return -EIO;
847
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500848 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400849 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700850 err = gdbackups;
851 goto exit_bh;
852 }
853
Mingming Cao617ba132006-10-11 01:20:53 -0700854 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700855 dind = sb_bread(sb, le32_to_cpu(*data));
856 if (!dind) {
857 err = -EIO;
858 goto exit_bh;
859 }
860
861 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700862 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500863 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Yongqiang Yang2f919712011-07-27 21:16:33 -0400864 group, gdblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865 err = -EINVAL;
866 goto exit_dind;
867 }
868
Theodore Ts'ob4097142011-01-10 12:46:59 -0500869 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
870 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871 goto exit_dind;
872
Yongqiang Yang2f919712011-07-27 21:16:33 -0400873 err = ext4_journal_get_write_access(handle, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500874 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875 goto exit_sbh;
876
Theodore Ts'ob4097142011-01-10 12:46:59 -0500877 err = ext4_journal_get_write_access(handle, dind);
878 if (unlikely(err))
879 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880
Mingming Cao617ba132006-10-11 01:20:53 -0700881 /* ext4_reserve_inode_write() gets a reference on the iloc */
Theodore Ts'ob4097142011-01-10 12:46:59 -0500882 err = ext4_reserve_inode_write(handle, inode, &iloc);
883 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 goto exit_dindj;
885
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400886 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
887 sizeof(struct buffer_head *),
888 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700889 if (!n_group_desc) {
890 err = -ENOMEM;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400891 ext4_warning(sb, "not enough memory for %lu groups",
892 gdb_num + 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700893 goto exit_inode;
894 }
895
896 /*
897 * Finally, we have all of the possible failures behind us...
898 *
899 * Remove new GDT block from inode double-indirect block and clear out
900 * the new GDT block for use (which also "frees" the backup GDT blocks
901 * from the reserved inode). We don't need to change the bitmaps for
902 * these blocks, because they are marked as in-use from being in the
903 * reserved inode, and will become GDT blocks (primary and backup).
904 */
Mingming Cao617ba132006-10-11 01:20:53 -0700905 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500906 err = ext4_handle_dirty_metadata(handle, NULL, dind);
907 if (unlikely(err)) {
908 ext4_std_error(sb, err);
909 goto exit_inode;
910 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700912 ext4_mark_iloc_dirty(handle, inode, &iloc);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400913 memset(gdb_bh->b_data, 0, sb->s_blocksize);
914 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500915 if (unlikely(err)) {
916 ext4_std_error(sb, err);
917 goto exit_inode;
918 }
919 brelse(dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920
Mingming Cao617ba132006-10-11 01:20:53 -0700921 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700922 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700923 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Yongqiang Yang2f919712011-07-27 21:16:33 -0400924 n_group_desc[gdb_num] = gdb_bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700925 EXT4_SB(sb)->s_group_desc = n_group_desc;
926 EXT4_SB(sb)->s_gdb_count++;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400927 ext4_kvfree(o_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700928
Marcin Slusarze8546d02008-04-17 10:38:59 -0400929 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500930 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
931 if (err)
932 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700933
Theodore Ts'ob4097142011-01-10 12:46:59 -0500934 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700935
936exit_inode:
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400937 ext4_kvfree(n_group_desc);
Amir Goldstein537a0312011-03-20 22:57:02 -0400938 /* ext4_handle_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700939 brelse(iloc.bh);
940exit_dindj:
Amir Goldstein537a0312011-03-20 22:57:02 -0400941 /* ext4_handle_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942exit_sbh:
Amir Goldstein537a0312011-03-20 22:57:02 -0400943 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700944exit_dind:
945 brelse(dind);
946exit_bh:
Yongqiang Yang2f919712011-07-27 21:16:33 -0400947 brelse(gdb_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948
Mingming Cao617ba132006-10-11 01:20:53 -0700949 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 return err;
951}
952
953/*
954 * Called when we are adding a new group which has a backup copy of each of
955 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
956 * We need to add these reserved backup GDT blocks to the resize inode, so
957 * that they are kept for future resizing and not allocated to files.
958 *
959 * Each reserved backup GDT block will go into a different indirect block.
960 * The indirect blocks are actually the primary reserved GDT blocks,
961 * so we know in advance what their block numbers are. We only get the
962 * double-indirect block to verify it is pointing to the primary reserved
963 * GDT blocks so we don't overwrite a data block by accident. The reserved
964 * backup GDT blocks are stored in their reserved primary GDT block.
