blob: bf76f405a5f91df5f276e2a2d1dcf2935178d4a7 [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);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010045 smp_mb__after_atomic();
Yongqiang Yang8f82f842011-07-26 21:35:44 -040046}
47
Yongqiang Yang01f795f2012-09-05 01:33:50 -040048static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
49 ext4_group_t group) {
50 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
51 EXT4_DESC_PER_BLOCK_BITS(sb);
52}
53
54static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
55 ext4_group_t group) {
56 group = ext4_meta_bg_first_group(sb, group);
57 return ext4_group_first_block_no(sb, group);
58}
59
60static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
61 ext4_group_t group) {
62 ext4_grpblk_t overhead;
63 overhead = ext4_bg_num_gdb(sb, group);
64 if (ext4_bg_has_super(sb, group))
65 overhead += 1 +
66 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
67 return overhead;
68}
69
Dave Kleikampac27a0e2006-10-11 01:20:50 -070070#define outside(b, first, last) ((b) < (first) || (b) >= (last))
71#define inside(b, first, last) ((b) >= (first) && (b) < (last))
72
73static int verify_group_input(struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -070074 struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070075{
Mingming Cao617ba132006-10-11 01:20:53 -070076 struct ext4_sb_info *sbi = EXT4_SB(sb);
77 struct ext4_super_block *es = sbi->s_es;
Laurent Vivierbd81d8e2006-10-11 01:21:10 -070078 ext4_fsblk_t start = ext4_blocks_count(es);
Mingming Cao617ba132006-10-11 01:20:53 -070079 ext4_fsblk_t end = start + input->blocks_count;
Avantika Mathurfd2d4292008-01-28 23:58:27 -050080 ext4_group_t group = input->group;
Mingming Cao617ba132006-10-11 01:20:53 -070081 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
Theodore Ts'ob302ef22013-06-06 11:14:31 -040082 unsigned overhead;
83 ext4_fsblk_t metaend;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070084 struct buffer_head *bh = NULL;
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070085 ext4_grpblk_t free_blocks_count, offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086 int err = -EINVAL;
87
Theodore Ts'ob302ef22013-06-06 11:14:31 -040088 if (group != sbi->s_groups_count) {
89 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
90 input->group, sbi->s_groups_count);
91 return -EINVAL;
92 }
93
94 overhead = ext4_group_overhead_blocks(sb, group);
95 metaend = start + overhead;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096 input->free_blocks_count = free_blocks_count =
97 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
98
99 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -0700100 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700101 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -0700102 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103 "no-super", input->group, input->blocks_count,
104 free_blocks_count, input->reserved_blocks);
105
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700106 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Theodore Ts'ob302ef22013-06-06 11:14:31 -0400107 if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500108 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500110 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700111 input->reserved_blocks);
112 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500113 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700114 input->blocks_count);
115 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500116 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700117 end - 1);
118 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500119 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700120 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700121 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500122 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700123 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700124 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400125 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500126 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700127 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700128 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500129 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700130 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700131 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500132 ext4_warning(sb, "Block bitmap (%llu) in inode table "
133 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700134 (unsigned long long)input->block_bitmap,
135 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700136 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500137 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
138 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700139 (unsigned long long)input->inode_bitmap,
140 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700141 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500142 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700143 (unsigned long long)input->block_bitmap,
144 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500146 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700147 (unsigned long long)input->inode_bitmap,
148 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700149 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400150 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500151 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
152 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700153 (unsigned long long)input->inode_table,
154 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700155 else
156 err = 0;
157 brelse(bh);
158
159 return err;
160}
161
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500162/*
163 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
164 * group each time.
165 */
166struct ext4_new_flex_group_data {
167 struct ext4_new_group_data *groups; /* new_group_data for groups
168 in the flex group */
169 __u16 *bg_flags; /* block group flags of groups
170 in @groups */
171 ext4_group_t count; /* number of groups in @groups
172 */
173};
174
175/*
176 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
177 * @flexbg_size.
178 *
179 * Returns NULL on failure otherwise address of the allocated structure.
180 */
181static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
182{
183 struct ext4_new_flex_group_data *flex_gd;
184
185 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
186 if (flex_gd == NULL)
187 goto out3;
188
Haogang Chen967ac8a2012-05-28 14:21:55 -0400189 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
190 goto out2;
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500191 flex_gd->count = flexbg_size;
192
193 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
194 flexbg_size, GFP_NOFS);
195 if (flex_gd->groups == NULL)
196 goto out2;
197
198 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
199 if (flex_gd->bg_flags == NULL)
200 goto out1;
201
202 return flex_gd;
203
204out1:
205 kfree(flex_gd->groups);
206out2:
207 kfree(flex_gd);
208out3:
209 return NULL;
210}
211
212static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
213{
214 kfree(flex_gd->bg_flags);
215 kfree(flex_gd->groups);
216 kfree(flex_gd);
217}
218
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500219/*
220 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
221 * and inode tables for a flex group.
222 *
223 * This function is used by 64bit-resize. Note that this function allocates
224 * group tables from the 1st group of groups contained by @flexgd, which may
225 * be a partial of a flex group.
