blob: 7a75e1086961eedb674b8c2bf1bb84e657f4dfa2 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/resize.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
Mingming Cao617ba132006-10-11 01:20:53 -07004 * Support for resizing an ext4 filesystem while it is mounted.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005 *
6 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7 *
8 * This could probably be made into a module, because it is not often in use.
9 */
10
11
Mingming Cao617ba132006-10-11 01:20:53 -070012#define EXT4FS_DEBUG
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013
Dave Kleikampac27a0e2006-10-11 01:20:50 -070014#include <linux/errno.h>
15#include <linux/slab.h>
16
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040017#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070018
Yongqiang Yang8f82f842011-07-26 21:35:44 -040019int ext4_resize_begin(struct super_block *sb)
20{
21 int ret = 0;
22
23 if (!capable(CAP_SYS_RESOURCE))
24 return -EPERM;
25
Yongqiang Yangce723c32011-07-26 21:39:09 -040026 /*
27 * We are not allowed to do online-resizing on a filesystem mounted
28 * with error, because it can destroy the filesystem easily.
29 */
30 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
31 ext4_warning(sb, "There are errors in the filesystem, "
32 "so online resizing is not allowed\n");
33 return -EPERM;
34 }
35
Yongqiang Yang8f82f842011-07-26 21:35:44 -040036 if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
37 ret = -EBUSY;
38
39 return ret;
40}
41
42void ext4_resize_end(struct super_block *sb)
43{
44 clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
45 smp_mb__after_clear_bit();
46}
47
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;
Yongqiang Yang01f795f2012-09-05 01:33:50 -040082 unsigned overhead = ext4_group_overhead_blocks(sb, group);
Mingming Cao617ba132006-10-11 01:20:53 -070083 ext4_fsblk_t metaend = start + overhead;
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
88 input->free_blocks_count = free_blocks_count =
89 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
90
91 if (test_opt(sb, DEBUG))
Mingming Cao617ba132006-10-11 01:20:53 -070092 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
Dave Kleikampac27a0e2006-10-11 01:20:50 -070093 "(%d free, %u reserved)\n",
Mingming Cao617ba132006-10-11 01:20:53 -070094 ext4_bg_has_super(sb, input->group) ? "normal" :
Dave Kleikampac27a0e2006-10-11 01:20:50 -070095 "no-super", input->group, input->blocks_count,
96 free_blocks_count, input->reserved_blocks);
97
Mingming Cao3a5b2ec2006-10-11 01:21:05 -070098 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070099 if (group != sbi->s_groups_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500100 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700101 input->group, sbi->s_groups_count);
Mingming Cao3a5b2ec2006-10-11 01:21:05 -0700102 else if (offset != 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500103 ext4_warning(sb, "Last group not full");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700104 else if (input->reserved_blocks > input->blocks_count / 5)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500105 ext4_warning(sb, "Reserved blocks too high (%u)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106 input->reserved_blocks);
107 else if (free_blocks_count < 0)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500108 ext4_warning(sb, "Bad blocks count %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700109 input->blocks_count);
110 else if (!(bh = sb_bread(sb, end - 1)))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500111 ext4_warning(sb, "Cannot read last block (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700112 end - 1);
113 else if (outside(input->block_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500114 ext4_warning(sb, "Block bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700115 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700116 else if (outside(input->inode_bitmap, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500117 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700118 (unsigned long long)input->inode_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700119 else if (outside(input->inode_table, start, end) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400120 outside(itend - 1, start, end))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500121 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700122 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700123 else if (input->inode_bitmap == input->block_bitmap)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500124 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700125 (unsigned long long)input->block_bitmap);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700126 else if (inside(input->block_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500127 ext4_warning(sb, "Block bitmap (%llu) in inode table "
128 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700129 (unsigned long long)input->block_bitmap,
130 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700131 else if (inside(input->inode_bitmap, input->inode_table, itend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500132 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
133 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700134 (unsigned long long)input->inode_bitmap,
135 (unsigned long long)input->inode_table, itend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700136 else if (inside(input->block_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500137 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700138 (unsigned long long)input->block_bitmap,
139 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700140 else if (inside(input->inode_bitmap, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500141 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700142 (unsigned long long)input->inode_bitmap,
143 start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700144 else if (inside(input->inode_table, start, metaend) ||
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400145 inside(itend - 1, start, metaend))
Eric Sandeen12062dd2010-02-15 14:19:27 -0500146 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
147 "(%llu-%llu)",
Randy Dunlap1939e492006-10-28 10:38:26 -0700148 (unsigned long long)input->inode_table,
149 itend - 1, start, metaend - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150 else
151 err = 0;
152 brelse(bh);
153
154 return err;
155}
156
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500157/*
158 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
159 * group each time.
160 */
161struct ext4_new_flex_group_data {
162 struct ext4_new_group_data *groups; /* new_group_data for groups
163 in the flex group */
164 __u16 *bg_flags; /* block group flags of groups
165 in @groups */
166 ext4_group_t count; /* number of groups in @groups
167 */
168};
169
170/*
171 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
172 * @flexbg_size.
173 *
174 * Returns NULL on failure otherwise address of the allocated structure.
175 */
176static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
177{
178 struct ext4_new_flex_group_data *flex_gd;
179
180 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
181 if (flex_gd == NULL)
182 goto out3;
183
Haogang Chen967ac8a2012-05-28 14:21:55 -0400184 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
185 goto out2;
Yongqiang Yang28c7bac2012-01-03 23:22:50 -0500186 flex_gd->count = flexbg_size;
187
188 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
189 flexbg_size, GFP_NOFS);
190 if (flex_gd->groups == NULL)
191 goto out2;
192
193 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
194 if (flex_gd->bg_flags == NULL)
195 goto out1;
196
197 return flex_gd;
198
199out1:
200 kfree(flex_gd->groups);
201out2:
202 kfree(flex_gd);
203out3:
204 return NULL;
205}
206
207static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
208{
209 kfree(flex_gd->bg_flags);
210 kfree(flex_gd->groups);
211 kfree(flex_gd);
212}
213
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500214/*
215 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
216 * and inode tables for a flex group.
217 *
218 * This function is used by 64bit-resize. Note that this function allocates
219 * group tables from the 1st group of groups contained by @flexgd, which may
220 * be a partial of a flex group.
221 *
222 * @sb: super block of fs to which the groups belongs
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400223 *
224 * Returns 0 on a successful allocation of the metadata blocks in the
225 * block group.
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500226 */
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400227static int ext4_alloc_group_tables(struct super_block *sb,
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500228 struct ext4_new_flex_group_data *flex_gd,
229 int flexbg_size)
230{
231 struct ext4_new_group_data *group_data = flex_gd->groups;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500232 ext4_fsblk_t start_blk;
233 ext4_fsblk_t last_blk;
234 ext4_group_t src_group;
235 ext4_group_t bb_index = 0;
236 ext4_group_t ib_index = 0;
237 ext4_group_t it_index = 0;
238 ext4_group_t group;
239 ext4_group_t last_group;
240 unsigned overhead;
241
242 BUG_ON(flex_gd->count == 0 || group_data == NULL);
243
244 src_group = group_data[0].group;
245 last_group = src_group + flex_gd->count - 1;
246
247 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
248 (last_group & ~(flexbg_size - 1))));
249next_group:
250 group = group_data[0].group;
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400251 if (src_group >= group_data[0].group + flex_gd->count)
252 return -ENOSPC;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500253 start_blk = ext4_group_first_block_no(sb, src_group);
254 last_blk = start_blk + group_data[src_group - group].blocks_count;
255
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400256 overhead = ext4_group_overhead_blocks(sb, src_group);
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500257
258 start_blk += overhead;
259
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500260 /* We collect contiguous blocks as much as possible. */
261 src_group++;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400262 for (; src_group <= last_group; src_group++) {
263 overhead = ext4_group_overhead_blocks(sb, src_group);
264 if (overhead != 0)
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500265 last_blk += group_data[src_group - group].blocks_count;
266 else
267 break;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400268 }
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500269
270 /* Allocate block bitmaps */
271 for (; bb_index < flex_gd->count; bb_index++) {
272 if (start_blk >= last_blk)
273 goto next_group;
274 group_data[bb_index].block_bitmap = start_blk++;
275 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
276 group -= group_data[0].group;
277 group_data[group].free_blocks_count--;
278 if (flexbg_size > 1)
279 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
280 }
281
282 /* Allocate inode bitmaps */
283 for (; ib_index < flex_gd->count; ib_index++) {
284 if (start_blk >= last_blk)
285 goto next_group;
286 group_data[ib_index].inode_bitmap = start_blk++;
287 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
288 group -= group_data[0].group;
289 group_data[group].free_blocks_count--;
290 if (flexbg_size > 1)
291 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
292 }
293
294 /* Allocate inode tables */
295 for (; it_index < flex_gd->count; it_index++) {
296 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
297 goto next_group;
298 group_data[it_index].inode_table = start_blk;
299 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
300 group -= group_data[0].group;
301 group_data[group].free_blocks_count -=
302 EXT4_SB(sb)->s_itb_per_group;
303 if (flexbg_size > 1)
304 flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
305
306 start_blk += EXT4_SB(sb)->s_itb_per_group;
307 }
308
309 if (test_opt(sb, DEBUG)) {
310 int i;
311 group = group_data[0].group;
312
313 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
314 "%d groups, flexbg size is %d:\n", flex_gd->count,
315 flexbg_size);
316
317 for (i = 0; i < flex_gd->count; i++) {
318 printk(KERN_DEBUG "adding %s group %u: %u "
319 "blocks (%d free)\n",
320 ext4_bg_has_super(sb, group + i) ? "normal" :
321 "no-super", group + i,
322 group_data[i].blocks_count,
323 group_data[i].free_blocks_count);
324 }
325 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -0400326 return 0;
Yongqiang Yang3fbea4b2012-01-03 23:44:38 -0500327}
328
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700329static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
Mingming Cao617ba132006-10-11 01:20:53 -0700330 ext4_fsblk_t blk)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700331{
332 struct buffer_head *bh;
333 int err;
334
335 bh = sb_getblk(sb, blk);
336 if (!bh)
337 return ERR_PTR(-EIO);
Mingming Cao617ba132006-10-11 01:20:53 -0700338 if ((err = ext4_journal_get_write_access(handle, bh))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700339 brelse(bh);
340 bh = ERR_PTR(err);
341 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700342 memset(bh->b_data, 0, sb->s_blocksize);
343 set_buffer_uptodate(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700344 }
345
346 return bh;
347}
348
349/*
Eric Sandeen14904102007-10-16 18:38:25 -0400350 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
351 * If that fails, restart the transaction & regain write access for the
352 * buffer head which is used for block_bitmap modifications.