965 */
966static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400967 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968{
969 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700970 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700971 struct buffer_head **primary;
972 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700973 struct ext4_iloc iloc;
974 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700975 __le32 *data, *end;
976 int gdbackups = 0;
977 int res, i;
978 int err;
979
Josef Bacik216553c2008-04-29 22:02:02 -0400980 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700981 if (!primary)
982 return -ENOMEM;
983
Mingming Cao617ba132006-10-11 01:20:53 -0700984 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 dind = sb_bread(sb, le32_to_cpu(*data));
986 if (!dind) {
987 err = -EIO;
988 goto exit_free;
989 }
990
Mingming Cao617ba132006-10-11 01:20:53 -0700991 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400992 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
993 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700994 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995
996 /* Get each reserved primary GDT block and verify it holds backups */
997 for (res = 0; res < reserved_gdb; res++, blk++) {
998 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500999 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 " not at offset %ld",
1001 blk,
1002 (long)(data - (__le32 *)dind->b_data));
1003 err = -EINVAL;
1004 goto exit_bh;
1005 }
1006 primary[res] = sb_bread(sb, blk);
1007 if (!primary[res]) {
1008 err = -EIO;
1009 goto exit_bh;
1010 }
Yongqiang Yangc72df9f2012-01-03 23:43:39 -05001011 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1012 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013 brelse(primary[res]);
1014 err = gdbackups;
1015 goto exit_bh;
1016 }
1017 if (++data >= end)
1018 data = (__le32 *)dind->b_data;
1019 }
1020
1021 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001022 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001023 /*
1024 int j;
1025 for (j = 0; j < i; j++)
Amir Goldstein537a0312011-03-20 22:57:02 -04001026 ext4_handle_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 */
1028 goto exit_bh;
1029 }
1030 }
1031
Mingming Cao617ba132006-10-11 01:20:53 -07001032 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 goto exit_bh;
1034
1035 /*
1036 * Finally we can add each of the reserved backup GDT blocks from
1037 * the new group to its reserved primary GDT block.
1038 */
Yongqiang Yang668f4dc2011-07-27 21:23:13 -04001039 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 for (i = 0; i < reserved_gdb; i++) {
1041 int err2;
1042 data = (__le32 *)primary[i]->b_data;
1043 /* printk("reserving backup %lu[%u] = %lu\n",
1044 primary[i]->b_blocknr, gdbackups,
1045 blk + primary[i]->b_blocknr); */
1046 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -05001047 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001048 if (!err)
1049 err = err2;
1050 }
1051 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -07001052 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053
1054exit_bh:
1055 while (--res >= 0)
1056 brelse(primary[res]);
1057 brelse(dind);
1058
1059exit_free:
1060 kfree(primary);
1061
1062 return err;
1063}
1064
1065/*
Mingming Cao617ba132006-10-11 01:20:53 -07001066 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001067 * of the main resize transaction, because e2fsck will re-write them if there
1068 * is a problem (basically only OOM will cause a problem). However, we
1069 * _should_ update the backups if possible, in case the primary gets trashed
1070 * for some reason and we need to run e2fsck from a backup superblock. The
1071 * important part is that the new block and inode counts are in the backup
1072 * superblocks, and the location of the new group metadata in the GDT backups.