226 *
227 * @sb: super block of fs to which the groups belongs
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400228 *
229 * Returns 0 on a successful allocation of the metadata blocks in the
230 * block group.
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500231 */
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400232static int ext4_alloc_group_tables(struct super_block *sb,
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500233 struct ext4_new_flex_group_data *flex_gd,
234 int flexbg_size)
235{
236 struct ext4_new_group_data *group_data = flex_gd->groups;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500237 ext4_fsblk_t start_blk;
238 ext4_fsblk_t last_blk;
239 ext4_group_t src_group;
240 ext4_group_t bb_index = 0;
241 ext4_group_t ib_index = 0;
242 ext4_group_t it_index = 0;
243 ext4_group_t group;
244 ext4_group_t last_group;
245 unsigned overhead;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500246 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500247
248 BUG_ON(flex_gd->count == 0 || group_data == NULL);
249
250 src_group = group_data[0].group;
251 last_group = src_group + flex_gd->count - 1;
252
253 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
254 (last_group & ~(flexbg_size - 1))));
255next_group:
256 group = group_data[0].group;
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400257 if (src_group >= group_data[0].group + flex_gd->count)
258 return -ENOSPC;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500259 start_blk = ext4_group_first_block_no(sb, src_group);
260 last_blk = start_blk + group_data[src_group - group].blocks_count;
261
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400262 overhead = ext4_group_overhead_blocks(sb, src_group);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500263
264 start_blk += overhead;
265
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500266 /* We collect contiguous blocks as much as possible. */
267 src_group++;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400268 for (; src_group <= last_group; src_group++) {
269 overhead = ext4_group_overhead_blocks(sb, src_group);
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500270 if (overhead == 0)
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500271 last_blk += group_data[src_group - group].blocks_count;
272 else
273 break;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400274 }
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500275
276 /* Allocate block bitmaps */
277 for (; bb_index < flex_gd->count; bb_index++) {
278 if (start_blk >= last_blk)
279 goto next_group;
280 group_data[bb_index].block_bitmap = start_blk++;
Lukas Czernerbd862982013-04-03 23:32:34 -0400281 group = ext4_get_group_number(sb, start_blk - 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500282 group -= group_data[0].group;
283 group_data[group].free_blocks_count--;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500284 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500285 }
286
287 /* Allocate inode bitmaps */
288 for (; ib_index < flex_gd->count; ib_index++) {
289 if (start_blk >= last_blk)
290 goto next_group;
291 group_data[ib_index].inode_bitmap = start_blk++;
Lukas Czernerbd862982013-04-03 23:32:34 -0400292 group = ext4_get_group_number(sb, start_blk - 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500293 group -= group_data[0].group;
294 group_data[group].free_blocks_count--;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500295 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500296 }
297
298 /* Allocate inode tables */
299 for (; it_index < flex_gd->count; it_index++) {
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500300 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
301 ext4_fsblk_t next_group_start;
302
303 if (start_blk + itb > last_blk)
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500304 goto next_group;
305 group_data[it_index].inode_table = start_blk;
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500306 group = ext4_get_group_number(sb, start_blk);
307 next_group_start = ext4_group_first_block_no(sb, group + 1);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500308 group -= group_data[0].group;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500309
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500310 if (start_blk + itb > next_group_start) {
311 flex_gd->bg_flags[group + 1] &= uninit_mask;
312 overhead = start_blk + itb - next_group_start;
313 group_data[group + 1].free_blocks_count -= overhead;
314 itb -= overhead;
315 }
316
317 group_data[group].free_blocks_count -= itb;
318 flex_gd->bg_flags[group] &= uninit_mask;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500319 start_blk += EXT4_SB(sb)->s_itb_per_group;
320 }
321
322 if (test_opt(sb, DEBUG)) {
323 int i;
324 group = group_data[0].group;
325
326 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
327 "%d groups, flexbg size is %d:\n", flex_gd->count,
328 flexbg_size);
329
330 for (i = 0; i < flex_gd->count; i++) {
331 printk(KERN_DEBUG "adding %s group %u: %u "
332 "blocks (%d free)\n",
333 ext4_bg_has_super(sb, group + i) ? "normal" :
334 "no-super", group + i,
335 group_data[i].blocks_count,
336 group_data[i].free_blocks_count);
337 }
338 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400339 return 0;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500340}
341
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700342static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700343 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700344{
345 struct buffer_head *bh;
346 int err;
347
348 bh = sb_getblk(sb, blk);
Wang Shilongaebf0242013-01-12 16:28:47 -0500349 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500350 return ERR_PTR(-ENOMEM);
liang xie5d601252014-05-12 22:06:43 -0400351 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -0700352 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700353 brelse(bh);
354 bh = ERR_PTR(err);
355 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700356 memset(bh->b_data, 0, sb->s_blocksize);
357 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700358 }
359
360 return bh;
361}
362
363/*
Eric Sandeen14904102007-10-16 18:38:25 -0400364 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
365 * If that fails, restart the transaction & regain write access for the
366 * buffer head which is used for block_bitmap modifications.
367 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400368static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400369{
370 int err;
371
Frank Mayhar03901312009-01-07 00:06:22 -0500372 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400373 return 0;
374
375 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
376 if (err < 0)
377 return err;
378 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400379 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
380 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400381 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400382 }
Eric Sandeen14904102007-10-16 18:38:25 -0400383
384 return 0;
385}
386
387/*
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500388 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
389 *
390 * Helper function for ext4_setup_new_group_blocks() which set .