353 */
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400354static int extend_or_restart_transaction(handle_t *handle, int thresh)
Eric Sandeen14904102007-10-16 18:38:25 -0400355{
356 int err;
357
Frank Mayhar03901312009-01-07 00:06:22 -0500358 if (ext4_handle_has_enough_credits(handle, thresh))
Eric Sandeen14904102007-10-16 18:38:25 -0400359 return 0;
360
361 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
362 if (err < 0)
363 return err;
364 if (err) {
Yongqiang Yang6d40bc52011-07-26 22:24:41 -0400365 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
366 if (err)
Eric Sandeen14904102007-10-16 18:38:25 -0400367 return err;
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400368 }
Eric Sandeen14904102007-10-16 18:38:25 -0400369
370 return 0;
371}
372
373/*
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500374 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
375 *
376 * Helper function for ext4_setup_new_group_blocks() which set .
377 *
378 * @sb: super block
379 * @handle: journal handle
380 * @flex_gd: flex group data
381 */
382static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
383 struct ext4_new_flex_group_data *flex_gd,
384 ext4_fsblk_t block, ext4_group_t count)
385{
386 ext4_group_t count2;
387
388 ext4_debug("mark blocks [%llu/%u] used\n", block, count);
389 for (count2 = count; count > 0; count -= count2, block += count2) {
390 ext4_fsblk_t start;
391 struct buffer_head *bh;
392 ext4_group_t group;
393 int err;
394
395 ext4_get_group_no_and_offset(sb, block, &group, NULL);
396 start = ext4_group_first_block_no(sb, group);
397 group -= flex_gd->groups[0].group;
398
399 count2 = sb->s_blocksize * 8 - (block - start);
400 if (count2 > count)
401 count2 = count;
402
403 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
404 BUG_ON(flex_gd->count > 1);
405 continue;
406 }
407
408 err = extend_or_restart_transaction(handle, 1);
409 if (err)
410 return err;
411
412 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
413 if (!bh)
414 return -EIO;
415
416 err = ext4_journal_get_write_access(handle, bh);
417 if (err)
418 return err;
419 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
420 block - start, count2);
421 ext4_set_bits(bh->b_data, block - start, count2);
422
423 err = ext4_handle_dirty_metadata(handle, NULL, bh);
424 if (unlikely(err))
425 return err;
426 brelse(bh);
427 }
428
429 return 0;
430}
431
432/*
433 * Set up the block and inode bitmaps, and the inode table for the new groups.
434 * This doesn't need to be part of the main transaction, since we are only
435 * changing blocks outside the actual filesystem. We still do journaling to
436 * ensure the recovery is correct in case of a failure just after resize.
437 * If any part of this fails, we simply abort the resize.
438 *
439 * setup_new_flex_group_blocks handles a flex group as follow:
440 * 1. copy super block and GDT, and initialize group tables if necessary.
441 * In this step, we only set bits in blocks bitmaps for blocks taken by
442 * super block and GDT.
443 * 2. allocate group tables in block bitmaps, that is, set bits in block
444 * bitmap for blocks taken by group tables.
445 */
446static int setup_new_flex_group_blocks(struct super_block *sb,
447 struct ext4_new_flex_group_data *flex_gd)
448{
449 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
450 ext4_fsblk_t start;
451 ext4_fsblk_t block;
452 struct ext4_sb_info *sbi = EXT4_SB(sb);
453 struct ext4_super_block *es = sbi->s_es;
454 struct ext4_new_group_data *group_data = flex_gd->groups;
455 __u16 *bg_flags = flex_gd->bg_flags;
456 handle_t *handle;
457 ext4_group_t group, count;
458 struct buffer_head *bh = NULL;
459 int reserved_gdb, i, j, err = 0, err2;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400460 int meta_bg;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500461
462 BUG_ON(!flex_gd->count || !group_data ||
463 group_data[0].group != sbi->s_groups_count);
464
465 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400466 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500467
468 /* This transaction may be extended/restarted along the way */
469 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
470 if (IS_ERR(handle))
471 return PTR_ERR(handle);
472
473 group = group_data[0].group;
474 for (i = 0; i < flex_gd->count; i++, group++) {
475 unsigned long gdblocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400476 ext4_grpblk_t overhead;
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500477
478 gdblocks = ext4_bg_num_gdb(sb, group);
479 start = ext4_group_first_block_no(sb, group);
480
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400481 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400482 goto handle_itb;
483
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400484 if (meta_bg == 1) {
485 ext4_group_t first_group;
486 first_group = ext4_meta_bg_first_group(sb, group);
487 if (first_group != group + 1 &&
488 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
489 goto handle_itb;
490 }
491
492 block = start + ext4_bg_has_super(sb, group);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500493 /* Copy all of the GDT blocks into the backup in this group */
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400494 for (j = 0; j < gdblocks; j++, block++) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500495 struct buffer_head *gdb;
496
497 ext4_debug("update backup group %#04llx\n", block);
498 err = extend_or_restart_transaction(handle, 1);
499 if (err)
500 goto out;
501
502 gdb = sb_getblk(sb, block);
503 if (!gdb) {
504 err = -EIO;
505 goto out;
506 }
507
508 err = ext4_journal_get_write_access(handle, gdb);
509 if (err) {
510 brelse(gdb);
511 goto out;
512 }
513 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
514 gdb->b_size);
515 set_buffer_uptodate(gdb);
516
517 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
518 if (unlikely(err)) {
519 brelse(gdb);
520 goto out;
521 }
522 brelse(gdb);
523 }
524
525 /* Zero out all of the reserved backup group descriptor
526 * table blocks
527 */
528 if (ext4_bg_has_super(sb, group)) {
529 err = sb_issue_zeroout(sb, gdblocks + start + 1,
530 reserved_gdb, GFP_NOFS);
531 if (err)
532 goto out;
533 }
534
Yongqiang Yang6df935a2012-09-05 01:25:50 -0400535handle_itb:
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500536 /* Initialize group tables of the grop @group */
537 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
538 goto handle_bb;
539
540 /* Zero out all of the inode table blocks */
541 block = group_data[i].inode_table;
542 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
543 block, sbi->s_itb_per_group);
544 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
545 GFP_NOFS);
546 if (err)
547 goto out;
548
549handle_bb:
550 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
551 goto handle_ib;
552
553 /* Initialize block bitmap of the @group */
554 block = group_data[i].block_bitmap;
555 err = extend_or_restart_transaction(handle, 1);
556 if (err)
557 goto out;
558
559 bh = bclean(handle, sb, block);
560 if (IS_ERR(bh)) {
561 err = PTR_ERR(bh);
562 goto out;
563 }
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400564 overhead = ext4_group_overhead_blocks(sb, group);
565 if (overhead != 0) {
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500566 ext4_debug("mark backup superblock %#04llx (+0)\n",
567 start);
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400568 ext4_set_bits(bh->b_data, 0, overhead);
Yongqiang Yang33afdcc2012-01-03 23:32:52 -0500569 }
570 ext4_mark_bitmap_end(group_data[i].blocks_count,
571 sb->s_blocksize * 8, bh->b_data);
572 err = ext4_handle_dirty_metadata(handle, NULL, bh);
573 if (err)
574 goto out;
575 brelse(bh);
576
577handle_ib:
578 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
579 continue;
580
581 /* Initialize inode bitmap of the @group */
582 block = group_data[i].inode_bitmap;
583 err = extend_or_restart_transaction(handle, 1);
584 if (err)
585 goto out;
586 /* Mark unused entries in inode bitmap used */
587 bh = bclean(handle, sb, block);
588 if (IS_ERR(bh)) {
589 err = PTR_ERR(bh);
590 goto out;
591 }
592
593 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
594 sb->s_blocksize * 8, bh->b_data);
595 err = ext4_handle_dirty_metadata(handle, NULL, bh);
596 if (err)
597 goto out;
598 brelse(bh);
599 }
600 bh = NULL;
601
602 /* Mark group tables in block bitmap */
603 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
604 count = group_table_count[j];
