]> git.ipfire.org Git - thirdparty/kernel/linux.git/blame - fs/ext4/resize.c
ext4: remove useless ext4_pin_inode()
[thirdparty/kernel/linux.git] / fs / ext4 / resize.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
ac27a0ec 2/*
617ba13b 3 * linux/fs/ext4/resize.c
ac27a0ec 4 *
617ba13b 5 * Support for resizing an ext4 filesystem while it is mounted.
ac27a0ec
DK
6 *
7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8 *
9 * This could probably be made into a module, because it is not often in use.
10 */
11
12
617ba13b 13#define EXT4FS_DEBUG
ac27a0ec 14
ac27a0ec
DK
15#include <linux/errno.h>
16#include <linux/slab.h>
17
3dcf5451 18#include "ext4_jbd2.h"
ac27a0ec 19
8f82f840
YY
20int ext4_resize_begin(struct super_block *sb)
21{
5f8c1093 22 struct ext4_sb_info *sbi = EXT4_SB(sb);
8f82f840
YY
23 int ret = 0;
24
25 if (!capable(CAP_SYS_RESOURCE))
26 return -EPERM;
27
011fa994
TT
28 /*
29 * If we are not using the primary superblock/GDT copy don't resize,
30 * because the user tools have no way of handling this. Probably a
31 * bad time to do it anyways.
32 */
5f8c1093 33 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
011fa994
TT
34 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
35 ext4_warning(sb, "won't resize using backup superblock at %llu",
36 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
37 return -EPERM;
38 }
39
ce723c31
YY
40 /*
41 * We are not allowed to do online-resizing on a filesystem mounted
42 * with error, because it can destroy the filesystem easily.
43 */
44 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
45 ext4_warning(sb, "There are errors in the filesystem, "
8d2ae1cb 46 "so online resizing is not allowed");
ce723c31
YY
47 return -EPERM;
48 }
49
9549a168
TT
50 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
51 &EXT4_SB(sb)->s_ext4_flags))
8f82f840
YY
52 ret = -EBUSY;
53
54 return ret;
55}
56
57void ext4_resize_end(struct super_block *sb)
58{
9549a168 59 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
4e857c58 60 smp_mb__after_atomic();
8f82f840
YY
61}
62
01f795f9
YY
63static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
64 ext4_group_t group) {
65 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
66 EXT4_DESC_PER_BLOCK_BITS(sb);
67}
68
69static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
70 ext4_group_t group) {
71 group = ext4_meta_bg_first_group(sb, group);
72 return ext4_group_first_block_no(sb, group);
73}
74
75static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
76 ext4_group_t group) {
77 ext4_grpblk_t overhead;
78 overhead = ext4_bg_num_gdb(sb, group);
79 if (ext4_bg_has_super(sb, group))
80 overhead += 1 +
81 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
82 return overhead;
83}
84
ac27a0ec
DK
85#define outside(b, first, last) ((b) < (first) || (b) >= (last))
86#define inside(b, first, last) ((b) >= (first) && (b) < (last))
87
88static int verify_group_input(struct super_block *sb,
617ba13b 89 struct ext4_new_group_data *input)
ac27a0ec 90{
617ba13b
MC
91 struct ext4_sb_info *sbi = EXT4_SB(sb);
92 struct ext4_super_block *es = sbi->s_es;
bd81d8ee 93 ext4_fsblk_t start = ext4_blocks_count(es);
617ba13b 94 ext4_fsblk_t end = start + input->blocks_count;
fd2d4291 95 ext4_group_t group = input->group;
617ba13b 96 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
b302ef2d
TT
97 unsigned overhead;
98 ext4_fsblk_t metaend;
ac27a0ec 99 struct buffer_head *bh = NULL;
3a5b2ecd 100 ext4_grpblk_t free_blocks_count, offset;
ac27a0ec
DK
101 int err = -EINVAL;
102
b302ef2d
TT
103 if (group != sbi->s_groups_count) {
104 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
105 input->group, sbi->s_groups_count);
106 return -EINVAL;
107 }
108
109 overhead = ext4_group_overhead_blocks(sb, group);
110 metaend = start + overhead;