1073 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001074 * We do not need take the s_resize_lock for this, because these
1075 * blocks are not otherwise touched by the filesystem code when it is
1076 * mounted. We don't need to worry about last changing from
1077 * sbi->s_groups_count, because the worst that can happen is that we
1078 * do not copy the full number of backups at this time. The resize
1079 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001080 */
1081static void update_backups(struct super_block *sb,
1082 int blk_off, char *data, int size)
1083{
Mingming Cao617ba132006-10-11 01:20:53 -07001084 struct ext4_sb_info *sbi = EXT4_SB(sb);
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001085 const ext4_group_t last = sbi->s_groups_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001086 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001087 unsigned three = 1;
1088 unsigned five = 5;
1089 unsigned seven = 7;
Avantika Mathurfd2d4292008-01-28 23:58:27 -05001090 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091 int rest = sb->s_blocksize - size;
1092 handle_t *handle;
1093 int err = 0, err2;
1094
Mingming Cao617ba132006-10-11 01:20:53 -07001095 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096 if (IS_ERR(handle)) {
1097 group = 1;
1098 err = PTR_ERR(handle);
1099 goto exit_err;
1100 }
1101
Mingming Cao617ba132006-10-11 01:20:53 -07001102 while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001103 struct buffer_head *bh;
1104
1105 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -05001106 if (ext4_handle_valid(handle) &&
1107 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -07001108 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1109 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001110 break;
1111
1112 bh = sb_getblk(sb, group * bpg + blk_off);
1113 if (!bh) {
1114 err = -EIO;
1115 break;
1116 }
Mingming Cao617ba132006-10-11 01:20:53 -07001117 ext4_debug("update metadata backup %#04lx\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001118 (unsigned long)bh->b_blocknr);
Mingming Cao617ba132006-10-11 01:20:53 -07001119 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001120 break;
1121 lock_buffer(bh);
1122 memcpy(bh->b_data, data, size);
1123 if (rest)
1124 memset(bh->b_data + size, 0, rest);
1125 set_buffer_uptodate(bh);
1126 unlock_buffer(bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05001127 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1128 if (unlikely(err))
1129 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001130 brelse(bh);
1131 }
Mingming Cao617ba132006-10-11 01:20:53 -07001132 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001133 err = err2;
1134
1135 /*
1136 * Ugh! Need to have e2fsck write the backup copies. It is too
1137 * late to revert the resize, we shouldn't fail just because of
1138 * the backup copies (they are only needed in case of corruption).
1139 *
1140 * However, if we got here we have a journal problem too, so we
1141 * can't really start a transaction to mark the superblock.
1142 * Chicken out and just set the flag on the hope it will be written
1143 * to disk, and if not - we will simply wait until next fsck.
1144 */
1145exit_err:
1146 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001147 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001148 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -07001149 sbi->s_mount_state &= ~EXT4_VALID_FS;
1150 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001151 mark_buffer_dirty(sbi->s_sbh);
1152 }
1153}
1154
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001155/*
1156 * ext4_add_new_descs() adds @count group descriptor of groups
1157 * starting at @group
1158 *
1159 * @handle: journal handle
1160 * @sb: super block
1161 * @group: the group no. of the first group desc to be added
1162 * @resize_inode: the resize inode
1163 * @count: number of group descriptors to be added
1164 */
1165static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1166 ext4_group_t group, struct inode *resize_inode,
1167 ext4_group_t count)
1168{
1169 struct ext4_sb_info *sbi = EXT4_SB(sb);
1170 struct ext4_super_block *es = sbi->s_es;
1171 struct buffer_head *gdb_bh;
1172 int i, gdb_off, gdb_num, err = 0;
1173
1174 for (i = 0; i < count; i++, group++) {
1175 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1176 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1177
1178 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1179 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1180
1181 /*
1182 * We will only either add reserved group blocks to a backup group
1183 * or remove reserved blocks for the first group in a new group block.
1184 * Doing both would be mean more complex code, and sane people don't
1185 * use non-sparse filesystems anymore. This is already checked above.
1186 */
1187 if (gdb_off) {
1188 gdb_bh = sbi->s_group_desc[gdb_num];
1189 err = ext4_journal_get_write_access(handle, gdb_bh);
1190
1191 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1192 err = reserve_backup_gdb(handle, resize_inode, group);
1193 } else
1194 err = add_new_gdb(handle, resize_inode, group);
1195 if (err)
1196 break;
1197 }
1198 return err;
1199}
1200
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001201/*
1202 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1203 */
1204static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1205 struct ext4_new_flex_group_data *flex_gd)
1206{
1207 struct ext4_new_group_data *group_data = flex_gd->groups;
1208 struct ext4_group_desc *gdp;
1209 struct ext4_sb_info *sbi = EXT4_SB(sb);
1210 struct buffer_head *gdb_bh;
1211 ext4_group_t group;
1212 __u16 *bg_flags = flex_gd->bg_flags;
1213 int i, gdb_off, gdb_num, err = 0;
1214
1215
1216 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1217 group = group_data->group;
1218
1219 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1220 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1221
1222 /*
1223 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1224 */
1225 gdb_bh = sbi->s_group_desc[gdb_num];
1226 /* Update group descriptor block for new group */
1227 gdp = (struct ext4_group_desc *)((char *)gdb_bh->b_data +
1228 gdb_off * EXT4_DESC_SIZE(sb));
1229
1230 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1231 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1232 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1233 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1234 ext4_free_group_clusters_set(sb, gdp,
1235 EXT4_B2C(sbi, group_data->free_blocks_count));
1236 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1237 gdp->bg_flags = cpu_to_le16(*bg_flags);
1238 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1239
1240 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1241 if (unlikely(err)) {
1242 ext4_std_error(sb, err);
1243 break;
1244 }
1245
1246 /*
1247 * We can allocate memory for mb_alloc based on the new group
1248 * descriptor
1249 */
1250 err = ext4_mb_add_groupinfo(sb, group, gdp);
1251 if (err)
1252 break;
1253 }
1254 return err;
1255}
1256
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001257/*
1258 * ext4_update_super() updates the super block so that the newly added
1259 * groups can be seen by the filesystem.