605 start = (&group_data[0].block_bitmap)[j];
606 block = start;
607 for (i = 1; i < flex_gd->count; i++) {
608 block += group_table_count[j];
609 if (block == (&group_data[i].block_bitmap)[j]) {
610 count += group_table_count[j];
611 continue;
612 }
613 err = set_flexbg_block_bitmap(sb, handle,
614 flex_gd, start, count);
615 if (err)
616 goto out;
617 count = group_table_count[j];
618 start = group_data[i].block_bitmap;
619 block = start;
620 }
621
622 if (count) {
623 err = set_flexbg_block_bitmap(sb, handle,
624 flex_gd, start, count);
625 if (err)
626 goto out;
627 }
628 }
629
630out:
631 brelse(bh);
632 err2 = ext4_journal_stop(handle);
633 if (err2 && !err)
634 err = err2;
635
636 return err;
637}
638
639/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700640 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
Mingming Cao617ba132006-10-11 01:20:53 -0700641 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 * calling this for the first time. In a sparse filesystem it will be the
643 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
644 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
645 */
Mingming Cao617ba132006-10-11 01:20:53 -0700646static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700647 unsigned *five, unsigned *seven)
648{
649 unsigned *min = three;
650 int mult = 3;
651 unsigned ret;
652
Mingming Cao617ba132006-10-11 01:20:53 -0700653 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
654 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700655 ret = *min;
656 *min += 1;
657 return ret;
658 }
659
660 if (*five < *min) {
661 min = five;
662 mult = 5;
663 }
664 if (*seven < *min) {
665 min = seven;
666 mult = 7;
667 }
668
669 ret = *min;
670 *min *= mult;
671
672 return ret;
673}
674
675/*
676 * Check that all of the backup GDT blocks are held in the primary GDT block.
677 * It is assumed that they are stored in group order. Returns the number of
678 * groups in current filesystem that have BACKUPS, or -ve error code.
679 */
680static int verify_reserved_gdb(struct super_block *sb,
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500681 ext4_group_t end,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700682 struct buffer_head *primary)
683{
Mingming Cao617ba132006-10-11 01:20:53 -0700684 const ext4_fsblk_t blk = primary->b_blocknr;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700685 unsigned three = 1;
686 unsigned five = 5;
687 unsigned seven = 7;
688 unsigned grp;
689 __le32 *p = (__le32 *)primary->b_data;
690 int gdbackups = 0;
691
Mingming Cao617ba132006-10-11 01:20:53 -0700692 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700693 if (le32_to_cpu(*p++) !=
694 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
Eric Sandeen12062dd2010-02-15 14:19:27 -0500695 ext4_warning(sb, "reserved GDT %llu"
Mingming Cao2ae02102006-10-11 01:21:11 -0700696 " missing grp %d (%llu)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700697 blk, grp,
Laurent Vivierbd81d8e2006-10-11 01:21:10 -0700698 grp *
699 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
700 blk);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700701 return -EINVAL;
702 }
Mingming Cao617ba132006-10-11 01:20:53 -0700703 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700704 return -EFBIG;
705 }
706
707 return gdbackups;
708}
709
710/*
711 * Called when we need to bring a reserved group descriptor table block into
712 * use from the resize inode. The primary copy of the new GDT block currently
713 * is an indirect block (under the double indirect block in the resize inode).
714 * The new backup GDT blocks will be stored as leaf blocks in this indirect
715 * block, in group order. Even though we know all the block numbers we need,
716 * we check to ensure that the resize inode has actually reserved these blocks.
717 *
718 * Don't need to update the block bitmaps because the blocks are still in use.
719 *
720 * We get all of the error cases out of the way, so that we are sure to not
721 * fail once we start modifying the data on disk, because JBD has no rollback.
722 */
723static int add_new_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang2f919712011-07-27 21:16:33 -0400724 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700725{
726 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700727 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400728 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
Mingming Cao617ba132006-10-11 01:20:53 -0700729 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700730 struct buffer_head **o_group_desc, **n_group_desc;
731 struct buffer_head *dind;
Yongqiang Yang2f919712011-07-27 21:16:33 -0400732 struct buffer_head *gdb_bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733 int gdbackups;
Mingming Cao617ba132006-10-11 01:20:53 -0700734 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700735 __le32 *data;
736 int err;
737
738 if (test_opt(sb, DEBUG))
739 printk(KERN_DEBUG
Mingming Cao617ba132006-10-11 01:20:53 -0700740 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700741 gdb_num);
742
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400743 /*
744 * If we are not using the primary superblock/GDT copy don't resize,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400745 * because the user tools have no way of handling this. Probably a
746 * bad time to do it anyways.
747 */
Mingming Cao617ba132006-10-11 01:20:53 -0700748 if (EXT4_SB(sb)->s_sbh->b_blocknr !=
749 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500750 ext4_warning(sb, "won't resize using backup superblock at %llu",
Mingming Cao617ba132006-10-11 01:20:53 -0700751 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752 return -EPERM;
753 }
754
Yongqiang Yang2f919712011-07-27 21:16:33 -0400755 gdb_bh = sb_bread(sb, gdblock);
756 if (!gdb_bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700757 return -EIO;
758
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500759 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400760 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700761 err = gdbackups;
762 goto exit_bh;
763 }
764
Mingming Cao617ba132006-10-11 01:20:53 -0700765 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700766 dind = sb_bread(sb, le32_to_cpu(*data));
767 if (!dind) {
768 err = -EIO;
769 goto exit_bh;
770 }
771
772 data = (__le32 *)dind->b_data;
Mingming Cao617ba132006-10-11 01:20:53 -0700773 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500774 ext4_warning(sb, "new group %u GDT block %llu not reserved",
Yongqiang Yang2f919712011-07-27 21:16:33 -0400775 group, gdblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776 err = -EINVAL;
777 goto exit_dind;
778 }
779
Theodore Ts'ob4097142011-01-10 12:46:59 -0500780 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
781 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 goto exit_dind;
783
Yongqiang Yang2f919712011-07-27 21:16:33 -0400784 err = ext4_journal_get_write_access(handle, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500785 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 goto exit_sbh;
787
Theodore Ts'ob4097142011-01-10 12:46:59 -0500788 err = ext4_journal_get_write_access(handle, dind);
789 if (unlikely(err))
790 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791
Mingming Cao617ba132006-10-11 01:20:53 -0700792 /* ext4_reserve_inode_write() gets a reference on the iloc */
Theodore Ts'ob4097142011-01-10 12:46:59 -0500793 err = ext4_reserve_inode_write(handle, inode, &iloc);
794 if (unlikely(err))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700795 goto exit_dindj;
796
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400797 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
798 sizeof(struct buffer_head *),
799 GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700800 if (!n_group_desc) {
801 err = -ENOMEM;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400802 ext4_warning(sb, "not enough memory for %lu groups",
803 gdb_num + 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804 goto exit_inode;
805 }
806
807 /*
808 * Finally, we have all of the possible failures behind us...
809 *
810 * Remove new GDT block from inode double-indirect block and clear out
811 * the new GDT block for use (which also "frees" the backup GDT blocks
812 * from the reserved inode). We don't need to change the bitmaps for
813 * these blocks, because they are marked as in-use from being in the
814 * reserved inode, and will become GDT blocks (primary and backup).