391 *
392 * @sb: super block
393 * @handle: journal handle
394 * @flex_gd: flex group data
395 */
396static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
397 struct ext4_new_flex_group_data *flex_gd,
398 ext4_fsblk_t block, ext4_group_t count)
399{
400 ext4_group_t count2;
401
402 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
403 for (count2 = count; count > 0; count -= count2, block += count2) {
404 ext4_fsblk_t start;
405 struct buffer_head *bh;
406 ext4_group_t group;
407 int err;
408
Lukas Czernerbd862982013-04-03 23:32:34 -0400409 group = ext4_get_group_number(sb, block);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500410 start = ext4_group_first_block_no(sb, group);
411 group -= flex_gd->groups[0].group;
412
Theodore Ts'o3d2660d2014-02-15 22:42:25 -0500413 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500414 if (count2 > count)
415 count2 = count;
416
417 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
418 BUG_ON(flex_gd->count > 1);
419 continue;
420 }
421
422 err = extend_or_restart_transaction(handle, 1);
423 if (err)
424 return err;
425
426 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
Wang Shilongaebf0242013-01-12 16:28:47 -0500427 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500428 return -ENOMEM;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500429
liang xie5d601252014-05-12 22:06:43 -0400430 BUFFER_TRACE(bh, "get_write_access");
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500431 err = ext4_journal_get_write_access(handle, bh);
432 if (err)
433 return err;
434 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
435 block - start, count2);
436 ext4_set_bits(bh->b_data, block - start, count2);
437
438 err = ext4_handle_dirty_metadata(handle, NULL, bh);
439 if (unlikely(err))
440 return err;
441 brelse(bh);
442 }
443
444 return 0;
445}
446
447/*
448 * Set up the block and inode bitmaps, and the inode table for the new groups.
449 * This doesn't need to be part of the main transaction, since we are only
450 * changing blocks outside the actual filesystem. We still do journaling to
451 * ensure the recovery is correct in case of a failure just after resize.
452 * If any part of this fails, we simply abort the resize.
453 *
454 * setup_new_flex_group_blocks handles a flex group as follow:
455 * 1. copy super block and GDT, and initialize group tables if necessary.
456 * In this step, we only set bits in blocks bitmaps for blocks taken by
457 * super block and GDT.
458 * 2. allocate group tables in block bitmaps, that is, set bits in block
459 * bitmap for blocks taken by group tables.
460 */
461static int setup_new_flex_group_blocks(struct super_block *sb,
462 struct ext4_new_flex_group_data *flex_gd)
463{
464 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
465 ext4_fsblk_t start;
466 ext4_fsblk_t block;
467 struct ext4_sb_info *sbi = EXT4_SB(sb);
468 struct ext4_super_block *es = sbi->s_es;
469 struct ext4_new_group_data *group_data = flex_gd->groups;
470 __u16 *bg_flags = flex_gd->bg_flags;
471 handle_t *handle;
472 ext4_group_t group, count;
473 struct buffer_head *bh = NULL;
474 int reserved_gdb, i, j, err = 0, err2;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400475 int meta_bg;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500476
477 BUG_ON(!flex_gd->count || !group_data ||
478 group_data[0].group != sbi->s_groups_count);
479
480 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400481 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500482
483 /* This transaction may be extended/restarted along the way */
Theodore Ts'o9924a922013-02-08 21:59:22 -0500484 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500485 if (IS_ERR(handle))
486 return PTR_ERR(handle);
487
488 group = group_data[0].group;
489 for (i = 0; i < flex_gd->count; i++, group++) {
490 unsigned long gdblocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400491 ext4_grpblk_t overhead;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500492
493 gdblocks = ext4_bg_num_gdb(sb, group);
494 start = ext4_group_first_block_no(sb, group);
495
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400496 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400497 goto handle_itb;
498
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400499 if (meta_bg == 1) {
500 ext4_group_t first_group;
501 first_group = ext4_meta_bg_first_group(sb, group);
502 if (first_group != group + 1 &&
503 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
504 goto handle_itb;
505 }
506
507 block = start + ext4_bg_has_super(sb, group);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500508 /* Copy all of the GDT blocks into the backup in this group */
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400509 for (j = 0; j < gdblocks; j++, block++) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500510 struct buffer_head *gdb;
511
512 ext4_debug("update backup group %#04llx\n", block);
513 err = extend_or_restart_transaction(handle, 1);
514 if (err)
515 goto out;
516
517 gdb = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -0500518 if (unlikely(!gdb)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500519 err = -ENOMEM;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500520 goto out;
521 }
522
liang xie5d601252014-05-12 22:06:43 -0400523 BUFFER_TRACE(gdb, "get_write_access");
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500524 err = ext4_journal_get_write_access(handle, gdb);
525 if (err) {
526 brelse(gdb);
527 goto out;
528 }
529 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
530 gdb->b_size);
531 set_buffer_uptodate(gdb);
532
533 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
534 if (unlikely(err)) {
535 brelse(gdb);
536 goto out;
537 }
538 brelse(gdb);
539 }
540
541 /* Zero out all of the reserved backup group descriptor
542 * table blocks
543 */
544 if (ext4_bg_has_super(sb, group)) {
545 err = sb_issue_zeroout(sb, gdblocks + start + 1,
546 reserved_gdb, GFP_NOFS);
547 if (err)
548 goto out;
549 }