1260 *
1261 * @sb: super block
1262 * @flex_gd: new added groups
1263 */
1264static void ext4_update_super(struct super_block *sb,
1265 struct ext4_new_flex_group_data *flex_gd)
1266{
1267 ext4_fsblk_t blocks_count = 0;
1268 ext4_fsblk_t free_blocks = 0;
1269 ext4_fsblk_t reserved_blocks = 0;
1270 struct ext4_new_group_data *group_data = flex_gd->groups;
1271 struct ext4_sb_info *sbi = EXT4_SB(sb);
1272 struct ext4_super_block *es = sbi->s_es;
1273 int i;
1274
1275 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1276 /*
1277 * Make the new blocks and inodes valid next. We do this before
1278 * increasing the group count so that once the group is enabled,
1279 * all of its blocks and inodes are already valid.
1280 *
1281 * We always allocate group-by-group, then block-by-block or
1282 * inode-by-inode within a group, so enabling these
1283 * blocks/inodes before the group is live won't actually let us
1284 * allocate the new space yet.
1285 */
1286 for (i = 0; i < flex_gd->count; i++) {
1287 blocks_count += group_data[i].blocks_count;
1288 free_blocks += group_data[i].free_blocks_count;
1289 }
1290
1291 reserved_blocks = ext4_r_blocks_count(es) * 100;
1292 do_div(reserved_blocks, ext4_blocks_count(es));
1293 reserved_blocks *= blocks_count;
1294 do_div(reserved_blocks, 100);
1295
1296 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1297 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1298 flex_gd->count);
1299
1300 /*
1301 * We need to protect s_groups_count against other CPUs seeing
1302 * inconsistent state in the superblock.
1303 *
1304 * The precise rules we use are:
1305 *
1306 * * Writers must perform a smp_wmb() after updating all
1307 * dependent data and before modifying the groups count
1308 *
1309 * * Readers must perform an smp_rmb() after reading the groups
1310 * count and before reading any dependent data.
1311 *
1312 * NB. These rules can be relaxed when checking the group count
1313 * while freeing data, as we can only allocate from a block
1314 * group after serialising against the group count, and we can
1315 * only then free after serialising in turn against that
1316 * allocation.
1317 */
1318 smp_wmb();
1319
1320 /* Update the global fs size fields */
1321 sbi->s_groups_count += flex_gd->count;
1322
1323 /* Update the reserved block counts only once the new group is
1324 * active. */
1325 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1326 reserved_blocks);
1327
1328 /* Update the free space counts */
1329 percpu_counter_add(&sbi->s_freeclusters_counter,
1330 EXT4_B2C(sbi, free_blocks));
1331 percpu_counter_add(&sbi->s_freeinodes_counter,
1332 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1333
1334 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1335 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1336 sbi->s_log_groups_per_flex) {
1337 ext4_group_t flex_group;
1338 flex_group = ext4_flex_group(sbi, group_data[0].group);
1339 atomic_add(EXT4_B2C(sbi, free_blocks),
1340 &sbi->s_flex_groups[flex_group].free_clusters);
1341 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1342 &sbi->s_flex_groups[flex_group].free_inodes);
1343 }
1344
1345 if (test_opt(sb, DEBUG))
1346 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1347 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1348 blocks_count, free_blocks, reserved_blocks);
1349}
1350
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001351/* Add a flex group to an fs. Ensure we handle all possible error conditions
1352 * _before_ we start modifying the filesystem, because we cannot abort the
1353 * transaction and not have it write the data to disk.
