blob: cf0c472047e3a89a84e262c7eeddd7ca72fd1b8c [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 /*
Theodore Ts'o011fa992014-12-26 23:58:21 -050027 * If we are not using the primary superblock/GDT copy don't resize,
28 * because the user tools have no way of handling this. Probably a
29 * bad time to do it anyways.
30 */
31 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
32 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
33 ext4_warning(sb, "won't resize using backup superblock at %llu",
34 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
35 return -EPERM;
36 }
37
38 /*
Yongqiang Yangce723c32011-07-26 21:39:09 -040039 * We are not allowed to do online-resizing on a filesystem mounted
40 * with error, because it can destroy the filesystem easily.
41 */
42 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
43 ext4_warning(sb, "There are errors in the filesystem, "
44 "so online resizing is not allowed\n");
45 return -EPERM;
46 }
47
Yongqiang Yang8f82f842011-07-26 21:35:44 -040048 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
49 ret = -EBUSY;
50
51 return ret;
52}
53
54void ext4_resize_end(struct super_block *sb)
55{
56 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010057 smp_mb__after_atomic();
Yongqiang Yang8f82f842011-07-26 21:35:44 -040058}
59
Yongqiang Yang01f795f2012-09-05 01:33:50 -040060static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
61 ext4_group_t group) {
62 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
63 EXT4_DESC_PER_BLOCK_BITS(sb);
64}
65
66static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
67 ext4_group_t group) {
68 group = ext4_meta_bg_first_group(sb, group);
69 return ext4_group_first_block_no(sb, group);
70}
71
72static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
73 ext4_group_t group) {
74 ext4_grpblk_t overhead;
75 overhead = ext4_bg_num_gdb(sb, group);
76 if (ext4_bg_has_super(sb, group))
77 overhead += 1 +
78 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
79 return overhead;
80}
81
Dave Kleikampac27a0e2006-10-11 01:20:50 -070082#define outside(b, first, last) ((b) < (first) || (b) >= (last))
83#define inside(b, first, last) ((b) >= (first) && (b) < (last))
84
85static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070086 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070087{
Mingming Cao617ba132006-10-11 01:20:53 -070088 struct ext4_sb_info *sbi = EXT4_SB(sb);
89 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070090 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070091 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050092 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070093 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
Theodore Ts'ob302ef22013-06-06 11:14:31 -040094 unsigned overhead;
95 ext4_fsblk_t metaend;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070097 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070098 int err = -EINVAL;
99
Theodore Ts'ob302ef22013-06-06 11:14:31 -0400100 if (group != sbi->s_groups_count) {
101 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
102 input->group, sbi->s_groups_count);
103 return -EINVAL;
104 }
105
106 overhead = ext4_group_overhead_blocks(sb, group);
107 metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700108 input->free_blocks_count = free_blocks_count =
109 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
110
111 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -0700112 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700113 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -0700114 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700115 "no-super", input->group, input->blocks_count,
116 free_blocks_count, input->reserved_blocks);
117
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700118 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Theodore Ts'ob302ef22013-06-06 11:14:31 -0400119 if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500120 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700121 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500122 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700123 input->reserved_blocks);
124 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500125 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700126 input->blocks_count);
127 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500128 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700129 end - 1);
130 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500131 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700132 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700133 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500134 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700135 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700136 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400137 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500138 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700139 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700140 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500141 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700142 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700143 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500144 ext4_warning(sb, "Block bitmap (%llu) in inode table "
145 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700146 (unsigned long long)input->block_bitmap,
147 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700148 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500149 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
150 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700151 (unsigned long long)input->inode_bitmap,
152 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700153 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500154 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700155 (unsigned long long)input->block_bitmap,
156 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700157 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500158 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700159 (unsigned long long)input->inode_bitmap,
160 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700161 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400162 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500163 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
164 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700165 (unsigned long long)input->inode_table,
166 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700167 else
168 err = 0;
169 brelse(bh);
170
171 return err;
172}
173
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500174/*
175 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
176 * group each time.
177 */
178struct ext4_new_flex_group_data {
179 struct ext4_new_group_data *groups; /* new_group_data for groups
180 in the flex group */
181 __u16 *bg_flags; /* block group flags of groups
182 in @groups */
183 ext4_group_t count; /* number of groups in @groups
184 */
185};
186
187/*
188 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
189 * @flexbg_size.
190 *
191 * Returns NULL on failure otherwise address of the allocated structure.
192 */
193static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
194{
195 struct ext4_new_flex_group_data *flex_gd;
196
197 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
198 if (flex_gd == NULL)
199 goto out3;
200
Haogang Chen967ac8a2012-05-28 14:21:55 -0400201 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
202 goto out2;
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500203 flex_gd->count = flexbg_size;
204
205 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
206 flexbg_size, GFP_NOFS);
207 if (flex_gd->groups == NULL)
208 goto out2;
209
210 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
211 if (flex_gd->bg_flags == NULL)
212 goto out1;
213
214 return flex_gd;
215
216out1:
217 kfree(flex_gd->groups);
218out2:
219 kfree(flex_gd);
220out3:
221 return NULL;
222}
223
224static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
225{
226 kfree(flex_gd->bg_flags);
227 kfree(flex_gd->groups);
228 kfree(flex_gd);
229}
230
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500231/*
232 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
233 * and inode tables for a flex group.
234 *
235 * This function is used by 64bit-resize. Note that this function allocates
236 * group tables from the 1st group of groups contained by @flexgd, which may
237 * be a partial of a flex group.
238 *
239 * @sb: super block of fs to which the groups belongs
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400240 *
241 * Returns 0 on a successful allocation of the metadata blocks in the
242 * block group.
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500243 */
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400244static int ext4_alloc_group_tables(struct super_block *sb,
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500245 struct ext4_new_flex_group_data *flex_gd,
246 int flexbg_size)
247{
248 struct ext4_new_group_data *group_data = flex_gd->groups;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500249 ext4_fsblk_t start_blk;
250 ext4_fsblk_t last_blk;
251 ext4_group_t src_group;
252 ext4_group_t bb_index = 0;
253 ext4_group_t ib_index = 0;
254 ext4_group_t it_index = 0;
255 ext4_group_t group;
256 ext4_group_t last_group;
257 unsigned overhead;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500258 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500259
260 BUG_ON(flex_gd->count == 0 || group_data == NULL);
261
262 src_group = group_data[0].group;
263 last_group = src_group + flex_gd->count - 1;
264
265 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
266 (last_group & ~(flexbg_size - 1))));
267next_group:
268 group = group_data[0].group;
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400269 if (src_group >= group_data[0].group + flex_gd->count)
270 return -ENOSPC;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500271 start_blk = ext4_group_first_block_no(sb, src_group);
272 last_blk = start_blk + group_data[src_group - group].blocks_count;
273
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400274 overhead = ext4_group_overhead_blocks(sb, src_group);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500275
276 start_blk += overhead;
277
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500278 /* We collect contiguous blocks as much as possible. */
279 src_group++;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400280 for (; src_group <= last_group; src_group++) {
281 overhead = ext4_group_overhead_blocks(sb, src_group);
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500282 if (overhead == 0)
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500283 last_blk += group_data[src_group - group].blocks_count;
284 else
285 break;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400286 }
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500287
288 /* Allocate block bitmaps */
289 for (; bb_index < flex_gd->count; bb_index++) {
290 if (start_blk >= last_blk)
291 goto next_group;
292 group_data[bb_index].block_bitmap = start_blk++;
Lukas Czernerbd862982013-04-03 23:32:34 -0400293 group = ext4_get_group_number(sb, start_blk - 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500294 group -= group_data[0].group;
295 group_data[group].free_blocks_count--;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500296 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500297 }
298
299 /* Allocate inode bitmaps */
300 for (; ib_index < flex_gd->count; ib_index++) {
301 if (start_blk >= last_blk)
302 goto next_group;
303 group_data[ib_index].inode_bitmap = start_blk++;
Lukas Czernerbd862982013-04-03 23:32:34 -0400304 group = ext4_get_group_number(sb, start_blk - 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500305 group -= group_data[0].group;
306 group_data[group].free_blocks_count--;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500307 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500308 }
309
310 /* Allocate inode tables */
311 for (; it_index < flex_gd->count; it_index++) {
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500312 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
313 ext4_fsblk_t next_group_start;
314
315 if (start_blk + itb > last_blk)
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500316 goto next_group;
317 group_data[it_index].inode_table = start_blk;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500318 group = ext4_get_group_number(sb, start_blk);
319 next_group_start = ext4_group_first_block_no(sb, group + 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500320 group -= group_data[0].group;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500321
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500322 if (start_blk + itb > next_group_start) {
323 flex_gd->bg_flags[group + 1] &= uninit_mask;
324 overhead = start_blk + itb - next_group_start;
325 group_data[group + 1].free_blocks_count -= overhead;
326 itb -= overhead;
327 }
328
329 group_data[group].free_blocks_count -= itb;
330 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500331 start_blk += EXT4_SB(sb)->s_itb_per_group;
332 }
333
334 if (test_opt(sb, DEBUG)) {
335 int i;
336 group = group_data[0].group;
337
338 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
339 "%d groups, flexbg size is %d:\n", flex_gd->count,
340 flexbg_size);
341
342 for (i = 0; i < flex_gd->count; i++) {
343 printk(KERN_DEBUG "adding %s group %u: %u "
344 "blocks (%d free)\n",
345 ext4_bg_has_super(sb, group + i) ? "normal" :
346 "no-super", group + i,
347 group_data[i].blocks_count,
348 group_data[i].free_blocks_count);
349 }
350 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400351 return 0;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500352}
353
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700354static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700355 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700356{
357 struct buffer_head *bh;
358 int err;
359
360 bh = sb_getblk(sb, blk);
Wang Shilongaebf0242013-01-12 16:28:47 -0500361 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500362 return ERR_PTR(-ENOMEM);
liang xie5d601252014-05-12 22:06:43 -0400363 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700364 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700365 brelse(bh);
366 bh = ERR_PTR(err);
367 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700368 memset(bh->b_data, 0, sb->s_blocksize);
369 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700370 }
371
372 return bh;
373}
374
375/*
Eric Sandeen14904102007-10-16 18:38:25 -0400376 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
377 * If that fails, restart the transaction & regain write access for the
378 * buffer head which is used for block_bitmap modifications.