550
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400551handle_itb:
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500552 /* Initialize group tables of the grop @group */
553 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
554 goto handle_bb;
555
556 /* Zero out all of the inode table blocks */
557 block = group_data[i].inode_table;
558 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
559 block, sbi->s_itb_per_group);
560 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
561 GFP_NOFS);
562 if (err)
563 goto out;
564
565handle_bb:
566 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
567 goto handle_ib;
568
569 /* Initialize block bitmap of the @group */
570 block = group_data[i].block_bitmap;
571 err = extend_or_restart_transaction(handle, 1);
572 if (err)
573 goto out;
574
575 bh = bclean(handle, sb, block);
576 if (IS_ERR(bh)) {
577 err = PTR_ERR(bh);
Theodore Ts'oa9cfcd62014-09-03 09:33:00 -0400578 bh = NULL;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500579 goto out;
580 }
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400581 overhead = ext4_group_overhead_blocks(sb, group);
582 if (overhead != 0) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500583 ext4_debug("mark backup superblock %#04llx (+0)\n",
584 start);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400585 ext4_set_bits(bh->b_data, 0, overhead);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500586 }
587 ext4_mark_bitmap_end(group_data[i].blocks_count,
588 sb->s_blocksize * 8, bh->b_data);
589 err = ext4_handle_dirty_metadata(handle, NULL, bh);
590 if (err)
591 goto out;
592 brelse(bh);
593
594handle_ib:
595 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
596 continue;
597
598 /* Initialize inode bitmap of the @group */
599 block = group_data[i].inode_bitmap;
600 err = extend_or_restart_transaction(handle, 1);
601 if (err)
602 goto out;
603 /* Mark unused entries in inode bitmap used */
604 bh = bclean(handle, sb, block);
605 if (IS_ERR(bh)) {
606 err = PTR_ERR(bh);
Theodore Ts'oa9cfcd62014-09-03 09:33:00 -0400607 bh = NULL;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500608 goto out;
609 }
610
611 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
612 sb->s_blocksize * 8, bh->b_data);
613 err = ext4_handle_dirty_metadata(handle, NULL, bh);
614 if (err)
615 goto out;
616 brelse(bh);
617 }
618 bh = NULL;
619
620 /* Mark group tables in block bitmap */
621 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
622 count = group_table_count[j];
623 start = (&group_data[0].block_bitmap)[j];
624 block = start;
625 for (i = 1; i < flex_gd->count; i++) {
626 block += group_table_count[j];
627 if (block == (&group_data[i].block_bitmap)[j]) {
628 count += group_table_count[j];
629 continue;
630 }
631 err = set_flexbg_block_bitmap(sb, handle,
632 flex_gd, start, count);
633 if (err)
634 goto out;
635 count = group_table_count[j];
Theodore Ts'ob93c9532014-02-15 21:33:13 -0500636 start = (&group_data[i].block_bitmap)[j];
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500637 block = start;
638 }
639
640 if (count) {
641 err = set_flexbg_block_bitmap(sb, handle,
642 flex_gd, start, count);
643 if (err)
644 goto out;
645 }
646 }
647
648out:
649 brelse(bh);
650 err2 = ext4_journal_stop(handle);
651 if (err2 && !err)
652 err = err2;
653
654 return err;
655}
656
657/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700658 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700659 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 * calling this for the first time. In a sparse filesystem it will be the
661 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
662 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
663 */
Mingming Cao617ba132006-10-11 01:20:53 -0700664static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700665 unsigned *five, unsigned *seven)
666{
667 unsigned *min = three;
668 int mult = 3;
669 unsigned ret;
670
Mingming Cao617ba132006-10-11 01:20:53 -0700671 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
672 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700673 ret = *min;
674 *min += 1;
675 return ret;
676 }
677
678 if (*five < *min) {
679 min = five;
680 mult = 5;
681 }
682 if (*seven < *min) {
683 min = seven;
684 mult = 7;
685 }
686
687 ret = *min;
688 *min *= mult;
689
690 return ret;
691}
692
693/*
694 * Check that all of the backup GDT blocks are held in the primary GDT block.
695 * It is assumed that they are stored in group order. Returns the number of
696 * groups in current filesystem that have BACKUPS, or -ve error code.
697 */
698static int verify_reserved_gdb(struct super_block *sb,
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500699 ext4_group_t end,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 struct buffer_head *primary)
701{
Mingming Cao617ba132006-10-11 01:20:53 -0700702 const ext4_fsblk_t blk = primary->b_blocknr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700703 unsigned three = 1;
704 unsigned five = 5;
705 unsigned seven = 7;
706 unsigned grp;
707 __le32 *p = (__le32 *)primary->b_data;
708 int gdbackups = 0;
709
Mingming Cao617ba132006-10-11 01:20:53 -0700710 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700711 if (le32_to_cpu(*p++) !=
712 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500713 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700714 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700716 grp *
717 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
718 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 return -EINVAL;
720 }
Mingming Cao617ba132006-10-11 01:20:53 -0700721 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700722 return -EFBIG;
723 }
724
725 return gdbackups;
726}
727
728/*
729 * Called when we need to bring a reserved group descriptor table block into
730 * use from the resize inode. The primary copy of the new GDT block currently
731 * is an indirect block (under the double indirect block in the resize inode).