1354 */
1355static int ext4_flex_group_add(struct super_block *sb,
1356 struct inode *resize_inode,
1357 struct ext4_new_flex_group_data *flex_gd)
1358{
1359 struct ext4_sb_info *sbi = EXT4_SB(sb);
1360 struct ext4_super_block *es = sbi->s_es;
1361 ext4_fsblk_t o_blocks_count;
1362 ext4_grpblk_t last;
1363 ext4_group_t group;
1364 handle_t *handle;
1365 unsigned reserved_gdb;
1366 int err = 0, err2 = 0, credit;
1367
1368 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1369
1370 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1371 o_blocks_count = ext4_blocks_count(es);
1372 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1373 BUG_ON(last);
1374
1375 err = setup_new_flex_group_blocks(sb, flex_gd);
1376 if (err)
1377 goto exit;
1378 /*
1379 * We will always be modifying at least the superblock and GDT
1380 * block. If we are adding a group past the last current GDT block,
1381 * we will also modify the inode and the dindirect block. If we
1382 * are adding a group with superblock/GDT backups we will also
1383 * modify each of the reserved GDT dindirect blocks.
1384 */
1385 credit = flex_gd->count * 4 + reserved_gdb;
1386 handle = ext4_journal_start_sb(sb, credit);
1387 if (IS_ERR(handle)) {
1388 err = PTR_ERR(handle);
1389 goto exit;
1390 }
1391
1392 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1393 if (err)
1394 goto exit_journal;
1395
1396 group = flex_gd->groups[0].group;
1397 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1398 err = ext4_add_new_descs(handle, sb, group,
1399 resize_inode, flex_gd->count);
1400 if (err)
1401 goto exit_journal;
1402
1403 err = ext4_setup_new_descs(handle, sb, flex_gd);
1404 if (err)
1405 goto exit_journal;
1406
1407 ext4_update_super(sb, flex_gd);
1408
1409 err = ext4_handle_dirty_super(handle, sb);
1410
1411exit_journal:
1412 err2 = ext4_journal_stop(handle);
1413 if (!err)
1414 err = err2;
1415
1416 if (!err) {
1417 int i;
1418 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1419 sizeof(struct ext4_super_block));
1420 for (i = 0; i < flex_gd->count; i++, group++) {
1421 struct buffer_head *gdb_bh;
1422 int gdb_num;
1423 gdb_num = group / EXT4_BLOCKS_PER_GROUP(sb);
1424 gdb_bh = sbi->s_group_desc[gdb_num];
1425 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1426 gdb_bh->b_size);
1427 }
1428 }
1429exit:
1430 return err;
1431}
1432
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001433/* Add group descriptor data to an existing or new group descriptor block.
1434 * Ensure we handle all possible error conditions _before_ we start modifying
1435 * the filesystem, because we cannot abort the transaction and not have it
1436 * write the data to disk.
1437 *
1438 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1439 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1440 *
1441 * We only need to hold the superblock lock while we are actually adding
1442 * in the new group's counts to the superblock. Prior to that we have
1443 * not really "added" the group at all. We re-check that we are still
1444 * adding in the last group in case things have changed since verifying.
1445 */
Mingming Cao617ba132006-10-11 01:20:53 -07001446int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001447{
Mingming Cao617ba132006-10-11 01:20:53 -07001448 struct ext4_sb_info *sbi = EXT4_SB(sb);
1449 struct ext4_super_block *es = sbi->s_es;
1450 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001451 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1452 struct buffer_head *primary = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001453 struct ext4_group_desc *gdp;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001454 struct inode *inode = NULL;
1455 handle_t *handle;
1456 int gdb_off, gdb_num;
1457 int err, err2;
1458
Mingming Cao617ba132006-10-11 01:20:53 -07001459 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1460 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001461
Mingming Cao617ba132006-10-11 01:20:53 -07001462 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1463 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001464 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001465 return -EPERM;
1466 }
1467
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001468 if (ext4_blocks_count(es) + input->blocks_count <
1469 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001470 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001471 return -EINVAL;
1472 }
1473
Mingming Cao617ba132006-10-11 01:20:53 -07001474 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001475 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001476 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001477 return -EINVAL;
1478 }
1479
1480 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001481 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -04001482 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1483 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001484 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001485 "No reserved GDT blocks, can't resize");
1486 return -EPERM;
1487 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001488 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1489 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001490 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -08001491 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001492 }
1493 }
1494
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001495
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001496 if ((err = verify_group_input(sb, input)))
1497 goto exit_put;
1498
1499 if ((err = setup_new_group_blocks(sb, input)))
1500 goto exit_put;
1501
1502 /*
1503 * We will always be modifying at least the superblock and a GDT
1504 * block. If we are adding a group past the last current GDT block,
1505 * we will also modify the inode and the dindirect block. If we
1506 * are adding a group with superblock/GDT backups we will also
1507 * modify each of the reserved GDT dindirect blocks.