379 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400380static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400381{
382 int err;
383
Frank Mayhar03901312009-01-07 00:06:22 -0500384 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400385 return 0;
386
387 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
388 if (err < 0)
389 return err;
390 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400391 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
392 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400393 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400394 }
Eric Sandeen14904102007-10-16 18:38:25 -0400395
396 return 0;
397}
398
399/*
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500400 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
401 *
402 * Helper function for ext4_setup_new_group_blocks() which set .
403 *
404 * @sb: super block
405 * @handle: journal handle
406 * @flex_gd: flex group data
407 */
408static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
409 struct ext4_new_flex_group_data *flex_gd,
410 ext4_fsblk_t block, ext4_group_t count)
411{
412 ext4_group_t count2;
413
414 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
415 for (count2 = count; count > 0; count -= count2, block += count2) {
416 ext4_fsblk_t start;
417 struct buffer_head *bh;
418 ext4_group_t group;
419 int err;
420
Lukas Czernerbd862982013-04-03 23:32:34 -0400421 group = ext4_get_group_number(sb, block);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500422 start = ext4_group_first_block_no(sb, group);
423 group -= flex_gd->groups[0].group;
424
Theodore Ts'o3d2660d2014-02-15 22:42:25 -0500425 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500426 if (count2 > count)
427 count2 = count;
428
429 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
430 BUG_ON(flex_gd->count > 1);
431 continue;
432 }
433
434 err = extend_or_restart_transaction(handle, 1);
435 if (err)
436 return err;
437
438 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
Wang Shilongaebf0242013-01-12 16:28:47 -0500439 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500440 return -ENOMEM;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500441
liang xie5d601252014-05-12 22:06:43 -0400442 BUFFER_TRACE(bh, "get_write_access");
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500443 err = ext4_journal_get_write_access(handle, bh);
444 if (err)
445 return err;
446 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
447 block - start, count2);
448 ext4_set_bits(bh->b_data, block - start, count2);
449
450 err = ext4_handle_dirty_metadata(handle, NULL, bh);
451 if (unlikely(err))
452 return err;
453 brelse(bh);
454 }
455
456 return 0;
457}
458
459/*
460 * Set up the block and inode bitmaps, and the inode table for the new groups.
461 * This doesn't need to be part of the main transaction, since we are only
462 * changing blocks outside the actual filesystem. We still do journaling to
463 * ensure the recovery is correct in case of a failure just after resize.
464 * If any part of this fails, we simply abort the resize.
465 *
466 * setup_new_flex_group_blocks handles a flex group as follow:
467 * 1. copy super block and GDT, and initialize group tables if necessary.
468 * In this step, we only set bits in blocks bitmaps for blocks taken by
469 * super block and GDT.
470 * 2. allocate group tables in block bitmaps, that is, set bits in block
471 * bitmap for blocks taken by group tables.
472 */
473static int setup_new_flex_group_blocks(struct super_block *sb,
474 struct ext4_new_flex_group_data *flex_gd)
475{
476 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
477 ext4_fsblk_t start;
478 ext4_fsblk_t block;
479 struct ext4_sb_info *sbi = EXT4_SB(sb);
480 struct ext4_super_block *es = sbi->s_es;
481 struct ext4_new_group_data *group_data = flex_gd->groups;
482 __u16 *bg_flags = flex_gd->bg_flags;
483 handle_t *handle;
484 ext4_group_t group, count;
485 struct buffer_head *bh = NULL;
486 int reserved_gdb, i, j, err = 0, err2;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400487 int meta_bg;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500488
489 BUG_ON(!flex_gd->count || !group_data ||
490 group_data[0].group != sbi->s_groups_count);
491
492 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400493 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500494
495 /* This transaction may be extended/restarted along the way */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500496 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500497 if (IS_ERR(handle))
498 return PTR_ERR(handle);
499
500 group = group_data[0].group;
501 for (i = 0; i < flex_gd->count; i++, group++) {
502 unsigned long gdblocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400503 ext4_grpblk_t overhead;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500504
505 gdblocks = ext4_bg_num_gdb(sb, group);
506 start = ext4_group_first_block_no(sb, group);
507
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400508 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400509 goto handle_itb;
510
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400511 if (meta_bg == 1) {
512 ext4_group_t first_group;
513 first_group = ext4_meta_bg_first_group(sb, group);
514 if (first_group != group + 1 &&
515 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
516 goto handle_itb;
517 }
518
519 block = start + ext4_bg_has_super(sb, group);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500520 /* Copy all of the GDT blocks into the backup in this group */
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400521 for (j = 0; j < gdblocks; j++, block++) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500522 struct buffer_head *gdb;
523
524 ext4_debug("update backup group %#04llx\n", block);
525 err = extend_or_restart_transaction(handle, 1);
526 if (err)
527 goto out;
528
529 gdb = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -0500530 if (unlikely(!gdb)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500531 err = -ENOMEM;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500532 goto out;
533 }
534
liang xie5d601252014-05-12 22:06:43 -0400535 BUFFER_TRACE(gdb, "get_write_access");
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500536 err = ext4_journal_get_write_access(handle, gdb);
537 if (err) {
538 brelse(gdb);
539 goto out;
540 }
541 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
542 gdb->b_size);
543 set_buffer_uptodate(gdb);
544
545 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
546 if (unlikely(err)) {
547 brelse(gdb);
548 goto out;
549 }
550 brelse(gdb);
551 }
552
553 /* Zero out all of the reserved backup group descriptor
554 * table blocks
555 */
556 if (ext4_bg_has_super(sb, group)) {
557 err = sb_issue_zeroout(sb, gdblocks + start + 1,
558 reserved_gdb, GFP_NOFS);
559 if (err)
560 goto out;
561 }
562
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400563handle_itb:
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500564 /* Initialize group tables of the grop @group */
565 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
566 goto handle_bb;
567
568 /* Zero out all of the inode table blocks */
569 block = group_data[i].inode_table;
570 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
571 block, sbi->s_itb_per_group);
572 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
573 GFP_NOFS);
574 if (err)
575 goto out;
576
577handle_bb:
578 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
579 goto handle_ib;
580
581 /* Initialize block bitmap of the @group */
582 block = group_data[i].block_bitmap;
583 err = extend_or_restart_transaction(handle, 1);
584 if (err)
585 goto out;
586
587 bh = bclean(handle, sb, block);
588 if (IS_ERR(bh)) {
589 err = PTR_ERR(bh);
Theodore Ts'oa9cfcd62014-09-03 09:33:00 -0400590 bh = NULL;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500591 goto out;
592 }
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400593 overhead = ext4_group_overhead_blocks(sb, group);
594 if (overhead != 0) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500595 ext4_debug("mark backup superblock %#04llx (+0)\n",
596 start);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400597 ext4_set_bits(bh->b_data, 0, overhead);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500598 }
599 ext4_mark_bitmap_end(group_data[i].blocks_count,
600 sb->s_blocksize * 8, bh->b_data);
601 err = ext4_handle_dirty_metadata(handle, NULL, bh);
602 if (err)
603 goto out;
604 brelse(bh);
605
606handle_ib:
607 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
608 continue;
609
610 /* Initialize inode bitmap of the @group */
611 block = group_data[i].inode_bitmap;
612 err = extend_or_restart_transaction(handle, 1);
613 if (err)
614 goto out;
615 /* Mark unused entries in inode bitmap used */
616 bh = bclean(handle, sb, block);
617 if (IS_ERR(bh)) {
618 err = PTR_ERR(bh);
Theodore Ts'oa9cfcd62014-09-03 09:33:00 -0400619 bh = NULL;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500620 goto out;
621 }
622
623 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
624 sb->s_blocksize * 8, bh->b_data);
625 err = ext4_handle_dirty_metadata(handle, NULL, bh);
626 if (err)
627 goto out;
628 brelse(bh);
629 }
630 bh = NULL;
631
632 /* Mark group tables in block bitmap */
633 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
634 count = group_table_count[j];
635 start = (&group_data[0].block_bitmap)[j];
636 block = start;
637 for (i = 1; i < flex_gd->count; i++) {
638 block += group_table_count[j];
639 if (block == (&group_data[i].block_bitmap)[j]) {
640 count += group_table_count[j];
641 continue;
642 }
643 err = set_flexbg_block_bitmap(sb, handle,
644 flex_gd, start, count);
645 if (err)
646 goto out;
647 count = group_table_count[j];
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500648 start = (&group_data[i].block_bitmap)[j];
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500649 block = start;
650 }
651
652 if (count) {
653 err = set_flexbg_block_bitmap(sb, handle,
654 flex_gd, start, count);
655 if (err)
656 goto out;
657 }
658 }
659
660out:
661 brelse(bh);
662 err2 = ext4_journal_stop(handle);
663 if (err2 && !err)
664 err = err2;
665
666 return err;
667}
668
669/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700670 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700671 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700672 * calling this for the first time. In a sparse filesystem it will be the
673 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
674 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
675 */
Mingming Cao617ba132006-10-11 01:20:53 -0700676static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700677 unsigned *five, unsigned *seven)
678{
679 unsigned *min = three;
680 int mult = 3;
681 unsigned ret;
682
Mingming Cao617ba132006-10-11 01:20:53 -0700683 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
684 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700685 ret = *min;
686 *min += 1;
687 return ret;
688 }
689
690 if (*five < *min) {
691 min = five;
692 mult = 5;
693 }
694 if (*seven < *min) {
695 min = seven;
696 mult = 7;
697 }
698
699 ret = *min;
700 *min *= mult;
701
702 return ret;
703}
704
705/*
706 * Check that all of the backup GDT blocks are held in the primary GDT block.
707 * It is assumed that they are stored in group order. Returns the number of
708 * groups in current filesystem that have BACKUPS, or -ve error code.
709 */
710static int verify_reserved_gdb(struct super_block *sb,
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500711 ext4_group_t end,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700712 struct buffer_head *primary)
713{
Mingming Cao617ba132006-10-11 01:20:53 -0700714 const ext4_fsblk_t blk = primary->b_blocknr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715 unsigned three = 1;
716 unsigned five = 5;
717 unsigned seven = 7;
718 unsigned grp;
719 __le32 *p = (__le32 *)primary->b_data;
720 int gdbackups = 0;
721
Mingming Cao617ba132006-10-11 01:20:53 -0700722 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700723 if (le32_to_cpu(*p++) !=
724 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500725 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700726 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700727 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700728 grp *
729 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
730 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731 return -EINVAL;
732 }
Mingming Cao617ba132006-10-11 01:20:53 -0700733 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700734 return -EFBIG;
735 }
736
737 return gdbackups;
738}
739
740/*
741 * Called when we need to bring a reserved group descriptor table block into
742 * use from the resize inode. The primary copy of the new GDT block currently
743 * is an indirect block (under the double indirect block in the resize inode).
744 * The new backup GDT blocks will be stored as leaf blocks in this indirect
745 * block, in group order. Even though we know all the block numbers we need,
746 * we check to ensure that the resize inode has actually reserved these blocks.
747 *
748 * Don't need to update the block bitmaps because the blocks are still in use.
749 *
750 * We get all of the error cases out of the way, so that we are sure to not
751 * fail once we start modifying the data on disk, because JBD has no rollback.
752 */
753static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400754 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700755{
756 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700757 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400758 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700759 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760 struct buffer_head **o_group_desc, **n_group_desc;
761 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400762 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700763 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700764 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765 __le32 *data;
766 int err;
767
768 if (test_opt(sb, DEBUG))
769 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700770 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700771 gdb_num);
772
Yongqiang Yang2f919712011-07-27 21:16:33 -0400773 gdb_bh = sb_bread(sb, gdblock);
774 if (!gdb_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700775 return -EIO;
776
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500777 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400778 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700779 err = gdbackups;
780 goto exit_bh;
781 }
782
Mingming Cao617ba132006-10-11 01:20:53 -0700783 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700784 dind = sb_bread(sb, le32_to_cpu(*data));
785 if (!dind) {
786 err = -EIO;
787 goto exit_bh;
788 }
789
790 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700791 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500792 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Yongqiang Yang2f919712011-07-27 21:16:33 -0400793 group, gdblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700794 err = -EINVAL;
795 goto exit_dind;
796 }
797
liang xie5d601252014-05-12 22:06:43 -0400798 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
Theodore Ts'ob4097142011-01-10 12:46:59 -0500799 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
800 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700801 goto exit_dind;
802
liang xie5d601252014-05-12 22:06:43 -0400803 BUFFER_TRACE(gdb_bh, "get_write_access");
Yongqiang Yang2f919712011-07-27 21:16:33 -0400804 err = ext4_journal_get_write_access(handle, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500805 if (unlikely(err))
Eric Sandeen37be2f52012-11-08 11:22:46 -0500806 goto exit_dind;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700807
liang xie5d601252014-05-12 22:06:43 -0400808 BUFFER_TRACE(dind, "get_write_access");
Theodore Ts'ob4097142011-01-10 12:46:59 -0500809 err = ext4_journal_get_write_access(handle, dind);
810 if (unlikely(err))
811 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812
Mingming Cao617ba132006-10-11 01:20:53 -0700813 /* ext4_reserve_inode_write() gets a reference on the iloc */
Theodore Ts'ob4097142011-01-10 12:46:59 -0500814 err = ext4_reserve_inode_write(handle, inode, &iloc);
815 if (unlikely(err))
Eric Sandeen37be2f52012-11-08 11:22:46 -0500816 goto exit_dind;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700817
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400818 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
819 sizeof(struct buffer_head *),
820 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821 if (!n_group_desc) {
822 err = -ENOMEM;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400823 ext4_warning(sb, "not enough memory for %lu groups",
824 gdb_num + 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700825 goto exit_inode;
826 }
827
828 /*
829 * Finally, we have all of the possible failures behind us...