732 * The new backup GDT blocks will be stored as leaf blocks in this indirect
733 * block, in group order. Even though we know all the block numbers we need,
734 * we check to ensure that the resize inode has actually reserved these blocks.
735 *
736 * Don't need to update the block bitmaps because the blocks are still in use.
737 *
738 * We get all of the error cases out of the way, so that we are sure to not
739 * fail once we start modifying the data on disk, because JBD has no rollback.
740 */
741static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400742 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700743{
744 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700745 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400746 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700747 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700748 struct buffer_head **o_group_desc, **n_group_desc;
749 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400750 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700752 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700753 __le32 *data;
754 int err;
755
756 if (test_opt(sb, DEBUG))
757 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700758 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759 gdb_num);
760
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400761 /*
762 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400763 * because the user tools have no way of handling this. Probably a
764 * bad time to do it anyways.
765 */
Mingming Cao617ba132006-10-11 01:20:53 -0700766 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
767 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500768 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700769 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700770 return -EPERM;
771 }
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 Monakhov9aa5d32b2014-10-13 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
1435 * block. If we are adding a group past the last current GDT block,
1436 * 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 */
1440 credit = flex_gd->count * 4 + reserved_gdb;
Theodore Ts'o9924a922013-02-08 21:59:22 -05001441 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001442 if (IS_ERR(handle)) {
1443 err = PTR_ERR(handle);
1444 goto exit;
1445 }
1446
liang xie5d601252014-05-12 22:06:43 -04001447 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001448 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1449 if (err)
1450 goto exit_journal;
1451
1452 group = flex_gd->groups[0].group;
1453 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1454 err = ext4_add_new_descs(handle, sb, group,
1455 resize_inode, flex_gd->count);
1456 if (err)
1457 goto exit_journal;
1458
1459 err = ext4_setup_new_descs(handle, sb, flex_gd);
1460 if (err)
1461 goto exit_journal;
1462
1463 ext4_update_super(sb, flex_gd);
1464
1465 err = ext4_handle_dirty_super(handle, sb);
1466
1467exit_journal:
1468 err2 = ext4_journal_stop(handle);
1469 if (!err)
1470 err = err2;
1471
1472 if (!err) {
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001473 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1474 int gdb_num_end = ((group + flex_gd->count - 1) /
1475 EXT4_DESC_PER_BLOCK(sb));
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001476 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1477 EXT4_FEATURE_INCOMPAT_META_BG);
Tao Ma0acdb882012-09-26 00:08:57 -04001478 sector_t old_gdb = 0;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001479
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001480 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001481 sizeof(struct ext4_super_block), 0);
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001482 for (; gdb_num <= gdb_num_end; gdb_num++) {
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001483 struct buffer_head *gdb_bh;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001484
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001485 gdb_bh = sbi->s_group_desc[gdb_num];
Tao Ma0acdb882012-09-26 00:08:57 -04001486 if (old_gdb == gdb_bh->b_blocknr)
1487 continue;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001488 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001489 gdb_bh->b_size, meta_bg);
Tao Ma0acdb882012-09-26 00:08:57 -04001490 old_gdb = gdb_bh->b_blocknr;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001491 }
1492 }
1493exit:
1494 return err;
1495}
1496
Yongqiang Yang19c52462012-01-04 17:09:44 -05001497static int ext4_setup_next_flex_gd(struct super_block *sb,
1498 struct ext4_new_flex_group_data *flex_gd,
1499 ext4_fsblk_t n_blocks_count,
1500 unsigned long flexbg_size)
1501{
1502 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1503 struct ext4_new_group_data *group_data = flex_gd->groups;
1504 ext4_fsblk_t o_blocks_count;
1505 ext4_group_t n_group;
1506 ext4_group_t group;
1507 ext4_group_t last_group;
1508 ext4_grpblk_t last;
1509 ext4_grpblk_t blocks_per_group;
1510 unsigned long i;
1511
1512 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1513
1514 o_blocks_count = ext4_blocks_count(es);
1515
1516 if (o_blocks_count == n_blocks_count)
1517 return 0;
1518
1519 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1520 BUG_ON(last);
1521 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1522
1523 last_group = group | (flexbg_size - 1);
1524 if (last_group > n_group)
1525 last_group = n_group;
1526
1527 flex_gd->count = last_group - group + 1;
1528
1529 for (i = 0; i < flex_gd->count; i++) {
1530 int overhead;
1531
1532 group_data[i].group = group + i;
1533 group_data[i].blocks_count = blocks_per_group;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001534 overhead = ext4_group_overhead_blocks(sb, group + i);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001535 group_data[i].free_blocks_count = blocks_per_group - overhead;
Theodore Ts'o7f511862013-01-13 08:41:45 -05001536 if (ext4_has_group_desc_csum(sb)) {
Yongqiang Yang19c52462012-01-04 17:09:44 -05001537 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1538 EXT4_BG_INODE_UNINIT;
Theodore Ts'o7f511862013-01-13 08:41:45 -05001539 if (!test_opt(sb, INIT_INODE_TABLE))
1540 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1541 } else
Yongqiang Yang19c52462012-01-04 17:09:44 -05001542 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1543 }
1544
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001545 if (last_group == n_group && ext4_has_group_desc_csum(sb))
Yongqiang Yang19c52462012-01-04 17:09:44 -05001546 /* We need to initialize block bitmap of last group. */
1547 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1548
1549 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1550 group_data[i - 1].blocks_count = last + 1;
1551 group_data[i - 1].free_blocks_count -= blocks_per_group-
1552 last - 1;
1553 }
1554
1555 return 1;
1556}
1557
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001558/* Add group descriptor data to an existing or new group descriptor block.