1508 */
Mingming Cao617ba132006-10-11 01:20:53 -07001509 handle = ext4_journal_start_sb(sb,
1510 ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001511 3 + reserved_gdb : 4);
1512 if (IS_ERR(handle)) {
1513 err = PTR_ERR(handle);
1514 goto exit_put;
1515 }
1516
Mingming Cao617ba132006-10-11 01:20:53 -07001517 if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001518 goto exit_journal;
1519
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001520 /*
1521 * We will only either add reserved group blocks to a backup group
1522 * or remove reserved blocks for the first group in a new group block.
1523 * Doing both would be mean more complex code, and sane people don't
1524 * use non-sparse filesystems anymore. This is already checked above.
1525 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001526 if (gdb_off) {
1527 primary = sbi->s_group_desc[gdb_num];
Mingming Cao617ba132006-10-11 01:20:53 -07001528 if ((err = ext4_journal_get_write_access(handle, primary)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001529 goto exit_journal;
1530
Yongqiang Yang668f4dc2011-07-27 21:23:13 -04001531 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
1532 err = reserve_backup_gdb(handle, inode, input->group);
1533 if (err)
1534 goto exit_journal;
1535 }
Yongqiang Yang2f919712011-07-27 21:16:33 -04001536 } else {
1537 /*
1538 * Note that we can access new group descriptor block safely
1539 * only if add_new_gdb() succeeds.
1540 */
1541 err = add_new_gdb(handle, inode, input->group);
1542 if (err)
1543 goto exit_journal;
1544 primary = sbi->s_group_desc[gdb_num];
1545 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001546
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001547 /*
1548 * OK, now we've set up the new group. Time to make it active.
1549 *
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001550 * so we have to be safe wrt. concurrent accesses the group
1551 * data. So we need to be careful to set all of the relevant
1552 * group descriptor data etc. *before* we enable the group.
1553 *
1554 * The key field here is sbi->s_groups_count: as long as
1555 * that retains its old value, nobody is going to access the new
1556 * group.
1557 *
1558 * So first we update all the descriptor metadata for the new
1559 * group; then we update the total disk blocks count; then we
1560 * update the groups count to enable the group; then finally we
1561 * update the free space counts so that the system can start
1562 * using the new disk blocks.
1563 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001564
1565 /* Update group descriptor block for new group */
Frederic Bohe28569222008-06-20 11:48:48 -04001566 gdp = (struct ext4_group_desc *)((char *)primary->b_data +
1567 gdb_off * EXT4_DESC_SIZE(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001568
Theodore Ts'ofdff73f2009-01-26 19:06:41 -05001569 memset(gdp, 0, EXT4_DESC_SIZE(sb));
Alexandre Ratchov8fadc142006-10-11 01:21:15 -07001570 ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
1571 ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
1572 ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
Theodore Ts'o021b65b2011-09-09 19:08:51 -04001573 ext4_free_group_clusters_set(sb, gdp, input->free_blocks_count);
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05001574 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'ofdff73f2009-01-26 19:06:41 -05001575 gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
Andreas Dilger717d50e2007-10-16 18:38:25 -04001576 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577
1578 /*
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001579 * We can allocate memory for mb_alloc based on the new group
1580 * descriptor
1581 */
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001582 err = ext4_mb_add_groupinfo(sb, input->group, gdp);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -04001583 if (err)
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001584 goto exit_journal;
1585
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001586 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587 * Make the new blocks and inodes valid next. We do this before
1588 * increasing the group count so that once the group is enabled,
1589 * all of its blocks and inodes are already valid.