830 *
831 * Remove new GDT block from inode double-indirect block and clear out
832 * the new GDT block for use (which also "frees" the backup GDT blocks
833 * from the reserved inode). We don't need to change the bitmaps for
834 * these blocks, because they are marked as in-use from being in the
835 * reserved inode, and will become GDT blocks (primary and backup).
836 */
Mingming Cao617ba132006-10-11 01:20:53 -0700837 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500838 err = ext4_handle_dirty_metadata(handle, NULL, dind);
839 if (unlikely(err)) {
840 ext4_std_error(sb, err);
841 goto exit_inode;
842 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700843 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700844 ext4_mark_iloc_dirty(handle, inode, &iloc);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400845 memset(gdb_bh->b_data, 0, sb->s_blocksize);
846 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500847 if (unlikely(err)) {
848 ext4_std_error(sb, err);
849 goto exit_inode;
850 }
851 brelse(dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700852
Mingming Cao617ba132006-10-11 01:20:53 -0700853 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700854 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700855 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Yongqiang Yang2f919712011-07-27 21:16:33 -0400856 n_group_desc[gdb_num] = gdb_bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700857 EXT4_SB(sb)->s_group_desc = n_group_desc;
858 EXT4_SB(sb)->s_gdb_count++;
Al Virob93b41d2014-11-20 12:19:11 -0500859 kvfree(o_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860
Marcin Slusarze8546d02008-04-17 10:38:59 -0400861 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -0400862 err = ext4_handle_dirty_super(handle, sb);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500863 if (err)
864 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700865
Theodore Ts'ob4097142011-01-10 12:46:59 -0500866 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867
868exit_inode:
Al Virob93b41d2014-11-20 12:19:11 -0500869 kvfree(n_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871exit_dind:
872 brelse(dind);
873exit_bh:
Yongqiang Yang2f919712011-07-27 21:16:33 -0400874 brelse(gdb_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875
Mingming Cao617ba132006-10-11 01:20:53 -0700876 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700877 return err;
878}
879
880/*
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400881 * add_new_gdb_meta_bg is the sister of add_new_gdb.
882 */
883static int add_new_gdb_meta_bg(struct super_block *sb,
884 handle_t *handle, ext4_group_t group) {
885 ext4_fsblk_t gdblock;
886 struct buffer_head *gdb_bh;
887 struct buffer_head **o_group_desc, **n_group_desc;
888 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
889 int err;
890
891 gdblock = ext4_meta_bg_first_block_no(sb, group) +
892 ext4_bg_has_super(sb, group);
893 gdb_bh = sb_bread(sb, gdblock);
894 if (!gdb_bh)
895 return -EIO;
896 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
897 sizeof(struct buffer_head *),
898 GFP_NOFS);
899 if (!n_group_desc) {
900 err = -ENOMEM;
901 ext4_warning(sb, "not enough memory for %lu groups",
902 gdb_num + 1);
903 return err;
904 }
905
906 o_group_desc = EXT4_SB(sb)->s_group_desc;
907 memcpy(n_group_desc, o_group_desc,
908 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
909 n_group_desc[gdb_num] = gdb_bh;
910 EXT4_SB(sb)->s_group_desc = n_group_desc;
911 EXT4_SB(sb)->s_gdb_count++;
Al Virob93b41d2014-11-20 12:19:11 -0500912 kvfree(o_group_desc);
liang xie5d601252014-05-12 22:06:43 -0400913 BUFFER_TRACE(gdb_bh, "get_write_access");
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400914 err = ext4_journal_get_write_access(handle, gdb_bh);
915 if (unlikely(err))
916 brelse(gdb_bh);
917 return err;
918}
919
920/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700921 * Called when we are adding a new group which has a backup copy of each of
922 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
923 * We need to add these reserved backup GDT blocks to the resize inode, so
924 * that they are kept for future resizing and not allocated to files.
925 *
926 * Each reserved backup GDT block will go into a different indirect block.
927 * The indirect blocks are actually the primary reserved GDT blocks,
928 * so we know in advance what their block numbers are. We only get the
929 * double-indirect block to verify it is pointing to the primary reserved
930 * GDT blocks so we don't overwrite a data block by accident. The reserved
931 * backup GDT blocks are stored in their reserved primary GDT block.
932 */
933static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400934 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700935{
936 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700937 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700938 struct buffer_head **primary;
939 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700940 struct ext4_iloc iloc;
941 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942 __le32 *data, *end;
943 int gdbackups = 0;
944 int res, i;
945 int err;
946
Josef Bacik216553c2008-04-29 22:02:02 -0400947 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948 if (!primary)
949 return -ENOMEM;
950
Mingming Cao617ba132006-10-11 01:20:53 -0700951 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700952 dind = sb_bread(sb, le32_to_cpu(*data));
953 if (!dind) {
954 err = -EIO;
955 goto exit_free;
956 }
957
Mingming Cao617ba132006-10-11 01:20:53 -0700958 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400959 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
960 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700961 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700962
963 /* Get each reserved primary GDT block and verify it holds backups */
964 for (res = 0; res < reserved_gdb; res++, blk++) {
965 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500966 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700967 " not at offset %ld",
968 blk,
969 (long)(data - (__le32 *)dind->b_data));
970 err = -EINVAL;
971 goto exit_bh;
972 }
973 primary[res] = sb_bread(sb, blk);
974 if (!primary[res]) {
975 err = -EIO;
976 goto exit_bh;
977 }
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500978 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
979 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980 brelse(primary[res]);
981 err = gdbackups;
982 goto exit_bh;
983 }
984 if (++data >= end)
985 data = (__le32 *)dind->b_data;
986 }
987
988 for (i = 0; i < reserved_gdb; i++) {
liang xie5d601252014-05-12 22:06:43 -0400989 BUFFER_TRACE(primary[i], "get_write_access");
Eric Sandeen37be2f52012-11-08 11:22:46 -0500990 if ((err = ext4_journal_get_write_access(handle, primary[i])))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700991 goto exit_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992 }
993
Mingming Cao617ba132006-10-11 01:20:53 -0700994 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 goto exit_bh;
996
997 /*
998 * Finally we can add each of the reserved backup GDT blocks from
999 * the new group to its reserved primary GDT block.
1000 */
Yongqiang Yang668f4dc2011-07-27 21:23:13 -04001001 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001002 for (i = 0; i < reserved_gdb; i++) {
1003 int err2;
1004 data = (__le32 *)primary[i]->b_data;
1005 /* printk("reserving backup %lu[%u] = %lu\n",
1006 primary[i]->b_blocknr, gdbackups,
1007 blk + primary[i]->b_blocknr); */
1008 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -05001009 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001010 if (!err)
1011 err = err2;
1012 }
1013 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -07001014 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015
1016exit_bh:
1017 while (--res >= 0)
1018 brelse(primary[res]);
1019 brelse(dind);
1020
1021exit_free:
1022 kfree(primary);
1023
1024 return err;
1025}
1026
1027/*
Mingming Cao617ba132006-10-11 01:20:53 -07001028 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001029 * of the main resize transaction, because e2fsck will re-write them if there
1030 * is a problem (basically only OOM will cause a problem). However, we
1031 * _should_ update the backups if possible, in case the primary gets trashed
1032 * for some reason and we need to run e2fsck from a backup superblock. The
1033 * important part is that the new block and inode counts are in the backup
1034 * superblocks, and the location of the new group metadata in the GDT backups.