815 */
Mingming Cao617ba132006-10-11 01:20:53 -0700816 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -0500817 err = ext4_handle_dirty_metadata(handle, NULL, dind);
818 if (unlikely(err)) {
819 ext4_std_error(sb, err);
820 goto exit_inode;
821 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -0700823 ext4_mark_iloc_dirty(handle, inode, &iloc);
Yongqiang Yang2f919712011-07-27 21:16:33 -0400824 memset(gdb_bh->b_data, 0, sb->s_blocksize);
825 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500826 if (unlikely(err)) {
827 ext4_std_error(sb, err);
828 goto exit_inode;
829 }
830 brelse(dind);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831
Mingming Cao617ba132006-10-11 01:20:53 -0700832 o_group_desc = EXT4_SB(sb)->s_group_desc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833 memcpy(n_group_desc, o_group_desc,
Mingming Cao617ba132006-10-11 01:20:53 -0700834 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Yongqiang Yang2f919712011-07-27 21:16:33 -0400835 n_group_desc[gdb_num] = gdb_bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700836 EXT4_SB(sb)->s_group_desc = n_group_desc;
837 EXT4_SB(sb)->s_gdb_count++;
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400838 ext4_kvfree(o_group_desc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700839
Marcin Slusarze8546d02008-04-17 10:38:59 -0400840 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -0400841 err = ext4_handle_dirty_super(handle, sb);
Theodore Ts'ob4097142011-01-10 12:46:59 -0500842 if (err)
843 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700844
Theodore Ts'ob4097142011-01-10 12:46:59 -0500845 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700846
847exit_inode:
Theodore Ts'of18a5f22011-08-01 08:45:38 -0400848 ext4_kvfree(n_group_desc);
Amir Goldstein537a0312011-03-20 22:57:02 -0400849 /* ext4_handle_release_buffer(handle, iloc.bh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700850 brelse(iloc.bh);
851exit_dindj:
Amir Goldstein537a0312011-03-20 22:57:02 -0400852 /* ext4_handle_release_buffer(handle, dind); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700853exit_sbh:
Amir Goldstein537a0312011-03-20 22:57:02 -0400854 /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700855exit_dind:
856 brelse(dind);
857exit_bh:
Yongqiang Yang2f919712011-07-27 21:16:33 -0400858 brelse(gdb_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859
Mingming Cao617ba132006-10-11 01:20:53 -0700860 ext4_debug("leaving with error %d\n", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 return err;
862}
863
864/*
Yongqiang Yang01f795f2012-09-05 01:33:50 -0400865 * add_new_gdb_meta_bg is the sister of add_new_gdb.
866 */
867static int add_new_gdb_meta_bg(struct super_block *sb,
868 handle_t *handle, ext4_group_t group) {
869 ext4_fsblk_t gdblock;
870 struct buffer_head *gdb_bh;
871 struct buffer_head **o_group_desc, **n_group_desc;
872 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
873 int err;
874
875 gdblock = ext4_meta_bg_first_block_no(sb, group) +
876 ext4_bg_has_super(sb, group);
877 gdb_bh = sb_bread(sb, gdblock);
878 if (!gdb_bh)
879 return -EIO;
880 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
881 sizeof(struct buffer_head *),
882 GFP_NOFS);
883 if (!n_group_desc) {
884 err = -ENOMEM;
885 ext4_warning(sb, "not enough memory for %lu groups",
886 gdb_num + 1);
887 return err;
888 }
889
890 o_group_desc = EXT4_SB(sb)->s_group_desc;
891 memcpy(n_group_desc, o_group_desc,
892 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
893 n_group_desc[gdb_num] = gdb_bh;
894 EXT4_SB(sb)->s_group_desc = n_group_desc;
895 EXT4_SB(sb)->s_gdb_count++;
896 ext4_kvfree(o_group_desc);
897 err = ext4_journal_get_write_access(handle, gdb_bh);
898 if (unlikely(err))
899 brelse(gdb_bh);
900 return err;
901}
902
903/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700904 * Called when we are adding a new group which has a backup copy of each of
905 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
906 * We need to add these reserved backup GDT blocks to the resize inode, so
907 * that they are kept for future resizing and not allocated to files.
908 *
909 * Each reserved backup GDT block will go into a different indirect block.
910 * The indirect blocks are actually the primary reserved GDT blocks,
911 * so we know in advance what their block numbers are. We only get the
912 * double-indirect block to verify it is pointing to the primary reserved
913 * GDT blocks so we don't overwrite a data block by accident. The reserved
914 * backup GDT blocks are stored in their reserved primary GDT block.
915 */
916static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400917 ext4_group_t group)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700918{
919 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -0700920 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700921 struct buffer_head **primary;
922 struct buffer_head *dind;
Mingming Cao617ba132006-10-11 01:20:53 -0700923 struct ext4_iloc iloc;
924 ext4_fsblk_t blk;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925 __le32 *data, *end;
926 int gdbackups = 0;
927 int res, i;
928 int err;
929
Josef Bacik216553c2008-04-29 22:02:02 -0400930 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 if (!primary)
932 return -ENOMEM;
933
Mingming Cao617ba132006-10-11 01:20:53 -0700934 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700935 dind = sb_bread(sb, le32_to_cpu(*data));
936 if (!dind) {
937 err = -EIO;
938 goto exit_free;
939 }
940
Mingming Cao617ba132006-10-11 01:20:53 -0700941 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
Josef Bacik94460092008-06-06 18:05:52 -0400942 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
943 EXT4_ADDR_PER_BLOCK(sb));
Mingming Cao617ba132006-10-11 01:20:53 -0700944 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700945
946 /* Get each reserved primary GDT block and verify it holds backups */
947 for (res = 0; res < reserved_gdb; res++, blk++) {
948 if (le32_to_cpu(*data) != blk) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500949 ext4_warning(sb, "reserved block %llu"
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 " not at offset %ld",
951 blk,
952 (long)(data - (__le32 *)dind->b_data));
953 err = -EINVAL;
954 goto exit_bh;
955 }
956 primary[res] = sb_bread(sb, blk);
957 if (!primary[res]) {
958 err = -EIO;
959 goto exit_bh;
960 }
Yongqiang Yangc72df9f2012-01-03 23:43:39 -0500961 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
962 if (gdbackups < 0) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963 brelse(primary[res]);
964 err = gdbackups;
965 goto exit_bh;
966 }
967 if (++data >= end)
968 data = (__le32 *)dind->b_data;
969 }
970
971 for (i = 0; i < reserved_gdb; i++) {
Mingming Cao617ba132006-10-11 01:20:53 -0700972 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973 /*
974 int j;
975 for (j = 0; j < i; j++)
Amir Goldstein537a0312011-03-20 22:57:02 -0400976 ext4_handle_release_buffer(handle, primary[j]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700977 */
978 goto exit_bh;
979 }
980 }
981
Mingming Cao617ba132006-10-11 01:20:53 -0700982 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700983 goto exit_bh;
984
985 /*
986 * Finally we can add each of the reserved backup GDT blocks from
987 * the new group to its reserved primary GDT block.
988 */
Yongqiang Yang668f4dc2011-07-27 21:23:13 -0400989 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700990 for (i = 0; i < reserved_gdb; i++) {
991 int err2;
992 data = (__le32 *)primary[i]->b_data;
993 /* printk("reserving backup %lu[%u] = %lu\n",
994 primary[i]->b_blocknr, gdbackups,
995 blk + primary[i]->b_blocknr); */
996 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
Frank Mayhar03901312009-01-07 00:06:22 -0500997 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998 if (!err)
999 err = err2;
1000 }
1001 inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
Mingming Cao617ba132006-10-11 01:20:53 -07001002 ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003
1004exit_bh:
1005 while (--res >= 0)
1006 brelse(primary[res]);
1007 brelse(dind);
1008
1009exit_free:
1010 kfree(primary);
1011
1012 return err;
1013}
1014
1015/*
Mingming Cao617ba132006-10-11 01:20:53 -07001016 * Update the backup copies of the ext4 metadata. These don't need to be part
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017 * of the main resize transaction, because e2fsck will re-write them if there
1018 * is a problem (basically only OOM will cause a problem). However, we
1019 * _should_ update the backups if possible, in case the primary gets trashed
1020 * for some reason and we need to run e2fsck from a backup superblock. The
1021 * important part is that the new block and inode counts are in the backup
1022 * superblocks, and the location of the new group metadata in the GDT backups.