d77147ff 111 input->free_clusters_count = free_blocks_count =
ac27a0ec
DK
112 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
113
114 if (test_opt(sb, DEBUG))
617ba13b 115 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
ac27a0ec 116 "(%d free, %u reserved)\n",
617ba13b 117 ext4_bg_has_super(sb, input->group) ? "normal" :
ac27a0ec
DK
118 "no-super", input->group, input->blocks_count,
119 free_blocks_count, input->reserved_blocks);
120
3a5b2ecd 121 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
b302ef2d 122 if (offset != 0)
12062ddd 123 ext4_warning(sb, "Last group not full");
ac27a0ec 124 else if (input->reserved_blocks > input->blocks_count / 5)
12062ddd 125 ext4_warning(sb, "Reserved blocks too high (%u)",
ac27a0ec
DK
126 input->reserved_blocks);
127 else if (free_blocks_count < 0)
12062ddd 128 ext4_warning(sb, "Bad blocks count %u",
ac27a0ec 129 input->blocks_count);
fb265c9c
TT
130 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
131 err = PTR_ERR(bh);
132 bh = NULL;
12062ddd 133 ext4_warning(sb, "Cannot read last block (%llu)",
ac27a0ec 134 end - 1);
fb265c9c 135 } else if (outside(input->block_bitmap, start, end))
12062ddd 136 ext4_warning(sb, "Block bitmap not in group (block %llu)",
1939e49a 137 (unsigned long long)input->block_bitmap);
ac27a0ec 138 else if (outside(input->inode_bitmap, start, end))
12062ddd 139 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
1939e49a 140 (unsigned long long)input->inode_bitmap);
ac27a0ec 141 else if (outside(input->inode_table, start, end) ||
2b2d6d01 142 outside(itend - 1, start, end))
12062ddd 143 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
1939e49a 144 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 145 else if (input->inode_bitmap == input->block_bitmap)
12062ddd 146 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
1939e49a 147 (unsigned long long)input->block_bitmap);
ac27a0ec 148 else if (inside(input->block_bitmap, input->inode_table, itend))
12062ddd
ES
149 ext4_warning(sb, "Block bitmap (%llu) in inode table "
150 "(%llu-%llu)",
1939e49a
RD
151 (unsigned long long)input->block_bitmap,
152 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 153 else if (inside(input->inode_bitmap, input->inode_table, itend))
12062ddd
ES
154 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
155 "(%llu-%llu)",
1939e49a
RD
156 (unsigned long long)input->inode_bitmap,
157 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 158 else if (inside(input->block_bitmap, start, metaend))
12062ddd 159 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
1939e49a
RD
160 (unsigned long long)input->block_bitmap,
161 start, metaend - 1);
ac27a0ec 162 else if (inside(input->inode_bitmap, start, metaend))
12062ddd 163 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
1939e49a
RD
164 (unsigned long long)input->inode_bitmap,
165 start, metaend - 1);
ac27a0ec 166 else if (inside(input->inode_table, start, metaend) ||
2b2d6d01 167 inside(itend - 1, start, metaend))
12062ddd
ES
168 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
169 "(%llu-%llu)",
1939e49a
RD
170 (unsigned long long)input->inode_table,
171 itend - 1, start, metaend - 1);
ac27a0ec
DK
172 else
173 err = 0;
174 brelse(bh);
175
176 return err;
177}
178
28c7bac0
YY
179/*
180 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
181 * group each time.
182 */
183struct ext4_new_flex_group_data {
184 struct ext4_new_group_data *groups; /* new_group_data for groups
185 in the flex group */
186 __u16 *bg_flags; /* block group flags of groups
187 in @groups */
188 ext4_group_t count; /* number of groups in @groups
189 */
190};
191
192/*
193 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
194 * @flexbg_size.
195 *
196 * Returns NULL on failure otherwise address of the allocated structure.