1559 * Ensure we handle all possible error conditions _before_ we start modifying
1560 * the filesystem, because we cannot abort the transaction and not have it
1561 * write the data to disk.
1562 *
1563 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1564 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1565 *
1566 * We only need to hold the superblock lock while we are actually adding
1567 * in the new group's counts to the superblock. Prior to that we have
1568 * not really "added" the group at all. We re-check that we are still
1569 * adding in the last group in case things have changed since verifying.
1570 */
Mingming Cao617ba132006-10-11 01:20:53 -07001571int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001572{
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001573 struct ext4_new_flex_group_data flex_gd;
Mingming Cao617ba132006-10-11 01:20:53 -07001574 struct ext4_sb_info *sbi = EXT4_SB(sb);
1575 struct ext4_super_block *es = sbi->s_es;
1576 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001578 struct inode *inode = NULL;
Jon Ernst03b40e32013-06-17 08:56:26 -04001579 int gdb_off;
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001580 int err;
1581 __u16 bg_flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001582
Mingming Cao617ba132006-10-11 01:20:53 -07001583 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584
Mingming Cao617ba132006-10-11 01:20:53 -07001585 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1586 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001587 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001588 return -EPERM;
1589 }
1590
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001591 if (ext4_blocks_count(es) + input->blocks_count <
1592 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001593 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001594 return -EINVAL;
1595 }
1596
Mingming Cao617ba132006-10-11 01:20:53 -07001597 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001598 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001599 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001600 return -EINVAL;
1601 }
1602
1603 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001604 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -04001605 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1606 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001607 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001608 "No reserved GDT blocks, can't resize");
1609 return -EPERM;
1610 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001611 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1612 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001613 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -08001614 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 }
1616 }
1617
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001618
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001619 err = verify_group_input(sb, input);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -04001620 if (err)
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001621 goto out;
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001622
Theodore Ts'o117fff12012-09-05 01:29:50 -04001623 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1624 if (err)
Theodore Ts'o7f511862013-01-13 08:41:45 -05001625 goto out;
Theodore Ts'o117fff12012-09-05 01:29:50 -04001626
Theodore Ts'o28623c22012-09-05 01:31:50 -04001627 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1628 if (err)
1629 goto out;
1630
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001631 flex_gd.count = 1;
1632 flex_gd.groups = input;
1633 flex_gd.bg_flags = &bg_flags;
1634 err = ext4_flex_group_add(sb, inode, &flex_gd);
1635out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001636 iput(inode);
1637 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001638} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001640/*
Yongqiang Yang18e31432012-01-03 23:18:50 -05001641 * extend a group without checking assuming that checking has been done.
1642 */
1643static int ext4_group_extend_no_check(struct super_block *sb,
1644 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1645{
1646 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1647 handle_t *handle;
1648 int err = 0, err2;
1649
1650 /* We will update the superblock, one block bitmap, and
1651 * one group descriptor via ext4_group_add_blocks().
1652 */
Theodore Ts'o9924a922013-02-08 21:59:22 -05001653 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001654 if (IS_ERR(handle)) {
1655 err = PTR_ERR(handle);
1656 ext4_warning(sb, "error %d on journal start", err);
1657 return err;
1658 }
1659
liang xie5d601252014-05-12 22:06:43 -04001660 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
Yongqiang Yang18e31432012-01-03 23:18:50 -05001661 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1662 if (err) {
1663 ext4_warning(sb, "error %d on journal write access", err);
1664 goto errout;
1665 }
1666
1667 ext4_blocks_count_set(es, o_blocks_count + add);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001668 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001669 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1670 o_blocks_count + add);
1671 /* We add the blocks to the bitmap and set the group need init bit */
1672 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1673 if (err)
1674 goto errout;
1675 ext4_handle_dirty_super(handle, sb);
1676 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1677 o_blocks_count + add);
1678errout:
1679 err2 = ext4_journal_stop(handle);
1680 if (err2 && !err)
1681 err = err2;
1682
1683 if (!err) {
1684 if (test_opt(sb, DEBUG))
1685 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1686 "blocks\n", ext4_blocks_count(es));
Maarten ter Huurne6ca792e2013-07-01 08:12:08 -04001687 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001688 (char *)es, sizeof(struct ext4_super_block), 0);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001689 }
1690 return err;
1691}
1692
1693/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001694 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001695 * point is only used to extend the current filesystem to the end of the last
1696 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1697 * for emergencies (because it has no dependencies on reserved blocks).