1035 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001036 * We do not need take the s_resize_lock for this, because these
1037 * blocks are not otherwise touched by the filesystem code when it is
1038 * mounted. We don't need to worry about last changing from
1039 * sbi->s_groups_count, because the worst that can happen is that we
1040 * do not copy the full number of backups at this time. The resize
1041 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042 */
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001043static void update_backups(struct super_block *sb, int blk_off, char *data,
1044 int size, int meta_bg)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001045{
Mingming Cao617ba132006-10-11 01:20:53 -07001046 struct ext4_sb_info *sbi = EXT4_SB(sb);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001047 ext4_group_t last;
Mingming Cao617ba132006-10-11 01:20:53 -07001048 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001049 unsigned three = 1;
1050 unsigned five = 5;
1051 unsigned seven = 7;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001052 ext4_group_t group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053 int rest = sb->s_blocksize - size;
1054 handle_t *handle;
1055 int err = 0, err2;
1056
Theodore Ts'o9924a922013-02-08 21:59:22 -05001057 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001058 if (IS_ERR(handle)) {
1059 group = 1;
1060 err = PTR_ERR(handle);
1061 goto exit_err;
1062 }
1063
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001064 if (meta_bg == 0) {
1065 group = ext4_list_backups(sb, &three, &five, &seven);
1066 last = sbi->s_groups_count;
1067 } else {
1068 group = ext4_meta_bg_first_group(sb, group) + 1;
1069 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1070 }
1071
1072 while (group < sbi->s_groups_count) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001073 struct buffer_head *bh;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001074 ext4_fsblk_t backup_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001075
1076 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -05001077 if (ext4_handle_valid(handle) &&
1078 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -07001079 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1080 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001081 break;
1082
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001083 if (meta_bg == 0)
Jan Kara9378c672014-10-30 10:52:57 -04001084 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001085 else
1086 backup_block = (ext4_group_first_block_no(sb, group) +
1087 ext4_bg_has_super(sb, group));
1088
1089 bh = sb_getblk(sb, backup_block);
Wang Shilongaebf0242013-01-12 16:28:47 -05001090 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -05001091 err = -ENOMEM;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001092 break;
1093 }
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001094 ext4_debug("update metadata backup %llu(+%llu)\n",
1095 backup_block, backup_block -
1096 ext4_group_first_block_no(sb, group));
liang xie5d601252014-05-12 22:06:43 -04001097 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001098 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001099 break;
1100 lock_buffer(bh);
1101 memcpy(bh->b_data, data, size);
1102 if (rest)
1103 memset(bh->b_data + size, 0, rest);
1104 set_buffer_uptodate(bh);
1105 unlock_buffer(bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05001106 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1107 if (unlikely(err))
1108 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001109 brelse(bh);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001110
1111 if (meta_bg == 0)
1112 group = ext4_list_backups(sb, &three, &five, &seven);
1113 else if (group == last)
1114 break;
1115 else
1116 group = last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001117 }
Mingming Cao617ba132006-10-11 01:20:53 -07001118 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001119 err = err2;
1120
1121 /*
1122 * Ugh! Need to have e2fsck write the backup copies. It is too
1123 * late to revert the resize, we shouldn't fail just because of
1124 * the backup copies (they are only needed in case of corruption).
1125 *
1126 * However, if we got here we have a journal problem too, so we
1127 * can't really start a transaction to mark the superblock.
1128 * Chicken out and just set the flag on the hope it will be written
1129 * to disk, and if not - we will simply wait until next fsck.
1130 */
1131exit_err:
1132 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001133 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001134 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -07001135 sbi->s_mount_state &= ~EXT4_VALID_FS;
1136 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137 mark_buffer_dirty(sbi->s_sbh);
1138 }
1139}
1140
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001141/*
1142 * ext4_add_new_descs() adds @count group descriptor of groups
1143 * starting at @group
1144 *
1145 * @handle: journal handle
1146 * @sb: super block
1147 * @group: the group no. of the first group desc to be added
1148 * @resize_inode: the resize inode
1149 * @count: number of group descriptors to be added
1150 */
1151static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1152 ext4_group_t group, struct inode *resize_inode,
1153 ext4_group_t count)
1154{
1155 struct ext4_sb_info *sbi = EXT4_SB(sb);
1156 struct ext4_super_block *es = sbi->s_es;
1157 struct buffer_head *gdb_bh;
1158 int i, gdb_off, gdb_num, err = 0;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001159 int meta_bg;
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001160
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001161 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001162 for (i = 0; i < count; i++, group++) {
1163 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1164 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1165
1166 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1167 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1168
1169 /*
1170 * We will only either add reserved group blocks to a backup group
1171 * or remove reserved blocks for the first group in a new group block.
1172 * Doing both would be mean more complex code, and sane people don't
1173 * use non-sparse filesystems anymore. This is already checked above.
1174 */
1175 if (gdb_off) {
1176 gdb_bh = sbi->s_group_desc[gdb_num];
liang xie5d601252014-05-12 22:06:43 -04001177 BUFFER_TRACE(gdb_bh, "get_write_access");
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001178 err = ext4_journal_get_write_access(handle, gdb_bh);
1179
1180 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1181 err = reserve_backup_gdb(handle, resize_inode, group);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001182 } else if (meta_bg != 0) {
1183 err = add_new_gdb_meta_bg(sb, handle, group);
1184 } else {
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001185 err = add_new_gdb(handle, resize_inode, group);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001186 }
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001187 if (err)
1188 break;
1189 }
1190 return err;
1191}
1192
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001193static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1194{
1195 struct buffer_head *bh = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -05001196 if (unlikely(!bh))
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001197 return NULL;
Dmitry Monakhov7f1468d2012-09-25 23:19:25 -04001198 if (!bh_uptodate_or_lock(bh)) {
1199 if (bh_submit_read(bh) < 0) {
1200 brelse(bh);
1201 return NULL;
1202 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001203 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001204
1205 return bh;
1206}
1207
1208static int ext4_set_bitmap_checksums(struct super_block *sb,
1209 ext4_group_t group,
1210 struct ext4_group_desc *gdp,
1211 struct ext4_new_group_data *group_data)
1212{
1213 struct buffer_head *bh;
1214
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -04001215 if (!ext4_has_metadata_csum(sb))
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001216 return 0;
1217
1218 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1219 if (!bh)
1220 return -EIO;
1221 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1222 EXT4_INODES_PER_GROUP(sb) / 8);
1223 brelse(bh);
1224
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -04001225 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1226 if (!bh)
1227 return -EIO;
Tao Ma79f1ba42012-10-22 00:34:32 -04001228 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -04001229 brelse(bh);
1230
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001231 return 0;
1232}
1233
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001234/*
1235 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1236 */
1237static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1238 struct ext4_new_flex_group_data *flex_gd)
1239{
1240 struct ext4_new_group_data *group_data = flex_gd->groups;
1241 struct ext4_group_desc *gdp;
1242 struct ext4_sb_info *sbi = EXT4_SB(sb);
1243 struct buffer_head *gdb_bh;
1244 ext4_group_t group;
1245 __u16 *bg_flags = flex_gd->bg_flags;
1246 int i, gdb_off, gdb_num, err = 0;
1247
1248
1249 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1250 group = group_data->group;
1251
1252 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1253 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1254
1255 /*
1256 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1257 */
1258 gdb_bh = sbi->s_group_desc[gdb_num];
1259 /* Update group descriptor block for new group */
Theodore Ts'o2716b802012-05-28 17:47:52 -04001260 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001261 gdb_off * EXT4_DESC_SIZE(sb));
1262
1263 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1264 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1265 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001266 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1267 if (err) {
1268 ext4_std_error(sb, err);
1269 break;
1270 }
1271
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001272 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1273 ext4_free_group_clusters_set(sb, gdp,
Lukas Czerner810da242013-03-02 17:18:58 -05001274 EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001275 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'o93f90522012-09-12 14:32:42 -04001276 if (ext4_has_group_desc_csum(sb))
1277 ext4_itable_unused_set(sb, gdp,
1278 EXT4_INODES_PER_GROUP(sb));
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001279 gdp->bg_flags = cpu_to_le16(*bg_flags);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001280 ext4_group_desc_csum_set(sb, group, gdp);
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001281
1282 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1283 if (unlikely(err)) {
1284 ext4_std_error(sb, err);
1285 break;
1286 }
1287
1288 /*
1289 * We can allocate memory for mb_alloc based on the new group
1290 * descriptor
1291 */
1292 err = ext4_mb_add_groupinfo(sb, group, gdp);
1293 if (err)
1294 break;
1295 }
1296 return err;
1297}
1298
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001299/*
1300 * ext4_update_super() updates the super block so that the newly added
1301 * groups can be seen by the filesystem.