197 */
198static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
199{
200 struct ext4_new_flex_group_data *flex_gd;
201
202 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
203 if (flex_gd == NULL)
204 goto out3;
205
46901760 206 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
967ac8af 207 goto out2;
28c7bac0
YY
208 flex_gd->count = flexbg_size;
209
6da2ec56
KC
210 flex_gd->groups = kmalloc_array(flexbg_size,
211 sizeof(struct ext4_new_group_data),
212 GFP_NOFS);
28c7bac0
YY
213 if (flex_gd->groups == NULL)
214 goto out2;
215
6da2ec56
KC
216 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
217 GFP_NOFS);
28c7bac0
YY
218 if (flex_gd->bg_flags == NULL)
219 goto out1;
220
221 return flex_gd;
222
223out1:
224 kfree(flex_gd->groups);
225out2:
226 kfree(flex_gd);
227out3:
228 return NULL;
229}
230
231static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
232{
233 kfree(flex_gd->bg_flags);
234 kfree(flex_gd->groups);
235 kfree(flex_gd);
236}
237
3fbea4b3
YY
238/*
239 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
240 * and inode tables for a flex group.
241 *
242 * This function is used by 64bit-resize. Note that this function allocates
243 * group tables from the 1st group of groups contained by @flexgd, which may
244 * be a partial of a flex group.
245 *
246 * @sb: super block of fs to which the groups belongs
03c1c290
YY
247 *
248 * Returns 0 on a successful allocation of the metadata blocks in the
249 * block group.
3fbea4b3 250 */
03c1c290 251static int ext4_alloc_group_tables(struct super_block *sb,
3fbea4b3
YY
252 struct ext4_new_flex_group_data *flex_gd,
253 int flexbg_size)
254{
255 struct ext4_new_group_data *group_data = flex_gd->groups;
3fbea4b3
YY
256 ext4_fsblk_t start_blk;
257 ext4_fsblk_t last_blk;
258 ext4_group_t src_group;
259 ext4_group_t bb_index = 0;
260 ext4_group_t ib_index = 0;
261 ext4_group_t it_index = 0;
262 ext4_group_t group;
263 ext4_group_t last_group;
264 unsigned overhead;
b93c9535 265 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
d77147ff 266 int i;
3fbea4b3
YY
267
268 BUG_ON(flex_gd->count == 0 || group_data == NULL);
269
270 src_group = group_data[0].group;
271 last_group = src_group + flex_gd->count - 1;
272
273 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
274 (last_group & ~(flexbg_size - 1))));
275next_group:
276 group = group_data[0].group;
03c1c290
YY
277 if (src_group >= group_data[0].group + flex_gd->count)
278 return -ENOSPC;
3fbea4b3
YY
279 start_blk = ext4_group_first_block_no(sb, src_group);
280 last_blk = start_blk + group_data[src_group - group].blocks_count;
281
01f795f9 282 overhead = ext4_group_overhead_blocks(sb, src_group);
3fbea4b3
YY
283
284 start_blk += overhead;
285
3fbea4b3
YY
286 /* We collect contiguous blocks as much as possible. */
287 src_group++;
01f795f9
YY
288 for (; src_group <= last_group; src_group++) {
289 overhead = ext4_group_overhead_blocks(sb, src_group);
b93c9535 290 if (overhead == 0)
3fbea4b3
YY
291 last_blk += group_data[src_group - group].blocks_count;
292 else
293 break;
01f795f9 294 }
3fbea4b3
YY
295
296 /* Allocate block bitmaps */
297 for (; bb_index < flex_gd->count; bb_index++) {
298 if (start_blk >= last_blk)
299 goto next_group;
300 group_data[bb_index].block_bitmap = start_blk++;
bd86298e 301 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 302 group -= group_data[0].group;
d77147ff 303 group_data[group].mdata_blocks++;
b93c9535 304 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
305 }
306
307 /* Allocate inode bitmaps */
308 for (; ib_index < flex_gd->count; ib_index++) {