1023 *
Theodore Ts'o32ed5052009-04-25 22:53:39 -04001024 * We do not need take the s_resize_lock for this, because these
1025 * blocks are not otherwise touched by the filesystem code when it is
1026 * mounted. We don't need to worry about last changing from
1027 * sbi->s_groups_count, because the worst that can happen is that we
1028 * do not copy the full number of backups at this time. The resize
1029 * which changed s_groups_count will backup again.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 */
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001031static void update_backups(struct super_block *sb, int blk_off, char *data,
1032 int size, int meta_bg)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033{
Mingming Cao617ba132006-10-11 01:20:53 -07001034 struct ext4_sb_info *sbi = EXT4_SB(sb);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001035 ext4_group_t last;
Mingming Cao617ba132006-10-11 01:20:53 -07001036 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037 unsigned three = 1;
1038 unsigned five = 5;
1039 unsigned seven = 7;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001040 ext4_group_t group = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001041 int rest = sb->s_blocksize - size;
1042 handle_t *handle;
1043 int err = 0, err2;
1044
Mingming Cao617ba132006-10-11 01:20:53 -07001045 handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 if (IS_ERR(handle)) {
1047 group = 1;
1048 err = PTR_ERR(handle);
1049 goto exit_err;
1050 }
1051
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001052 if (meta_bg == 0) {
1053 group = ext4_list_backups(sb, &three, &five, &seven);
1054 last = sbi->s_groups_count;
1055 } else {
1056 group = ext4_meta_bg_first_group(sb, group) + 1;
1057 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1058 }
1059
1060 while (group < sbi->s_groups_count) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001061 struct buffer_head *bh;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001062 ext4_fsblk_t backup_block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063
1064 /* Out of journal space, and can't get more - abort - so sad */
Frank Mayhar03901312009-01-07 00:06:22 -05001065 if (ext4_handle_valid(handle) &&
1066 handle->h_buffer_credits == 0 &&
Mingming Cao617ba132006-10-11 01:20:53 -07001067 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1068 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001069 break;
1070
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001071 if (meta_bg == 0)
1072 backup_block = group * bpg + blk_off;
1073 else
1074 backup_block = (ext4_group_first_block_no(sb, group) +
1075 ext4_bg_has_super(sb, group));
1076
1077 bh = sb_getblk(sb, backup_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078 if (!bh) {
1079 err = -EIO;
1080 break;
1081 }
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001082 ext4_debug("update metadata backup %llu(+%llu)\n",
1083 backup_block, backup_block -
1084 ext4_group_first_block_no(sb, group));
Mingming Cao617ba132006-10-11 01:20:53 -07001085 if ((err = ext4_journal_get_write_access(handle, bh)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086 break;
1087 lock_buffer(bh);
1088 memcpy(bh->b_data, data, size);
1089 if (rest)
1090 memset(bh->b_data + size, 0, rest);
1091 set_buffer_uptodate(bh);
1092 unlock_buffer(bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05001093 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1094 if (unlikely(err))
1095 ext4_std_error(sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096 brelse(bh);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001097
1098 if (meta_bg == 0)
1099 group = ext4_list_backups(sb, &three, &five, &seven);
1100 else if (group == last)
1101 break;
1102 else
1103 group = last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001104 }
Mingming Cao617ba132006-10-11 01:20:53 -07001105 if ((err2 = ext4_journal_stop(handle)) && !err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106 err = err2;
1107
1108 /*
1109 * Ugh! Need to have e2fsck write the backup copies. It is too
1110 * late to revert the resize, we shouldn't fail just because of
1111 * the backup copies (they are only needed in case of corruption).
1112 *
1113 * However, if we got here we have a journal problem too, so we
1114 * can't really start a transaction to mark the superblock.
1115 * Chicken out and just set the flag on the hope it will be written
1116 * to disk, and if not - we will simply wait until next fsck.
1117 */
1118exit_err:
1119 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001120 ext4_warning(sb, "can't update backup for group %u (err %d), "
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001121 "forcing fsck on next reboot", group, err);
Mingming Cao617ba132006-10-11 01:20:53 -07001122 sbi->s_mount_state &= ~EXT4_VALID_FS;
1123 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001124 mark_buffer_dirty(sbi->s_sbh);
1125 }
1126}
1127
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001128/*
1129 * ext4_add_new_descs() adds @count group descriptor of groups
1130 * starting at @group
1131 *
1132 * @handle: journal handle
1133 * @sb: super block
1134 * @group: the group no. of the first group desc to be added
1135 * @resize_inode: the resize inode
1136 * @count: number of group descriptors to be added
1137 */
1138static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1139 ext4_group_t group, struct inode *resize_inode,
1140 ext4_group_t count)
1141{
1142 struct ext4_sb_info *sbi = EXT4_SB(sb);
1143 struct ext4_super_block *es = sbi->s_es;
1144 struct buffer_head *gdb_bh;
1145 int i, gdb_off, gdb_num, err = 0;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001146 int meta_bg;
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001147
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001148 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001149 for (i = 0; i < count; i++, group++) {
1150 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1151 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1152
1153 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1154 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1155
1156 /*
1157 * We will only either add reserved group blocks to a backup group
1158 * or remove reserved blocks for the first group in a new group block.
1159 * Doing both would be mean more complex code, and sane people don't
1160 * use non-sparse filesystems anymore. This is already checked above.
1161 */
1162 if (gdb_off) {
1163 gdb_bh = sbi->s_group_desc[gdb_num];
1164 err = ext4_journal_get_write_access(handle, gdb_bh);
1165
1166 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1167 err = reserve_backup_gdb(handle, resize_inode, group);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001168 } else if (meta_bg != 0) {
1169 err = add_new_gdb_meta_bg(sb, handle, group);
1170 } else {
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001171 err = add_new_gdb(handle, resize_inode, group);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001172 }
Yongqiang Yangbb08c1e2012-01-03 23:20:50 -05001173 if (err)
1174 break;
1175 }
1176 return err;
1177}
1178
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001179static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1180{
1181 struct buffer_head *bh = sb_getblk(sb, block);
1182 if (!bh)
1183 return NULL;
Dmitry Monakhov7f1468d2012-09-25 23:19:25 -04001184 if (!bh_uptodate_or_lock(bh)) {
1185 if (bh_submit_read(bh) < 0) {
1186 brelse(bh);
1187 return NULL;
1188 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001189 }
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001190
1191 return bh;
1192}
1193
1194static int ext4_set_bitmap_checksums(struct super_block *sb,
1195 ext4_group_t group,
1196 struct ext4_group_desc *gdp,
1197 struct ext4_new_group_data *group_data)
1198{
1199 struct buffer_head *bh;
1200
1201 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1202 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1203 return 0;
1204
1205 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1206 if (!bh)
1207 return -EIO;
1208 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1209 EXT4_INODES_PER_GROUP(sb) / 8);
1210 brelse(bh);
1211
Darrick J. Wongfa77dcf2012-04-29 18:35:10 -04001212 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1213 if (!bh)
1214 return -EIO;
1215 ext4_block_bitmap_csum_set(sb, group, gdp, bh,
1216 EXT4_BLOCKS_PER_GROUP(sb) / 8);
1217 brelse(bh);
1218
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001219 return 0;
1220}
1221
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001222/*
1223 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1224 */
1225static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1226 struct ext4_new_flex_group_data *flex_gd)
1227{
1228 struct ext4_new_group_data *group_data = flex_gd->groups;
1229 struct ext4_group_desc *gdp;
1230 struct ext4_sb_info *sbi = EXT4_SB(sb);
1231 struct buffer_head *gdb_bh;
1232 ext4_group_t group;
1233 __u16 *bg_flags = flex_gd->bg_flags;
1234 int i, gdb_off, gdb_num, err = 0;
1235
1236
1237 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1238 group = group_data->group;
1239
1240 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1241 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1242
1243 /*
1244 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1245 */
1246 gdb_bh = sbi->s_group_desc[gdb_num];
1247 /* Update group descriptor block for new group */
Theodore Ts'o2716b802012-05-28 17:47:52 -04001248 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001249 gdb_off * EXT4_DESC_SIZE(sb));
1250
1251 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1252 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1253 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
Darrick J. Wong41a246d2012-04-29 18:33:10 -04001254 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1255 if (err) {
1256 ext4_std_error(sb, err);
1257 break;
1258 }
1259
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001260 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1261 ext4_free_group_clusters_set(sb, gdp,
1262 EXT4_B2C(sbi, group_data->free_blocks_count));
1263 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
Theodore Ts'o93f90522012-09-12 14:32:42 -04001264 if (ext4_has_group_desc_csum(sb))
1265 ext4_itable_unused_set(sb, gdp,
1266 EXT4_INODES_PER_GROUP(sb));
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001267 gdp->bg_flags = cpu_to_le16(*bg_flags);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001268 ext4_group_desc_csum_set(sb, group, gdp);
Yongqiang Yang083f5b22012-01-03 23:37:31 -05001269
1270 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1271 if (unlikely(err)) {
1272 ext4_std_error(sb, err);
1273 break;
1274 }
1275
1276 /*
1277 * We can allocate memory for mb_alloc based on the new group
1278 * descriptor
1279 */
1280 err = ext4_mb_add_groupinfo(sb, group, gdp);
1281 if (err)
1282 break;
1283 }
1284 return err;
1285}
1286
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001287/*
1288 * ext4_update_super() updates the super block so that the newly added
1289 * groups can be seen by the filesystem.