1590 *
1591 * We always allocate group-by-group, then block-by-block or
1592 * inode-by-inode within a group, so enabling these
1593 * blocks/inodes before the group is live won't actually let us
1594 * allocate the new space yet.
1595 */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001596 ext4_blocks_count_set(es, ext4_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001597 input->blocks_count);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001598 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001599
1600 /*
1601 * We need to protect s_groups_count against other CPUs seeing
1602 * inconsistent state in the superblock.
1603 *
1604 * The precise rules we use are:
1605 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001606 * * Writers must perform a smp_wmb() after updating all dependent
1607 * data and before modifying the groups count
1608 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001609 * * Readers must perform an smp_rmb() after reading the groups count
1610 * and before reading any dependent data.
1611 *
1612 * NB. These rules can be relaxed when checking the group count
1613 * while freeing data, as we can only allocate from a block
1614 * group after serialising against the group count, and we can
1615 * only then free after serialising in turn against that
1616 * allocation.
1617 */
1618 smp_wmb();
1619
1620 /* Update the global fs size fields */
1621 sbi->s_groups_count++;
1622
Theodore Ts'ob4097142011-01-10 12:46:59 -05001623 err = ext4_handle_dirty_metadata(handle, NULL, primary);
1624 if (unlikely(err)) {
1625 ext4_std_error(sb, err);
1626 goto exit_journal;
1627 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628
1629 /* Update the reserved block counts only once the new group is
1630 * active. */
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001631 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001632 input->reserved_blocks);
1633
1634 /* Update the free space counts */
Theodore Ts'o57042652011-09-09 18:56:51 -04001635 percpu_counter_add(&sbi->s_freeclusters_counter,
1636 EXT4_B2C(sbi, input->free_blocks_count));
Peter Zijlstraaa0dff22007-10-16 23:25:42 -07001637 percpu_counter_add(&sbi->s_freeinodes_counter,
Mingming Cao617ba132006-10-11 01:20:53 -07001638 EXT4_INODES_PER_GROUP(sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639
Eric Sandeen42007ef2010-05-16 01:00:00 -04001640 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1641 sbi->s_log_groups_per_flex) {
Frederic Bohec62a11f2008-09-08 10:20:24 -04001642 ext4_group_t flex_group;
1643 flex_group = ext4_flex_group(sbi, input->group);
Theodore Ts'o24aaa8e2011-09-09 18:58:51 -04001644 atomic_add(EXT4_B2C(sbi, input->free_blocks_count),
1645 &sbi->s_flex_groups[flex_group].free_clusters);
Theodore Ts'o9f24e422009-03-04 19:09:10 -05001646 atomic_add(EXT4_INODES_PER_GROUP(sb),
1647 &sbi->s_flex_groups[flex_group].free_inodes);
Frederic Bohec62a11f2008-09-08 10:20:24 -04001648 }
1649
Theodore Ts'oa0375152010-06-11 23:14:04 -04001650 ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001651
1652exit_journal:
Mingming Cao617ba132006-10-11 01:20:53 -07001653 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001654 err = err2;
Yongqiang Yang2f919712011-07-27 21:16:33 -04001655 if (!err && primary) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001656 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Mingming Cao617ba132006-10-11 01:20:53 -07001657 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001658 update_backups(sb, primary->b_blocknr, primary->b_data,
1659 primary->b_size);
1660 }
1661exit_put:
1662 iput(inode);
1663 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001664} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001665
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001666/*
Yongqiang Yang18e31432012-01-03 23:18:50 -05001667 * extend a group without checking assuming that checking has been done.
1668 */
1669static int ext4_group_extend_no_check(struct super_block *sb,
1670 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1671{
1672 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1673 handle_t *handle;
1674 int err = 0, err2;
1675
1676 /* We will update the superblock, one block bitmap, and
1677 * one group descriptor via ext4_group_add_blocks().