309 if (start_blk >= last_blk)
310 goto next_group;
311 group_data[ib_index].inode_bitmap = start_blk++;
bd86298e 312 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 313 group -= group_data[0].group;
d77147ff 314 group_data[group].mdata_blocks++;
b93c9535 315 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
316 }
317
318 /* Allocate inode tables */
319 for (; it_index < flex_gd->count; it_index++) {
b93c9535
TT
320 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
321 ext4_fsblk_t next_group_start;
322
323 if (start_blk + itb > last_blk)
3fbea4b3
YY
324 goto next_group;
325 group_data[it_index].inode_table = start_blk;
b93c9535
TT
326 group = ext4_get_group_number(sb, start_blk);
327 next_group_start = ext4_group_first_block_no(sb, group + 1);
3fbea4b3 328 group -= group_data[0].group;
3fbea4b3 329
b93c9535
TT
330 if (start_blk + itb > next_group_start) {
331 flex_gd->bg_flags[group + 1] &= uninit_mask;
332 overhead = start_blk + itb - next_group_start;
d77147ff 333 group_data[group + 1].mdata_blocks += overhead;
b93c9535
TT
334 itb -= overhead;
335 }
336
d77147ff 337 group_data[group].mdata_blocks += itb;
b93c9535 338 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
339 start_blk += EXT4_SB(sb)->s_itb_per_group;
340 }
341
d77147ff 342 /* Update free clusters count to exclude metadata blocks */
343 for (i = 0; i < flex_gd->count; i++) {
344 group_data[i].free_clusters_count -=
345 EXT4_NUM_B2C(EXT4_SB(sb),
346 group_data[i].mdata_blocks);
347 }
348
3fbea4b3
YY
349 if (test_opt(sb, DEBUG)) {
350 int i;
351 group = group_data[0].group;
352
353 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
354 "%d groups, flexbg size is %d:\n", flex_gd->count,
355 flexbg_size);
356
357 for (i = 0; i < flex_gd->count; i++) {
d77147ff 358 ext4_debug(
359 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
3fbea4b3
YY
360 ext4_bg_has_super(sb, group + i) ? "normal" :
361 "no-super", group + i,
362 group_data[i].blocks_count,
d77147ff 363 group_data[i].free_clusters_count,
364 group_data[i].mdata_blocks);
3fbea4b3
YY
365 }
366 }
03c1c290 367 return 0;
3fbea4b3
YY
368}
369
ac27a0ec 370static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
617ba13b 371 ext4_fsblk_t blk)
ac27a0ec
DK
372{
373 struct buffer_head *bh;
374 int err;
375
376 bh = sb_getblk(sb, blk);
aebf0243 377 if (unlikely(!bh))
860d21e2 378 return ERR_PTR(-ENOMEM);
5d601255 379 BUFFER_TRACE(bh, "get_write_access");
617ba13b 380 if ((err = ext4_journal_get_write_access(handle, bh))) {
ac27a0ec
DK
381 brelse(bh);
382 bh = ERR_PTR(err);
383 } else {
ac27a0ec
DK
384 memset(bh->b_data, 0, sb->s_blocksize);
385 set_buffer_uptodate(bh);
ac27a0ec
DK
386 }
387
388 return bh;
389}
390
14904107
ES
391/*
392 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
393 * If that fails, restart the transaction & regain write access for the
394 * buffer head which is used for block_bitmap modifications.
395 */
6d40bc5a 396static int extend_or_restart_transaction(handle_t *handle, int thresh)
14904107
ES
397{
398 int err;
399
0390131b 400 if (ext4_handle_has_enough_credits(handle, thresh))
14904107
ES
401 return 0;
402
403 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
404 if (err < 0)
405 return err;
406 if (err) {
6d40bc5a
YY
407 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
408 if (err)
14904107 409 return err;
2b2d6d01 410 }
14904107
ES
411
412 return 0;
413}
414
33afdcc5 415/*
d77147ff 416 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
33afdcc5
YY
417 *
418 * Helper function for ext4_setup_new_group_blocks() which set .