1290 *
1291 * @sb: super block
1292 * @flex_gd: new added groups
1293 */
1294static void ext4_update_super(struct super_block *sb,
1295 struct ext4_new_flex_group_data *flex_gd)
1296{
1297 ext4_fsblk_t blocks_count = 0;
1298 ext4_fsblk_t free_blocks = 0;
1299 ext4_fsblk_t reserved_blocks = 0;
1300 struct ext4_new_group_data *group_data = flex_gd->groups;
1301 struct ext4_sb_info *sbi = EXT4_SB(sb);
1302 struct ext4_super_block *es = sbi->s_es;
Theodore Ts'o8a991842012-07-22 20:23:31 -04001303 int i;
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001304
1305 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1306 /*
1307 * Make the new blocks and inodes valid next. We do this before
1308 * increasing the group count so that once the group is enabled,
1309 * all of its blocks and inodes are already valid.
1310 *
1311 * We always allocate group-by-group, then block-by-block or
1312 * inode-by-inode within a group, so enabling these
1313 * blocks/inodes before the group is live won't actually let us
1314 * allocate the new space yet.
1315 */
1316 for (i = 0; i < flex_gd->count; i++) {
1317 blocks_count += group_data[i].blocks_count;
1318 free_blocks += group_data[i].free_blocks_count;
1319 }
1320
1321 reserved_blocks = ext4_r_blocks_count(es) * 100;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001322 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001323 reserved_blocks *= blocks_count;
1324 do_div(reserved_blocks, 100);
1325
1326 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001327 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001328 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1329 flex_gd->count);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001330 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1331 flex_gd->count);
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001332
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001333 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001334 /*
1335 * We need to protect s_groups_count against other CPUs seeing
1336 * inconsistent state in the superblock.
1337 *
1338 * The precise rules we use are:
1339 *
1340 * * Writers must perform a smp_wmb() after updating all
1341 * dependent data and before modifying the groups count
1342 *
1343 * * Readers must perform an smp_rmb() after reading the groups
1344 * count and before reading any dependent data.
1345 *
1346 * NB. These rules can be relaxed when checking the group count
1347 * while freeing data, as we can only allocate from a block
1348 * group after serialising against the group count, and we can
1349 * only then free after serialising in turn against that
1350 * allocation.
1351 */
1352 smp_wmb();
1353
1354 /* Update the global fs size fields */
1355 sbi->s_groups_count += flex_gd->count;
1356
1357 /* Update the reserved block counts only once the new group is
1358 * active. */
1359 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1360 reserved_blocks);
1361
1362 /* Update the free space counts */
1363 percpu_counter_add(&sbi->s_freeclusters_counter,
1364 EXT4_B2C(sbi, free_blocks));
1365 percpu_counter_add(&sbi->s_freeinodes_counter,
1366 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1367
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001368 ext4_debug("free blocks count %llu",
1369 percpu_counter_read(&sbi->s_freeclusters_counter));
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001370 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1371 EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1372 sbi->s_log_groups_per_flex) {
1373 ext4_group_t flex_group;
1374 flex_group = ext4_flex_group(sbi, group_data[0].group);
1375 atomic_add(EXT4_B2C(sbi, free_blocks),
1376 &sbi->s_flex_groups[flex_group].free_clusters);
1377 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1378 &sbi->s_flex_groups[flex_group].free_inodes);
1379 }
1380
Theodore Ts'o952fc182012-07-09 16:27:05 -04001381 /*
1382 * Update the fs overhead information
1383 */
1384 ext4_calculate_overhead(sb);
1385
Yongqiang Yang2e10e2f2012-01-03 23:41:39 -05001386 if (test_opt(sb, DEBUG))
1387 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1388 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1389 blocks_count, free_blocks, reserved_blocks);
1390}
1391
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001392/* Add a flex group to an fs. Ensure we handle all possible error conditions
1393 * _before_ we start modifying the filesystem, because we cannot abort the
1394 * transaction and not have it write the data to disk.
1395 */
1396static int ext4_flex_group_add(struct super_block *sb,
1397 struct inode *resize_inode,
1398 struct ext4_new_flex_group_data *flex_gd)
1399{
1400 struct ext4_sb_info *sbi = EXT4_SB(sb);
1401 struct ext4_super_block *es = sbi->s_es;
1402 ext4_fsblk_t o_blocks_count;
1403 ext4_grpblk_t last;
1404 ext4_group_t group;
1405 handle_t *handle;
1406 unsigned reserved_gdb;
1407 int err = 0, err2 = 0, credit;
1408
1409 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1410
1411 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1412 o_blocks_count = ext4_blocks_count(es);
1413 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1414 BUG_ON(last);
1415
1416 err = setup_new_flex_group_blocks(sb, flex_gd);
1417 if (err)
1418 goto exit;
1419 /*
1420 * We will always be modifying at least the superblock and GDT
1421 * block. If we are adding a group past the last current GDT block,
1422 * we will also modify the inode and the dindirect block. If we
1423 * are adding a group with superblock/GDT backups we will also
1424 * modify each of the reserved GDT dindirect blocks.
1425 */
1426 credit = flex_gd->count * 4 + reserved_gdb;
1427 handle = ext4_journal_start_sb(sb, credit);
1428 if (IS_ERR(handle)) {
1429 err = PTR_ERR(handle);
1430 goto exit;
1431 }
1432
1433 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1434 if (err)
1435 goto exit_journal;
1436
1437 group = flex_gd->groups[0].group;
1438 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1439 err = ext4_add_new_descs(handle, sb, group,
1440 resize_inode, flex_gd->count);
1441 if (err)
1442 goto exit_journal;
1443
1444 err = ext4_setup_new_descs(handle, sb, flex_gd);
1445 if (err)
1446 goto exit_journal;
1447
1448 ext4_update_super(sb, flex_gd);
1449
1450 err = ext4_handle_dirty_super(handle, sb);
1451
1452exit_journal:
1453 err2 = ext4_journal_stop(handle);
1454 if (!err)
1455 err = err2;
1456
1457 if (!err) {
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001458 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1459 int gdb_num_end = ((group + flex_gd->count - 1) /
1460 EXT4_DESC_PER_BLOCK(sb));
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001461 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1462 EXT4_FEATURE_INCOMPAT_META_BG);
Tao Ma0acdb882012-09-26 00:08:57 -04001463 sector_t old_gdb = 0;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001464
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001465 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001466 sizeof(struct ext4_super_block), 0);
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001467 for (; gdb_num <= gdb_num_end; gdb_num++) {
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001468 struct buffer_head *gdb_bh;
Yongqiang Yang2ebd1702012-09-05 01:27:50 -04001469
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001470 gdb_bh = sbi->s_group_desc[gdb_num];
Tao Ma0acdb882012-09-26 00:08:57 -04001471 if (old_gdb == gdb_bh->b_blocknr)
1472 continue;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001473 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001474 gdb_bh->b_size, meta_bg);
Tao Ma0acdb882012-09-26 00:08:57 -04001475 old_gdb = gdb_bh->b_blocknr;
Yongqiang Yang4bac1f82012-01-03 23:44:38 -05001476 }
1477 }
1478exit:
1479 return err;
1480}
1481
Yongqiang Yang19c52462012-01-04 17:09:44 -05001482static int ext4_setup_next_flex_gd(struct super_block *sb,
1483 struct ext4_new_flex_group_data *flex_gd,
1484 ext4_fsblk_t n_blocks_count,
1485 unsigned long flexbg_size)
1486{
1487 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1488 struct ext4_new_group_data *group_data = flex_gd->groups;
1489 ext4_fsblk_t o_blocks_count;
1490 ext4_group_t n_group;
1491 ext4_group_t group;
1492 ext4_group_t last_group;
1493 ext4_grpblk_t last;
1494 ext4_grpblk_t blocks_per_group;
1495 unsigned long i;
1496
1497 blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1498
1499 o_blocks_count = ext4_blocks_count(es);
1500
1501 if (o_blocks_count == n_blocks_count)
1502 return 0;
1503
1504 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1505 BUG_ON(last);
1506 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1507
1508 last_group = group | (flexbg_size - 1);
1509 if (last_group > n_group)
1510 last_group = n_group;
1511
1512 flex_gd->count = last_group - group + 1;
1513
1514 for (i = 0; i < flex_gd->count; i++) {
1515 int overhead;
1516
1517 group_data[i].group = group + i;
1518 group_data[i].blocks_count = blocks_per_group;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001519 overhead = ext4_group_overhead_blocks(sb, group + i);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001520 group_data[i].free_blocks_count = blocks_per_group - overhead;
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001521 if (ext4_has_group_desc_csum(sb))
Yongqiang Yang19c52462012-01-04 17:09:44 -05001522 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1523 EXT4_BG_INODE_UNINIT;
1524 else
1525 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1526 }
1527
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04001528 if (last_group == n_group && ext4_has_group_desc_csum(sb))
Yongqiang Yang19c52462012-01-04 17:09:44 -05001529 /* We need to initialize block bitmap of last group. */
1530 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1531
1532 if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1533 group_data[i - 1].blocks_count = last + 1;
1534 group_data[i - 1].free_blocks_count -= blocks_per_group-
1535 last - 1;
1536 }
1537
1538 return 1;
1539}
1540
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541/* Add group descriptor data to an existing or new group descriptor block.