1302 *
1303 * @sb: super block
1304 * @flex_gd: new added groups
1305 */
1306static void ext4_update_super(struct super_block *sb,
1307 struct ext4_new_flex_group_data *flex_gd)
1308{
1309 ext4_fsblk_t blocks_count = 0;
1310 ext4_fsblk_t free_blocks = 0;
1311 ext4_fsblk_t reserved_blocks = 0;
1312 struct ext4_new_group_data *group_data = flex_gd->groups;
1313 struct ext4_sb_info *sbi = EXT4_SB(sb);
1314 struct ext4_super_block *es = sbi->s_es;
Theodore Ts'o8a991842012-07-22 20:23:31 -04001315 int i;
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001316
1317 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1318 /*
1319 * Make the new blocks and inodes valid next. We do this before
1320 * increasing the group count so that once the group is enabled,
1321 * all of its blocks and inodes are already valid.
1322 *
1323 * We always allocate group-by-group, then block-by-block or
1324 * inode-by-inode within a group, so enabling these
1325 * blocks/inodes before the group is live won't actually let us
1326 * allocate the new space yet.
1327 */
1328 for (i = 0; i < flex_gd->count; i++) {
1329 blocks_count += group_data[i].blocks_count;
1330 free_blocks += group_data[i].free_blocks_count;
1331 }
1332
1333 reserved_blocks = ext4_r_blocks_count(es) * 100;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001334 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001335 reserved_blocks *= blocks_count;
1336 do_div(reserved_blocks, 100);
1337
1338 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001339 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001340 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1341 flex_gd->count);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001342 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1343 flex_gd->count);
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001344
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001345 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001346 /*
1347 * We need to protect s_groups_count against other CPUs seeing
1348 * inconsistent state in the superblock.
1349 *
1350 * The precise rules we use are:
1351 *
1352 * * Writers must perform a smp_wmb() after updating all
1353 * dependent data and before modifying the groups count
1354 *
1355 * * Readers must perform an smp_rmb() after reading the groups
1356 * count and before reading any dependent data.
1357 *
1358 * NB. These rules can be relaxed when checking the group count
1359 * while freeing data, as we can only allocate from a block
1360 * group after serialising against the group count, and we can
1361 * only then free after serialising in turn against that
1362 * allocation.
1363 */
1364 smp_wmb();
1365
1366 /* Update the global fs size fields */
1367 sbi->s_groups_count += flex_gd->count;
Theodore Ts'oc5c72d82013-04-21 20:19:43 -04001368 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1369 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001370
1371 /* Update the reserved block counts only once the new group is
1372 * active. */
1373 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1374 reserved_blocks);
1375
1376 /* Update the free space counts */
1377 percpu_counter_add(&sbi->s_freeclusters_counter,
Lukas Czerner810da242013-03-02 17:18:58 -05001378 EXT4_NUM_B2C(sbi, free_blocks));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001379 percpu_counter_add(&sbi->s_freeinodes_counter,
1380 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1381
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001382 ext4_debug("free blocks count %llu",
1383 percpu_counter_read(&sbi->s_freeclusters_counter));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001384 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1385 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1386 sbi->s_log_groups_per_flex) {
1387 ext4_group_t flex_group;
1388 flex_group = ext4_flex_group(sbi, group_data[0].group);
Theodore Ts'o90ba9832013-03-11 23:39:59 -04001389 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1390 &sbi->s_flex_groups[flex_group].free_clusters);
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001391 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1392 &sbi->s_flex_groups[flex_group].free_inodes);
1393 }
1394
Theodore Ts'o952fc182012-07-09 16:27:05 -04001395 /*
1396 * Update the fs overhead information
1397 */
1398 ext4_calculate_overhead(sb);
1399
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001400 if (test_opt(sb, DEBUG))
1401 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1402 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1403 blocks_count, free_blocks, reserved_blocks);
1404}
1405
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001406/* Add a flex group to an fs. Ensure we handle all possible error conditions
1407 * _before_ we start modifying the filesystem, because we cannot abort the
1408 * transaction and not have it write the data to disk.
1409 */
1410static int ext4_flex_group_add(struct super_block *sb,
1411 struct inode *resize_inode,
1412 struct ext4_new_flex_group_data *flex_gd)
1413{
1414 struct ext4_sb_info *sbi = EXT4_SB(sb);
1415 struct ext4_super_block *es = sbi->s_es;
1416 ext4_fsblk_t o_blocks_count;
1417 ext4_grpblk_t last;
1418 ext4_group_t group;
1419 handle_t *handle;
1420 unsigned reserved_gdb;
1421 int err = 0, err2 = 0, credit;
1422
1423 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1424
1425 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1426 o_blocks_count = ext4_blocks_count(es);
1427 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1428 BUG_ON(last);
1429
1430 err = setup_new_flex_group_blocks(sb, flex_gd);
1431 if (err)
1432 goto exit;
1433 /*
1434 * We will always be modifying at least the superblock and GDT
Jan Kara2c869b22015-05-02 23:58:32 -04001435 * blocks. If we are adding a group past the last current GDT block,
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001436 * we will also modify the inode and the dindirect block. If we
1437 * are adding a group with superblock/GDT backups we will also
1438 * modify each of the reserved GDT dindirect blocks.
1439 */
Jan Kara2c869b22015-05-02 23:58:32 -04001440 credit = 3; /* sb, resize inode, resize inode dindirect */
1441 /* GDT blocks */
1442 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1443 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
Theodore Ts'o9924a922013-02-08 21:59:22 -05001444 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001445 if (IS_ERR(handle)) {
1446 err = PTR_ERR(handle);
1447 goto exit;
1448 }
1449
liang xie5d601252014-05-12 22:06:43 -04001450 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001451 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1452 if (err)
1453 goto exit_journal;
1454
1455 group = flex_gd->groups[0].group;
1456 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1457 err = ext4_add_new_descs(handle, sb, group,
1458 resize_inode, flex_gd->count);
1459 if (err)
1460 goto exit_journal;
1461
1462 err = ext4_setup_new_descs(handle, sb, flex_gd);
1463 if (err)
1464 goto exit_journal;
1465
1466 ext4_update_super(sb, flex_gd);
1467
1468 err = ext4_handle_dirty_super(handle, sb);
1469
1470exit_journal:
1471 err2 = ext4_journal_stop(handle);
1472 if (!err)
1473 err = err2;
1474
1475 if (!err) {
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001476 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1477 int gdb_num_end = ((group + flex_gd->count - 1) /
1478 EXT4_DESC_PER_BLOCK(sb));
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001479 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1480 EXT4_FEATURE_INCOMPAT_META_BG);
Tao Ma0acdb882012-09-26 00:08:57 -04001481 sector_t old_gdb = 0;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001482
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001483 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001484 sizeof(struct ext4_super_block), 0);
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001485 for (; gdb_num <= gdb_num_end; gdb_num++) {
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001486 struct buffer_head *gdb_bh;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001487
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001488 gdb_bh = sbi->s_group_desc[gdb_num];
Tao Ma0acdb882012-09-26 00:08:57 -04001489 if (old_gdb == gdb_bh->b_blocknr)
1490 continue;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001491 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001492 gdb_bh->b_size, meta_bg);
Tao Ma0acdb882012-09-26 00:08:57 -04001493 old_gdb = gdb_bh->b_blocknr;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001494 }
1495 }
1496exit:
1497 return err;
1498}
1499
Yongqiang Yang19c52462012-01-04 17:09:44 -05001500static int ext4_setup_next_flex_gd(struct super_block *sb,
1501 struct ext4_new_flex_group_data *flex_gd,
1502 ext4_fsblk_t n_blocks_count,
1503 unsigned long flexbg_size)
1504{
1505 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1506 struct ext4_new_group_data *group_data = flex_gd->groups;
1507 ext4_fsblk_t o_blocks_count;
1508 ext4_group_t n_group;
1509 ext4_group_t group;
1510 ext4_group_t last_group;
1511 ext4_grpblk_t last;
1512 ext4_grpblk_t blocks_per_group;
1513 unsigned long i;
1514
1515 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1516
1517 o_blocks_count = ext4_blocks_count(es);
1518
1519 if (o_blocks_count == n_blocks_count)
1520 return 0;
1521
1522 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1523 BUG_ON(last);
1524 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1525
1526 last_group = group | (flexbg_size - 1);
1527 if (last_group > n_group)
1528 last_group = n_group;
1529
1530 flex_gd->count = last_group - group + 1;
1531
1532 for (i = 0; i < flex_gd->count; i++) {
1533 int overhead;
1534
1535 group_data[i].group = group + i;
1536 group_data[i].blocks_count = blocks_per_group;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001537 overhead = ext4_group_overhead_blocks(sb, group + i);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001538 group_data[i].free_blocks_count = blocks_per_group - overhead;
Theodore Ts'o7f511862013-01-13 08:41:45 -05001539 if (ext4_has_group_desc_csum(sb)) {
Yongqiang Yang19c52462012-01-04 17:09:44 -05001540 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1541 EXT4_BG_INODE_UNINIT;
Theodore Ts'o7f511862013-01-13 08:41:45 -05001542 if (!test_opt(sb, INIT_INODE_TABLE))
1543 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1544 } else
Yongqiang Yang19c52462012-01-04 17:09:44 -05001545 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1546 }
1547
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001548 if (last_group == n_group && ext4_has_group_desc_csum(sb))
Yongqiang Yang19c52462012-01-04 17:09:44 -05001549 /* We need to initialize block bitmap of last group. */
1550 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1551
1552 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1553 group_data[i - 1].blocks_count = last + 1;
1554 group_data[i - 1].free_blocks_count -= blocks_per_group-
1555 last - 1;
1556 }
1557
1558 return 1;
1559}
1560
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001561/* Add group descriptor data to an existing or new group descriptor block.