419 *
420 * @sb: super block
421 * @handle: journal handle
422 * @flex_gd: flex group data
423 */
424static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
425 struct ext4_new_flex_group_data *flex_gd,
d77147ff 426 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
33afdcc5 427{
d77147ff 428 struct ext4_sb_info *sbi = EXT4_SB(sb);
429 ext4_group_t count = last_cluster - first_cluster + 1;
33afdcc5
YY
430 ext4_group_t count2;
431
d77147ff 432 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
433 last_cluster);
434 for (count2 = count; count > 0;
435 count -= count2, first_cluster += count2) {
33afdcc5
YY
436 ext4_fsblk_t start;
437 struct buffer_head *bh;
438 ext4_group_t group;
439 int err;
440
d77147ff 441 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
442 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
33afdcc5
YY
443 group -= flex_gd->groups[0].group;
444
d77147ff 445 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
33afdcc5
YY
446 if (count2 > count)
447 count2 = count;
448
449 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
450 BUG_ON(flex_gd->count > 1);
451 continue;
452 }
453
454 err = extend_or_restart_transaction(handle, 1);
455 if (err)
456 return err;
457
458 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
aebf0243 459 if (unlikely(!bh))
860d21e2 460 return -ENOMEM;
33afdcc5 461
5d601255 462 BUFFER_TRACE(bh, "get_write_access");
33afdcc5 463 err = ext4_journal_get_write_access(handle, bh);
cea57941
VA
464 if (err) {
465 brelse(bh);
33afdcc5 466 return err;
cea57941 467 }
d77147ff 468 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
469 first_cluster, first_cluster - start, count2);
470 ext4_set_bits(bh->b_data, first_cluster - start, count2);
33afdcc5
YY
471
472 err = ext4_handle_dirty_metadata(handle, NULL, bh);
cea57941 473 brelse(bh);
33afdcc5
YY
474 if (unlikely(err))
475 return err;
33afdcc5
YY
476 }
477
478 return 0;
479}
480
481/*
482 * Set up the block and inode bitmaps, and the inode table for the new groups.
483 * This doesn't need to be part of the main transaction, since we are only
484 * changing blocks outside the actual filesystem. We still do journaling to
485 * ensure the recovery is correct in case of a failure just after resize.
486 * If any part of this fails, we simply abort the resize.
487 *
488 * setup_new_flex_group_blocks handles a flex group as follow:
489 * 1. copy super block and GDT, and initialize group tables if necessary.
490 * In this step, we only set bits in blocks bitmaps for blocks taken by
491 * super block and GDT.
492 * 2. allocate group tables in block bitmaps, that is, set bits in block
493 * bitmap for blocks taken by group tables.
494 */
495static int setup_new_flex_group_blocks(struct super_block *sb,
496 struct ext4_new_flex_group_data *flex_gd)
497{
498 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
499 ext4_fsblk_t start;
500 ext4_fsblk_t block;
501 struct ext4_sb_info *sbi = EXT4_SB(sb);
502 struct ext4_super_block *es = sbi->s_es;
503 struct ext4_new_group_data *group_data = flex_gd->groups;
504 __u16 *bg_flags = flex_gd->bg_flags;
505 handle_t *handle;
506 ext4_group_t group, count;
507 struct buffer_head *bh = NULL;
508 int reserved_gdb, i, j, err = 0, err2;
01f795f9 509 int meta_bg;
33afdcc5
YY
510
511 BUG_ON(!flex_gd->count || !group_data ||
512 group_data[0].group != sbi->s_groups_count);
513
514 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
e2b911c5 515 meta_bg = ext4_has_feature_meta_bg(sb);
33afdcc5
YY
516
517 /* This transaction may be extended/restarted along the way */
9924a92a 518 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
33afdcc5
YY
519 if (IS_ERR(handle))
520 return PTR_ERR(handle);
521
522 group = group_data[0].group;
523 for (i = 0; i < flex_gd->count; i++, group++) {
524 unsigned long gdblocks;
01f795f9 525 ext4_grpblk_t overhead;
33afdcc5
YY
526
527 gdblocks = ext4_bg_num_gdb(sb, group);
528 start = ext4_group_first_block_no(sb, group);
529
01f795f9 530 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