1542 * Ensure we handle all possible error conditions _before_ we start modifying
1543 * the filesystem, because we cannot abort the transaction and not have it
1544 * write the data to disk.
1545 *
1546 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1547 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1548 *
1549 * We only need to hold the superblock lock while we are actually adding
1550 * in the new group's counts to the superblock. Prior to that we have
1551 * not really "added" the group at all. We re-check that we are still
1552 * adding in the last group in case things have changed since verifying.
1553 */
Mingming Cao617ba132006-10-11 01:20:53 -07001554int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001555{
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001556 struct ext4_new_flex_group_data flex_gd;
Mingming Cao617ba132006-10-11 01:20:53 -07001557 struct ext4_sb_info *sbi = EXT4_SB(sb);
1558 struct ext4_super_block *es = sbi->s_es;
1559 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001561 struct inode *inode = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001562 int gdb_off, gdb_num;
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001563 int err;
1564 __u16 bg_flags = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001565
Mingming Cao617ba132006-10-11 01:20:53 -07001566 gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1567 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001568
Mingming Cao617ba132006-10-11 01:20:53 -07001569 if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1570 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001571 ext4_warning(sb, "Can't resize non-sparse filesystem further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001572 return -EPERM;
1573 }
1574
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001575 if (ext4_blocks_count(es) + input->blocks_count <
1576 ext4_blocks_count(es)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001577 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001578 return -EINVAL;
1579 }
1580
Mingming Cao617ba132006-10-11 01:20:53 -07001581 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001582 le32_to_cpu(es->s_inodes_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001583 ext4_warning(sb, "inodes_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584 return -EINVAL;
1585 }
1586
1587 if (reserved_gdb || gdb_off == 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001588 if (!EXT4_HAS_COMPAT_FEATURE(sb,
Josef Bacik37609fd2008-08-19 22:13:41 -04001589 EXT4_FEATURE_COMPAT_RESIZE_INODE)
1590 || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001591 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001592 "No reserved GDT blocks, can't resize");
1593 return -EPERM;
1594 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001595 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1596 if (IS_ERR(inode)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001597 ext4_warning(sb, "Error opening resize inode");
David Howells1d1fe1e2008-02-07 00:15:37 -08001598 return PTR_ERR(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001599 }
1600 }
1601
Aneesh Kumar K.V920313a2009-01-05 21:36:19 -05001602
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001603 err = verify_group_input(sb, input);
Aneesh Kumar K.V08c3a812009-09-09 23:50:17 -04001604 if (err)
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001605 goto out;
Theodore Ts'oc2ea3fd2008-10-10 09:40:52 -04001606
Theodore Ts'o117fff12012-09-05 01:29:50 -04001607 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1608 if (err)
1609 return err;
1610
Theodore Ts'o28623c22012-09-05 01:31:50 -04001611 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1612 if (err)
1613 goto out;
1614
Yongqiang Yang61f296c2012-01-04 17:09:50 -05001615 flex_gd.count = 1;
1616 flex_gd.groups = input;
1617 flex_gd.bg_flags = &bg_flags;
1618 err = ext4_flex_group_add(sb, inode, &flex_gd);
1619out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001620 iput(inode);
1621 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001622} /* ext4_group_add */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001623
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001624/*
Yongqiang Yang18e31432012-01-03 23:18:50 -05001625 * extend a group without checking assuming that checking has been done.
1626 */
1627static int ext4_group_extend_no_check(struct super_block *sb,
1628 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1629{
1630 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1631 handle_t *handle;
1632 int err = 0, err2;
1633
1634 /* We will update the superblock, one block bitmap, and
1635 * one group descriptor via ext4_group_add_blocks().
1636 */
1637 handle = ext4_journal_start_sb(sb, 3);
1638 if (IS_ERR(handle)) {
1639 err = PTR_ERR(handle);
1640 ext4_warning(sb, "error %d on journal start", err);
1641 return err;
1642 }
1643
1644 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1645 if (err) {
1646 ext4_warning(sb, "error %d on journal write access", err);
1647 goto errout;
1648 }
1649
1650 ext4_blocks_count_set(es, o_blocks_count + add);
Darrick J. Wong636d7e2e2012-03-20 15:46:11 -04001651 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001652 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1653 o_blocks_count + add);
1654 /* We add the blocks to the bitmap and set the group need init bit */
1655 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1656 if (err)
1657 goto errout;
1658 ext4_handle_dirty_super(handle, sb);
1659 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1660 o_blocks_count + add);
1661errout:
1662 err2 = ext4_journal_stop(handle);
1663 if (err2 && !err)
1664 err = err2;
1665
1666 if (!err) {
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001667 ext4_fsblk_t first_block;
1668 first_block = ext4_group_first_block_no(sb, 0);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001669 if (test_opt(sb, DEBUG))
1670 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1671 "blocks\n", ext4_blocks_count(es));
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001672 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr - first_block,
1673 (char *)es, sizeof(struct ext4_super_block), 0);
Yongqiang Yang18e31432012-01-03 23:18:50 -05001674 }
1675 return err;
1676}
1677
1678/*
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001679 * Extend the filesystem to the new number of blocks specified. This entry
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001680 * point is only used to extend the current filesystem to the end of the last
1681 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1682 * for emergencies (because it has no dependencies on reserved blocks).
1683 *
Mingming Cao617ba132006-10-11 01:20:53 -07001684 * If we _really_ wanted, we could use default values to call ext4_group_add()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001685 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1686 * GDT blocks are reserved to grow to the desired size.
1687 */
Mingming Cao617ba132006-10-11 01:20:53 -07001688int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1689 ext4_fsblk_t n_blocks_count)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001690{
Mingming Cao617ba132006-10-11 01:20:53 -07001691 ext4_fsblk_t o_blocks_count;
Mingming Cao617ba132006-10-11 01:20:53 -07001692 ext4_grpblk_t last;
1693 ext4_grpblk_t add;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001694 struct buffer_head *bh;
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001695 int err;
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001696 ext4_group_t group;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001697
Laurent Vivierbd81d8e2006-10-11 01:21:10 -07001698 o_blocks_count = ext4_blocks_count(es);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001699
1700 if (test_opt(sb, DEBUG))
Theodore Ts'o92b97812012-03-19 23:41:49 -04001701 ext4_msg(sb, KERN_DEBUG,
1702 "extending last group from %llu to %llu blocks",
1703 o_blocks_count, n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001704
1705 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1706 return 0;
1707
1708 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
Theodore Ts'o92b97812012-03-19 23:41:49 -04001709 ext4_msg(sb, KERN_ERR,
1710 "filesystem too large to resize to %llu blocks safely",
1711 n_blocks_count);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001712 if (sizeof(sector_t) < 8)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001713 ext4_warning(sb, "CONFIG_LBDAF not enabled");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001714 return -EINVAL;
1715 }
1716
1717 if (n_blocks_count < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001718 ext4_warning(sb, "can't shrink FS - resize aborted");
Yongqiang Yang8f82f842011-07-26 21:35:44 -04001719 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001720 }
1721
1722 /* Handle the remaining blocks in the last group only. */
Frederic Bohe5f21b0e2008-07-11 19:27:31 -04001723 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001724
1725 if (last == 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001726 ext4_warning(sb, "need to use ext2online to resize further");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 return -EPERM;
1728 }
1729
Mingming Cao617ba132006-10-11 01:20:53 -07001730 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001731
1732 if (o_blocks_count + add < o_blocks_count) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001733 ext4_warning(sb, "blocks_count overflow");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001734 return -EINVAL;
1735 }
1736
1737 if (o_blocks_count + add > n_blocks_count)
1738 add = n_blocks_count - o_blocks_count;
1739
1740 if (o_blocks_count + add < n_blocks_count)
Eric Sandeen12062dd2010-02-15 14:19:27 -05001741 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742 o_blocks_count + add, add);