1562 * Ensure we handle all possible error conditions _before_ we start modifying
1563 * the filesystem, because we cannot abort the transaction and not have it
1564 * write the data to disk.
1565 *
1566 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1567 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1568 *
1569 * We only need to hold the superblock lock while we are actually adding
1570 * in the new group's counts to the superblock. Prior to that we have
1571 * not really "added" the group at all. We re-check that we are still
1572 * adding in the last group in case things have changed since verifying.
1573 */
Mingming Cao617ba132006-10-11 01:20:53 -07001574int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001575{
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001576 struct ext4_new_flex_group_data flex_gd;
Mingming Cao617ba132006-10-11 01:20:53 -07001577 struct ext4_sb_info *sbi = EXT4_SB(sb);
1578 struct ext4_super_block *es = sbi->s_es;
1579 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001580 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001581 struct inode *inode = NULL;
Jon Ernst03b40e32013-06-17 08:56:26 -04001582 int gdb_off;
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001583 int err;
1584 __u16 bg_flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001585
Mingming Cao617ba132006-10-11 01:20:53 -07001586 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587
Mingming Cao617ba132006-10-11 01:20:53 -07001588 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1589 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001590 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001591 return -EPERM;
1592 }
1593
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001594 if (ext4_blocks_count(es) + input->blocks_count <
1595 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001596 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001597 return -EINVAL;
1598 }
1599
Mingming Cao617ba132006-10-11 01:20:53 -07001600 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001601 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001602 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001603 return -EINVAL;
1604 }
1605
1606 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001607 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -04001608 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1609 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001610 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001611 "No reserved GDT blocks, can't resize");
1612 return -EPERM;
1613 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001614 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1615 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001616 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -08001617 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001618 }
1619 }
1620
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001621
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001622 err = verify_group_input(sb, input);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -04001623 if (err)
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001624 goto out;
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001625
Theodore Ts'o117fff12012-09-05 01:29:50 -04001626 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1627 if (err)
Theodore Ts'o7f511862013-01-13 08:41:45 -05001628 goto out;
Theodore Ts'o117fff12012-09-05 01:29:50 -04001629
Theodore Ts'o28623c22012-09-05 01:31:50 -04001630 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1631 if (err)
1632 goto out;
1633
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001634 flex_gd.count = 1;
1635 flex_gd.groups = input;
1636 flex_gd.bg_flags = &bg_flags;
1637 err = ext4_flex_group_add(sb, inode, &flex_gd);
1638out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639 iput(inode);
1640 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001641} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001642
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001643/*
Yongqiang Yang18e31432012-01-03 23:18:50 -05001644 * extend a group without checking assuming that checking has been done.
1645 */
1646static int ext4_group_extend_no_check(struct super_block *sb,
1647 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1648{
1649 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1650 handle_t *handle;
1651 int err = 0, err2;
1652
1653 /* We will update the superblock, one block bitmap, and
1654 * one group descriptor via ext4_group_add_blocks().
1655 */
Theodore Ts'o9924a922013-02-08 21:59:22 -05001656 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001657 if (IS_ERR(handle)) {
1658 err = PTR_ERR(handle);
1659 ext4_warning(sb, "error %d on journal start", err);
1660 return err;
1661 }
1662
liang xie5d601252014-05-12 22:06:43 -04001663 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
Yongqiang Yang18e31432012-01-03 23:18:50 -05001664 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1665 if (err) {
1666 ext4_warning(sb, "error %d on journal write access", err);
1667 goto errout;
1668 }
1669
1670 ext4_blocks_count_set(es, o_blocks_count + add);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001671 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001672 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1673 o_blocks_count + add);
1674 /* We add the blocks to the bitmap and set the group need init bit */
1675 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1676 if (err)
1677 goto errout;
1678 ext4_handle_dirty_super(handle, sb);
1679 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1680 o_blocks_count + add);
1681errout:
1682 err2 = ext4_journal_stop(handle);
1683 if (err2 && !err)
1684 err = err2;
1685
1686 if (!err) {
1687 if (test_opt(sb, DEBUG))
1688 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1689 "blocks\n", ext4_blocks_count(es));
Maarten ter Huurne6ca792e2013-07-01 08:12:08 -04001690 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001691 (char *)es, sizeof(struct ext4_super_block), 0);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001692 }
1693 return err;
1694}
1695
1696/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001697 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001698 * point is only used to extend the current filesystem to the end of the last
1699 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1700 * for emergencies (because it has no dependencies on reserved blocks).
1701 *
Mingming Cao617ba132006-10-11 01:20:53 -07001702 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001703 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1704 * GDT blocks are reserved to grow to the desired size.