1698 *
Mingming Cao617ba132006-10-11 01:20:53 -07001699 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001700 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1701 * GDT blocks are reserved to grow to the desired size.
1702 */
Mingming Cao617ba132006-10-11 01:20:53 -07001703int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1704 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705{
Mingming Cao617ba132006-10-11 01:20:53 -07001706 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001707 ext4_grpblk_t last;
1708 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001709 struct buffer_head *bh;
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001710 int err;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001711 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001712
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001713 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001714
1715 if (test_opt(sb, DEBUG))
Theodore Ts'o92b97812012-03-19 23:41:49 -04001716 ext4_msg(sb, KERN_DEBUG,
1717 "extending last group from %llu to %llu blocks",
1718 o_blocks_count, n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001719
1720 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1721 return 0;
1722
1723 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Theodore Ts'o92b97812012-03-19 23:41:49 -04001724 ext4_msg(sb, KERN_ERR,
1725 "filesystem too large to resize to %llu blocks safely",
1726 n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001728 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001729 return -EINVAL;
1730 }
1731
1732 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001733 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001734 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 }
1736
1737 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001738 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001739
1740 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001741 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742 return -EPERM;
1743 }
1744
Mingming Cao617ba132006-10-11 01:20:53 -07001745 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001746
1747 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001748 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001749 return -EINVAL;
1750 }
1751
1752 if (o_blocks_count + add > n_blocks_count)
1753 add = n_blocks_count - o_blocks_count;
1754
1755 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001756 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001757 o_blocks_count + add, add);
1758
1759 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001760 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001761 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001762 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001763 return -ENOSPC;
1764 }
1765 brelse(bh);
1766
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001767 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001768 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001769} /* ext4_group_extend */
Yongqiang Yang19c52462012-01-04 17:09:44 -05001770
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001771
1772static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1773{
1774 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1775}
1776
1777/*
1778 * Release the resize inode and drop the resize_inode feature if there
1779 * are no more reserved gdt blocks, and then convert the file system
1780 * to enable meta_bg
1781 */
1782static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1783{
1784 handle_t *handle;
1785 struct ext4_sb_info *sbi = EXT4_SB(sb);
1786 struct ext4_super_block *es = sbi->s_es;
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001787 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001788 ext4_fsblk_t nr;
1789 int i, ret, err = 0;
1790 int credits = 1;
1791
1792 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001793 if (inode) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001794 if (es->s_reserved_gdt_blocks) {
1795 ext4_error(sb, "Unexpected non-zero "
1796 "s_reserved_gdt_blocks");
1797 return -EPERM;
1798 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001799
1800 /* Do a quick sanity check of the resize inode */
1801 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1802 goto invalid_resize_inode;
1803 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1804 if (i == EXT4_DIND_BLOCK) {
1805 if (ei->i_data[i])
1806 continue;
1807 else
1808 goto invalid_resize_inode;
1809 }
1810 if (ei->i_data[i])
1811 goto invalid_resize_inode;
1812 }
1813 credits += 3; /* block bitmap, bg descriptor, resize inode */
1814 }
1815
Theodore Ts'o9924a922013-02-08 21:59:22 -05001816 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001817 if (IS_ERR(handle))
1818 return PTR_ERR(handle);
1819
liang xie5d601252014-05-12 22:06:43 -04001820 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001821 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1822 if (err)
1823 goto errout;
1824
1825 EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1826 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1827 sbi->s_es->s_first_meta_bg =
1828 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1829
1830 err = ext4_handle_dirty_super(handle, sb);
1831 if (err) {
1832 ext4_std_error(sb, err);
1833 goto errout;
1834 }
1835
1836 if (inode) {
1837 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1838 ext4_free_blocks(handle, inode, NULL, nr, 1,
1839 EXT4_FREE_BLOCKS_METADATA |
1840 EXT4_FREE_BLOCKS_FORGET);
1841 ei->i_data[EXT4_DIND_BLOCK] = 0;
1842 inode->i_blocks = 0;
1843
1844 err = ext4_mark_inode_dirty(handle, inode);
1845 if (err)
1846 ext4_std_error(sb, err);
1847 }
1848
1849errout:
1850 ret = ext4_journal_stop(handle);
1851 if (!err)
1852 err = ret;
1853 return ret;
1854
1855invalid_resize_inode:
1856 ext4_error(sb, "corrupted/inconsistent resize inode");
1857 return -EINVAL;
1858}
1859
Yongqiang Yang19c52462012-01-04 17:09:44 -05001860/*
1861 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1862 *
1863 * @sb: super block of the fs to be resized
1864 * @n_blocks_count: the number of blocks resides in the resized fs
1865 */
1866int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1867{
1868 struct ext4_new_flex_group_data *flex_gd = NULL;
1869 struct ext4_sb_info *sbi = EXT4_SB(sb);