1743
1744 /* See if the device is actually as big as what was requested */
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -04001745 bh = sb_bread(sb, o_blocks_count + add - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001746 if (!bh) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001747 ext4_warning(sb, "can't read last block, resize aborted");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001748 return -ENOSPC;
1749 }
1750 brelse(bh);
1751
Yongqiang Yangd89651c2012-01-04 17:09:48 -05001752 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001753 return err;
Mingming Cao617ba132006-10-11 01:20:53 -07001754} /* ext4_group_extend */
Yongqiang Yang19c52462012-01-04 17:09:44 -05001755
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001756
1757static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1758{
1759 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1760}
1761
1762/*
1763 * Release the resize inode and drop the resize_inode feature if there
1764 * are no more reserved gdt blocks, and then convert the file system
1765 * to enable meta_bg
1766 */
1767static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1768{
1769 handle_t *handle;
1770 struct ext4_sb_info *sbi = EXT4_SB(sb);
1771 struct ext4_super_block *es = sbi->s_es;
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001772 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001773 ext4_fsblk_t nr;
1774 int i, ret, err = 0;
1775 int credits = 1;
1776
1777 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001778 if (inode) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001779 if (es->s_reserved_gdt_blocks) {
1780 ext4_error(sb, "Unexpected non-zero "
1781 "s_reserved_gdt_blocks");
1782 return -EPERM;
1783 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001784
1785 /* Do a quick sanity check of the resize inode */
1786 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1787 goto invalid_resize_inode;
1788 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1789 if (i == EXT4_DIND_BLOCK) {
1790 if (ei->i_data[i])
1791 continue;
1792 else
1793 goto invalid_resize_inode;
1794 }
1795 if (ei->i_data[i])
1796 goto invalid_resize_inode;
1797 }
1798 credits += 3; /* block bitmap, bg descriptor, resize inode */
1799 }
1800
1801 handle = ext4_journal_start_sb(sb, credits);
1802 if (IS_ERR(handle))
1803 return PTR_ERR(handle);
1804
1805 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1806 if (err)
1807 goto errout;
1808
1809 EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1810 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1811 sbi->s_es->s_first_meta_bg =
1812 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1813
1814 err = ext4_handle_dirty_super(handle, sb);
1815 if (err) {
1816 ext4_std_error(sb, err);
1817 goto errout;
1818 }
1819
1820 if (inode) {
1821 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1822 ext4_free_blocks(handle, inode, NULL, nr, 1,
1823 EXT4_FREE_BLOCKS_METADATA |
1824 EXT4_FREE_BLOCKS_FORGET);
1825 ei->i_data[EXT4_DIND_BLOCK] = 0;
1826 inode->i_blocks = 0;
1827
1828 err = ext4_mark_inode_dirty(handle, inode);
1829 if (err)
1830 ext4_std_error(sb, err);
1831 }
1832
1833errout:
1834 ret = ext4_journal_stop(handle);
1835 if (!err)
1836 err = ret;
1837 return ret;
1838
1839invalid_resize_inode:
1840 ext4_error(sb, "corrupted/inconsistent resize inode");
1841 return -EINVAL;
1842}
1843
Yongqiang Yang19c52462012-01-04 17:09:44 -05001844/*
1845 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1846 *
1847 * @sb: super block of the fs to be resized
1848 * @n_blocks_count: the number of blocks resides in the resized fs
1849 */
1850int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1851{
1852 struct ext4_new_flex_group_data *flex_gd = NULL;
1853 struct ext4_sb_info *sbi = EXT4_SB(sb);
1854 struct ext4_super_block *es = sbi->s_es;
1855 struct buffer_head *bh;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001856 struct inode *resize_inode = NULL;
1857 ext4_grpblk_t add, offset;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001858 unsigned long n_desc_blocks;
1859 unsigned long o_desc_blocks;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001860 ext4_group_t o_group;
1861 ext4_group_t n_group;
1862 ext4_fsblk_t o_blocks_count;
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001863 ext4_fsblk_t n_blocks_count_retry = 0;
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001864 unsigned long last_update_time = 0;
Theodore Ts'o117fff12012-09-05 01:29:50 -04001865 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001866 int meta_bg;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001867
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001868 /* See if the device is actually as big as what was requested */
1869 bh = sb_bread(sb, n_blocks_count - 1);
1870 if (!bh) {
1871 ext4_warning(sb, "can't read last block, resize aborted");
1872 return -ENOSPC;
1873 }
1874 brelse(bh);
1875
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001876retry:
Yongqiang Yang19c52462012-01-04 17:09:44 -05001877 o_blocks_count = ext4_blocks_count(es);
1878
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001879 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1880 "to %llu blocks", o_blocks_count, n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001881
1882 if (n_blocks_count < o_blocks_count) {
1883 /* On-line shrinking not supported */
1884 ext4_warning(sb, "can't shrink FS - resize aborted");
1885 return -EINVAL;
1886 }
1887
1888 if (n_blocks_count == o_blocks_count)
1889 /* Nothing need to do */
1890 return 0;
1891
1892 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001893 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001894
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001895 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1896 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001897
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001898 meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
Yongqiang Yang19c52462012-01-04 17:09:44 -05001899
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001900 if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1901 if (meta_bg) {
1902 ext4_error(sb, "resize_inode and meta_bg enabled "
1903 "simultaneously");
1904 return -EINVAL;
1905 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001906 if (n_desc_blocks > o_desc_blocks +
1907 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1908 n_blocks_count_retry = n_blocks_count;
1909 n_desc_blocks = o_desc_blocks +
1910 le16_to_cpu(es->s_reserved_gdt_blocks);
1911 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1912 n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1913 n_group--; /* set to last group number */
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001914 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001915
1916 if (!resize_inode)
1917 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001918 if (IS_ERR(resize_inode)) {
1919 ext4_warning(sb, "Error opening resize inode");
1920 return PTR_ERR(resize_inode);
1921 }
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001922 }
1923
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001924 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001925 err = ext4_convert_meta_bg(sb, resize_inode);
1926 if (err)
1927 goto out;
1928 if (resize_inode) {
1929 iput(resize_inode);
1930 resize_inode = NULL;
1931 }
1932 if (n_blocks_count_retry) {
1933 n_blocks_count = n_blocks_count_retry;
1934 n_blocks_count_retry = 0;
1935 goto retry;
1936 }
Yongqiang Yang19c52462012-01-04 17:09:44 -05001937 }
1938
Lukas Czernera0ade1d2012-02-20 23:02:06 -05001939 /* extend the last group */
1940 if (n_group == o_group)
1941 add = n_blocks_count - o_blocks_count;
1942 else
1943 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1944 if (add > 0) {
Yongqiang Yang19c52462012-01-04 17:09:44 -05001945 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1946 if (err)
1947 goto out;
1948 }
1949
Yongqiang Yangd7574ad2012-09-05 01:23:50 -04001950 if (ext4_blocks_count(es) == n_blocks_count)
Yongqiang Yang19c52462012-01-04 17:09:44 -05001951 goto out;
1952
Theodore Ts'o117fff12012-09-05 01:29:50 -04001953 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1954 if (err)
1955 return err;
1956
Theodore Ts'o28623c22012-09-05 01:31:50 -04001957 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1958 if (err)
1959 goto out;
1960
Yongqiang Yang19c52462012-01-04 17:09:44 -05001961 flex_gd = alloc_flex_gd(flexbg_size);
1962 if (flex_gd == NULL) {
1963 err = -ENOMEM;
1964 goto out;
1965 }
1966
1967 /* Add flex groups. Note that a regular group is a
1968 * flex group with 1 group.
1969 */
1970 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1971 flexbg_size)) {
Theodore Ts'o4da4a562012-09-13 10:24:21 -04001972 if (jiffies - last_update_time > HZ * 10) {
1973 if (last_update_time)
1974 ext4_msg(sb, KERN_INFO,
1975 "resized to %llu blocks",
1976 ext4_blocks_count(es));
1977 last_update_time = jiffies;
1978 }
Yongqiang Yang03c1c292012-09-05 01:21:50 -04001979 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1980 break;
Yongqiang Yang19c52462012-01-04 17:09:44 -05001981 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1982 if (unlikely(err))
1983 break;
1984 }
1985
Theodore Ts'o1c6bd712012-09-13 10:19:24 -04001986 if (!err && n_blocks_count_retry) {
1987 n_blocks_count = n_blocks_count_retry;
1988 n_blocks_count_retry = 0;
1989 free_flex_gd(flex_gd);
1990 flex_gd = NULL;
1991 goto retry;
1992 }
1993
Yongqiang Yang19c52462012-01-04 17:09:44 -05001994out:
1995 if (flex_gd)
1996 free_flex_gd(flex_gd);
Yongqiang Yang01f795f2012-09-05 01:33:50 -04001997 if (resize_inode != NULL)
1998 iput(resize_inode);
Theodore Ts'o59e31c12012-09-19 00:55:56 -04001999 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
Yongqiang Yang19c52462012-01-04 17:09:44 -05002000 return err;
2001}