1705 */
Mingming Cao617ba132006-10-11 01:20:53 -07001706int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1707 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001708{
Mingming Cao617ba132006-10-11 01:20:53 -07001709 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001710 ext4_grpblk_t last;
1711 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001712 struct buffer_head *bh;
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001713 int err;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001714 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001715
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001716 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001717
1718 if (test_opt(sb, DEBUG))
Theodore Ts'o92b97812012-03-19 23:41:49 -04001719 ext4_msg(sb, KERN_DEBUG,
1720 "extending last group from %llu to %llu blocks",
1721 o_blocks_count, n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001722
1723 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1724 return 0;
1725
1726 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Theodore Ts'o92b97812012-03-19 23:41:49 -04001727 ext4_msg(sb, KERN_ERR,
1728 "filesystem too large to resize to %llu blocks safely",
1729 n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001730 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001731 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001732 return -EINVAL;
1733 }
1734
1735 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001736 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001737 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001738 }
1739
1740 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001741 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742
1743 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001744 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001745 return -EPERM;
1746 }
1747
Mingming Cao617ba132006-10-11 01:20:53 -07001748 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001749
1750 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001751 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752 return -EINVAL;
1753 }
1754
1755 if (o_blocks_count + add > n_blocks_count)
1756 add = n_blocks_count - o_blocks_count;
1757
1758 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001759 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001760 o_blocks_count + add, add);
1761
1762 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001763 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001765 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001766 return -ENOSPC;
1767 }
1768 brelse(bh);
1769
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001770 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001771 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001772} /* ext4_group_extend */
Yongqiang Yang19c52462012-01-04 17:09:44 -05001773
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001774
1775static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1776{
1777 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1778}
1779
1780/*
1781 * Release the resize inode and drop the resize_inode feature if there
1782 * are no more reserved gdt blocks, and then convert the file system
1783 * to enable meta_bg
1784 */
1785static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1786{
1787 handle_t *handle;
1788 struct ext4_sb_info *sbi = EXT4_SB(sb);
1789 struct ext4_super_block *es = sbi->s_es;
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001790 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001791 ext4_fsblk_t nr;
1792 int i, ret, err = 0;
1793 int credits = 1;
1794
1795 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001796 if (inode) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001797 if (es->s_reserved_gdt_blocks) {
1798 ext4_error(sb, "Unexpected non-zero "
1799 "s_reserved_gdt_blocks");
1800 return -EPERM;
1801 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001802
1803 /* Do a quick sanity check of the resize inode */
1804 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1805 goto invalid_resize_inode;
1806 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1807 if (i == EXT4_DIND_BLOCK) {
1808 if (ei->i_data[i])
1809 continue;
1810 else
1811 goto invalid_resize_inode;
1812 }
1813 if (ei->i_data[i])
1814 goto invalid_resize_inode;
1815 }
1816 credits += 3; /* block bitmap, bg descriptor, resize inode */
1817 }
1818
Theodore Ts'o9924a922013-02-08 21:59:22 -05001819 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001820 if (IS_ERR(handle))
1821 return PTR_ERR(handle);
1822
liang xie5d601252014-05-12 22:06:43 -04001823 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001824 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1825 if (err)
1826 goto errout;
1827
1828 EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1829 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1830 sbi->s_es->s_first_meta_bg =
1831 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1832
1833 err = ext4_handle_dirty_super(handle, sb);
1834 if (err) {
1835 ext4_std_error(sb, err);
1836 goto errout;
1837 }
1838
1839 if (inode) {
1840 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1841 ext4_free_blocks(handle, inode, NULL, nr, 1,
1842 EXT4_FREE_BLOCKS_METADATA |
1843 EXT4_FREE_BLOCKS_FORGET);
1844 ei->i_data[EXT4_DIND_BLOCK] = 0;
1845 inode->i_blocks = 0;
1846
1847 err = ext4_mark_inode_dirty(handle, inode);
1848 if (err)
1849 ext4_std_error(sb, err);
1850 }
1851
1852errout:
1853 ret = ext4_journal_stop(handle);
1854 if (!err)
1855 err = ret;
1856 return ret;
1857
1858invalid_resize_inode:
1859 ext4_error(sb, "corrupted/inconsistent resize inode");
1860 return -EINVAL;
1861}
1862
Yongqiang Yang19c52462012-01-04 17:09:44 -05001863/*
1864 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1865 *
1866 * @sb: super block of the fs to be resized
1867 * @n_blocks_count: the number of blocks resides in the resized fs
1868 */
1869int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1870{
1871 struct ext4_new_flex_group_data *flex_gd = NULL;
1872 struct ext4_sb_info *sbi = EXT4_SB(sb);
1873 struct ext4_super_block *es = sbi->s_es;
1874 struct buffer_head *bh;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001875 struct inode *resize_inode = NULL;
1876 ext4_grpblk_t add, offset;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001877 unsigned long n_desc_blocks;
1878 unsigned long o_desc_blocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001879 ext4_group_t o_group;
1880 ext4_group_t n_group;
1881 ext4_fsblk_t o_blocks_count;
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001882 ext4_fsblk_t n_blocks_count_retry = 0;
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001883 unsigned long last_update_time = 0;
Theodore Ts'o117fff12012-09-05 01:29:50 -04001884 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001885 int meta_bg;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001886
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001887 /* See if the device is actually as big as what was requested */
1888 bh = sb_bread(sb, n_blocks_count - 1);
1889 if (!bh) {
1890 ext4_warning(sb, "can't read last block, resize aborted");
1891 return -ENOSPC;
1892 }
1893 brelse(bh);
1894
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001895retry:
Yongqiang Yang19c52462012-01-04 17:09:44 -05001896 o_blocks_count = ext4_blocks_count(es);
1897
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001898 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1899 "to %llu blocks", o_blocks_count, n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001900
1901 if (n_blocks_count < o_blocks_count) {
1902 /* On-line shrinking not supported */
1903 ext4_warning(sb, "can't shrink FS - resize aborted");
1904 return -EINVAL;
1905 }
1906
1907 if (n_blocks_count == o_blocks_count)
1908 /* Nothing need to do */
1909 return 0;
1910
Lukas Czernerbd862982013-04-03 23:32:34 -04001911 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
Theodore Ts'o3f8a6412013-04-21 22:56:32 -04001912 if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1913 ext4_warning(sb, "resize would cause inodes_count overflow");
1914 return -EINVAL;
1915 }
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001916 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001917
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001918 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1919 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001920
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001921 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001922
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001923 if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1924 if (meta_bg) {
1925 ext4_error(sb, "resize_inode and meta_bg enabled "
1926 "simultaneously");
1927 return -EINVAL;
1928 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001929 if (n_desc_blocks > o_desc_blocks +
1930 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1931 n_blocks_count_retry = n_blocks_count;
1932 n_desc_blocks = o_desc_blocks +
1933 le16_to_cpu(es->s_reserved_gdt_blocks);
1934 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1935 n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1936 n_group--; /* set to last group number */
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001937 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001938
1939 if (!resize_inode)
1940 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001941 if (IS_ERR(resize_inode)) {
1942 ext4_warning(sb, "Error opening resize inode");
1943 return PTR_ERR(resize_inode);
1944 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001945 }
1946
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001947 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001948 err = ext4_convert_meta_bg(sb, resize_inode);
1949 if (err)
1950 goto out;
1951 if (resize_inode) {
1952 iput(resize_inode);
1953 resize_inode = NULL;
1954 }
1955 if (n_blocks_count_retry) {
1956 n_blocks_count = n_blocks_count_retry;
1957 n_blocks_count_retry = 0;
1958 goto retry;
1959 }
Yongqiang Yang19c52462012-01-04 17:09:44 -05001960 }
1961
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001962 /* extend the last group */
1963 if (n_group == o_group)
1964 add = n_blocks_count - o_blocks_count;
1965 else
1966 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1967 if (add > 0) {
Yongqiang Yang19c52462012-01-04 17:09:44 -05001968 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1969 if (err)
1970 goto out;
1971 }
1972
Yongqiang Yangd7574ad2012-09-05 01:23:50 -04001973 if (ext4_blocks_count(es) == n_blocks_count)
Yongqiang Yang19c52462012-01-04 17:09:44 -05001974 goto out;
1975
Theodore Ts'o117fff12012-09-05 01:29:50 -04001976 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1977 if (err)
1978 return err;
1979
Theodore Ts'o28623c22012-09-05 01:31:50 -04001980 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1981 if (err)
1982 goto out;
1983
Yongqiang Yang19c52462012-01-04 17:09:44 -05001984 flex_gd = alloc_flex_gd(flexbg_size);
1985 if (flex_gd == NULL) {
1986 err = -ENOMEM;
1987 goto out;
1988 }
1989
1990 /* Add flex groups. Note that a regular group is a
1991 * flex group with 1 group.
1992 */
1993 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1994 flexbg_size)) {
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001995 if (jiffies - last_update_time > HZ * 10) {
1996 if (last_update_time)
1997 ext4_msg(sb, KERN_INFO,
1998 "resized to %llu blocks",
1999 ext4_blocks_count(es));
2000 last_update_time = jiffies;
2001 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -04002002 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2003 break;
Yongqiang Yang19c52462012-01-04 17:09:44 -05002004 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2005 if (unlikely(err))
2006 break;
2007 }
2008
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04002009 if (!err && n_blocks_count_retry) {
2010 n_blocks_count = n_blocks_count_retry;
2011 n_blocks_count_retry = 0;
2012 free_flex_gd(flex_gd);
2013 flex_gd = NULL;
2014 goto retry;
2015 }
2016
Yongqiang Yang19c52462012-01-04 17:09:44 -05002017out:
2018 if (flex_gd)
2019 free_flex_gd(flex_gd);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04002020 if (resize_inode != NULL)
2021 iput(resize_inode);
Theodore Ts'o59e31c12012-09-19 00:55:56 -04002022 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05002023 return err;
2024}