1870 struct ext4_super_block *es = sbi->s_es;
1871 struct buffer_head *bh;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001872 struct inode *resize_inode = NULL;
1873 ext4_grpblk_t add, offset;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001874 unsigned long n_desc_blocks;
1875 unsigned long o_desc_blocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001876 ext4_group_t o_group;
1877 ext4_group_t n_group;
1878 ext4_fsblk_t o_blocks_count;
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001879 ext4_fsblk_t n_blocks_count_retry = 0;
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001880 unsigned long last_update_time = 0;
Theodore Ts'o117fff12012-09-05 01:29:50 -04001881 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001882 int meta_bg;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001883
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001884 /* See if the device is actually as big as what was requested */
1885 bh = sb_bread(sb, n_blocks_count - 1);
1886 if (!bh) {
1887 ext4_warning(sb, "can't read last block, resize aborted");
1888 return -ENOSPC;
1889 }
1890 brelse(bh);
1891
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001892retry:
Yongqiang Yang19c52462012-01-04 17:09:44 -05001893 o_blocks_count = ext4_blocks_count(es);
1894
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001895 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1896 "to %llu blocks", o_blocks_count, n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001897
1898 if (n_blocks_count < o_blocks_count) {
1899 /* On-line shrinking not supported */
1900 ext4_warning(sb, "can't shrink FS - resize aborted");
1901 return -EINVAL;
1902 }
1903
1904 if (n_blocks_count == o_blocks_count)
1905 /* Nothing need to do */
1906 return 0;
1907
Lukas Czernerbd862982013-04-03 23:32:34 -04001908 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
Theodore Ts'o3f8a6412013-04-21 22:56:32 -04001909 if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1910 ext4_warning(sb, "resize would cause inodes_count overflow");
1911 return -EINVAL;
1912 }
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001913 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001914
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001915 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1916 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001917
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001918 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001919
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001920 if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1921 if (meta_bg) {
1922 ext4_error(sb, "resize_inode and meta_bg enabled "
1923 "simultaneously");
1924 return -EINVAL;
1925 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001926 if (n_desc_blocks > o_desc_blocks +
1927 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1928 n_blocks_count_retry = n_blocks_count;
1929 n_desc_blocks = o_desc_blocks +
1930 le16_to_cpu(es->s_reserved_gdt_blocks);
1931 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1932 n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1933 n_group--; /* set to last group number */
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001934 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001935
1936 if (!resize_inode)
1937 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001938 if (IS_ERR(resize_inode)) {
1939 ext4_warning(sb, "Error opening resize inode");
1940 return PTR_ERR(resize_inode);
1941 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001942 }
1943
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001944 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001945 err = ext4_convert_meta_bg(sb, resize_inode);
1946 if (err)
1947 goto out;
1948 if (resize_inode) {
1949 iput(resize_inode);
1950 resize_inode = NULL;
1951 }
1952 if (n_blocks_count_retry) {
1953 n_blocks_count = n_blocks_count_retry;
1954 n_blocks_count_retry = 0;
1955 goto retry;
1956 }
Yongqiang Yang19c52462012-01-04 17:09:44 -05001957 }
1958
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001959 /* extend the last group */
1960 if (n_group == o_group)
1961 add = n_blocks_count - o_blocks_count;
1962 else
1963 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1964 if (add > 0) {
Yongqiang Yang19c52462012-01-04 17:09:44 -05001965 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1966 if (err)
1967 goto out;
1968 }
1969
Yongqiang Yangd7574ad2012-09-05 01:23:50 -04001970 if (ext4_blocks_count(es) == n_blocks_count)
Yongqiang Yang19c52462012-01-04 17:09:44 -05001971 goto out;
1972
Theodore Ts'o117fff12012-09-05 01:29:50 -04001973 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1974 if (err)
1975 return err;
1976
Theodore Ts'o28623c22012-09-05 01:31:50 -04001977 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1978 if (err)
1979 goto out;
1980
Yongqiang Yang19c52462012-01-04 17:09:44 -05001981 flex_gd = alloc_flex_gd(flexbg_size);
1982 if (flex_gd == NULL) {
1983 err = -ENOMEM;
1984 goto out;
1985 }
1986
1987 /* Add flex groups. Note that a regular group is a
1988 * flex group with 1 group.
1989 */
1990 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1991 flexbg_size)) {
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001992 if (jiffies - last_update_time > HZ * 10) {
1993 if (last_update_time)
1994 ext4_msg(sb, KERN_INFO,
1995 "resized to %llu blocks",
1996 ext4_blocks_count(es));
1997 last_update_time = jiffies;
1998 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -04001999 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2000 break;
Yongqiang Yang19c52462012-01-04 17:09:44 -05002001 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2002 if (unlikely(err))
2003 break;
2004 }
2005
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04002006 if (!err && n_blocks_count_retry) {
2007 n_blocks_count = n_blocks_count_retry;
2008 n_blocks_count_retry = 0;
2009 free_flex_gd(flex_gd);
2010 flex_gd = NULL;
2011 goto retry;
2012 }
2013
Yongqiang Yang19c52462012-01-04 17:09:44 -05002014out:
2015 if (flex_gd)
2016 free_flex_gd(flex_gd);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04002017 if (resize_inode != NULL)
2018 iput(resize_inode);
Theodore Ts'o59e31c12012-09-19 00:55:56 -04002019 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05002020 return err;
2021}