1678 */
1679 handle = ext4_journal_start_sb(sb, 3);
1680 if (IS_ERR(handle)) {
1681 err = PTR_ERR(handle);
1682 ext4_warning(sb, "error %d on journal start", err);
1683 return err;
1684 }
1685
1686 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1687 if (err) {
1688 ext4_warning(sb, "error %d on journal write access", err);
1689 goto errout;
1690 }
1691
1692 ext4_blocks_count_set(es, o_blocks_count + add);
1693 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1694 o_blocks_count + add);
1695 /* We add the blocks to the bitmap and set the group need init bit */
1696 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1697 if (err)
1698 goto errout;
1699 ext4_handle_dirty_super(handle, sb);
1700 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1701 o_blocks_count + add);
1702errout:
1703 err2 = ext4_journal_stop(handle);
1704 if (err2 && !err)
1705 err = err2;
1706
1707 if (!err) {
1708 if (test_opt(sb, DEBUG))
1709 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1710 "blocks\n", ext4_blocks_count(es));
1711 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1712 sizeof(struct ext4_super_block));
1713 }
1714 return err;
1715}
1716
1717/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001718 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001719 * point is only used to extend the current filesystem to the end of the last
1720 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1721 * for emergencies (because it has no dependencies on reserved blocks).
1722 *
Mingming Cao617ba132006-10-11 01:20:53 -07001723 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001724 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1725 * GDT blocks are reserved to grow to the desired size.
1726 */
Mingming Cao617ba132006-10-11 01:20:53 -07001727int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1728 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001729{
Mingming Cao617ba132006-10-11 01:20:53 -07001730 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001731 ext4_grpblk_t last;
1732 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001733 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001734 handle_t *handle;
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001735 int err, err2;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001736 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001737
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001738 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001739
1740 if (test_opt(sb, DEBUG))
Yongqiang Yang2b79b092011-07-26 21:53:35 -04001741 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742 o_blocks_count, n_blocks_count);
1743
1744 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1745 return 0;
1746
1747 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001748 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
Mingming Cao2ae02102006-10-11 01:21:11 -07001749 " too large to resize to %llu blocks safely\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001750 sb->s_id, n_blocks_count);
1751 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001752 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001753 return -EINVAL;
1754 }
1755
1756 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001757 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001758 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001759 }
1760
1761 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001762 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001763
1764 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001765 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001766 return -EPERM;
1767 }
1768
Mingming Cao617ba132006-10-11 01:20:53 -07001769 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001770
1771 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001772 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773 return -EINVAL;
1774 }
1775
1776 if (o_blocks_count + add > n_blocks_count)
1777 add = n_blocks_count - o_blocks_count;
1778
1779 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001780 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001781 o_blocks_count + add, add);
1782
1783 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001784 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001785 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001786 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001787 return -ENOSPC;
1788 }
1789 brelse(bh);
1790
1791 /* We will update the superblock, one block bitmap, and
Mingming Cao617ba132006-10-11 01:20:53 -07001792 * one group descriptor via ext4_free_blocks().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001793 */
Mingming Cao617ba132006-10-11 01:20:53 -07001794 handle = ext4_journal_start_sb(sb, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001795 if (IS_ERR(handle)) {
1796 err = PTR_ERR(handle);
Eric Sandeen12062dd2010-02-15 14:19:27 -05001797 ext4_warning(sb, "error %d on journal start", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001798 goto exit_put;
1799 }
1800
Mingming Cao617ba132006-10-11 01:20:53 -07001801 if ((err = ext4_journal_get_write_access(handle,
1802 EXT4_SB(sb)->s_sbh))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001803 ext4_warning(sb, "error %d on journal write access", err);
Mingming Cao617ba132006-10-11 01:20:53 -07001804 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001805 goto exit_put;
1806 }
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001807 ext4_blocks_count_set(es, o_blocks_count + add);
Eric Sandeenc549a952008-01-28 23:58:27 -05001808 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001809 o_blocks_count + add);
Aneesh Kumar K.Ve21675d2009-01-05 21:36:02 -05001810 /* We add the blocks to the bitmap and set the group need init bit */
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001811 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
Theodore Ts'oa0375152010-06-11 23:14:04 -04001812 ext4_handle_dirty_super(handle, sb);
Mingming Cao2ae02102006-10-11 01:21:11 -07001813 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814 o_blocks_count + add);
Yongqiang Yangcc7365d2011-07-26 21:46:07 -04001815 err2 = ext4_journal_stop(handle);
1816 if (!err && err2)
1817 err = err2;
1818
1819 if (err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001820 goto exit_put;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001821
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001822 if (test_opt(sb, DEBUG))
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001823 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1824 ext4_blocks_count(es));
Mingming Cao617ba132006-10-11 01:20:53 -07001825 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1826 sizeof(struct ext4_super_block));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827exit_put:
1828 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001829} /* ext4_group_extend */