6df935ad
YY
531 goto handle_itb;
532
01f795f9
YY
533 if (meta_bg == 1) {
534 ext4_group_t first_group;
535 first_group = ext4_meta_bg_first_group(sb, group);
536 if (first_group != group + 1 &&
537 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
538 goto handle_itb;
539 }
540
541 block = start + ext4_bg_has_super(sb, group);
33afdcc5 542 /* Copy all of the GDT blocks into the backup in this group */
01f795f9 543 for (j = 0; j < gdblocks; j++, block++) {
33afdcc5
YY
544 struct buffer_head *gdb;
545
546 ext4_debug("update backup group %#04llx\n", block);
547 err = extend_or_restart_transaction(handle, 1);
548 if (err)
549 goto out;
550
551 gdb = sb_getblk(sb, block);
aebf0243 552 if (unlikely(!gdb)) {
860d21e2 553 err = -ENOMEM;
33afdcc5
YY
554 goto out;
555 }
556
5d601255 557 BUFFER_TRACE(gdb, "get_write_access");
33afdcc5
YY
558 err = ext4_journal_get_write_access(handle, gdb);
559 if (err) {
560 brelse(gdb);
561 goto out;
562 }
563 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
564 gdb->b_size);
565 set_buffer_uptodate(gdb);
566
567 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
568 if (unlikely(err)) {
569 brelse(gdb);
570 goto out;
571 }
572 brelse(gdb);
573 }
574
575 /* Zero out all of the reserved backup group descriptor
576 * table blocks
577 */
578 if (ext4_bg_has_super(sb, group)) {
579 err = sb_issue_zeroout(sb, gdblocks + start + 1,
580 reserved_gdb, GFP_NOFS);
581 if (err)
582 goto out;
583 }
584
6df935ad 585handle_itb:
33afdcc5
YY
586 /* Initialize group tables of the grop @group */
587 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
588 goto handle_bb;
589
590 /* Zero out all of the inode table blocks */
591 block = group_data[i].inode_table;
592 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
593 block, sbi->s_itb_per_group);
594 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
595 GFP_NOFS);
596 if (err)
597 goto out;
598
599handle_bb:
600 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
601 goto handle_ib;
602
603 /* Initialize block bitmap of the @group */
604 block = group_data[i].block_bitmap;
605 err = extend_or_restart_transaction(handle, 1);
606 if (err)
607 goto out;
608
609 bh = bclean(handle, sb, block);
610 if (IS_ERR(bh)) {
611 err = PTR_ERR(bh);
612 goto out;
613 }
01f795f9
YY
614 overhead = ext4_group_overhead_blocks(sb, group);
615 if (overhead != 0) {
33afdcc5
YY
616 ext4_debug("mark backup superblock %#04llx (+0)\n",
617 start);
d77147ff 618 ext4_set_bits(bh->b_data, 0,
619 EXT4_NUM_B2C(sbi, overhead));
33afdcc5 620 }
d77147ff 621 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
33afdcc5
YY
622 sb->s_blocksize * 8, bh->b_data);
623 err = ext4_handle_dirty_metadata(handle, NULL, bh);
9e402893 624 brelse(bh);
33afdcc5
YY
625 if (err)
626 goto out;
33afdcc5
YY
627
628handle_ib:
629 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
630 continue;
631
632 /* Initialize inode bitmap of the @group */
633 block = group_data[i].inode_bitmap;
634 err = extend_or_restart_transaction(handle, 1);
635 if (err)
636 goto out;
637 /* Mark unused entries in inode bitmap used */
638 bh = bclean(handle, sb, block);
639 if (IS_ERR(bh)) {
640 err = PTR_ERR(bh);
641 goto out;
642 }
643
644 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
645 sb->s_blocksize * 8, bh->b_data);
646 err = ext4_handle_dirty_metadata(handle, NULL, bh);
9e402893 647 brelse(bh);
33afdcc5
YY
648 if (err)
649 goto out;
33afdcc5 650 }
33afdcc5
YY
651
652 /* Mark group tables in block bitmap */
653 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
654 count = group_table_count[j];
655 start = (&group_data[0].block_bitmap)[j];
656 block = start;
657 for (i = 1; i < flex_gd->count; i++) {
658 block += group_table_count[j];
659 if (block == (&group_data[i].block_bitmap)[j]) {
660 count += group_table_count[j];
661 continue;
662 }
663 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 664 flex_gd,
665 EXT4_B2C(sbi, start),
666 EXT4_B2C(sbi,
667 start + count
668 - 1));
33afdcc5
YY
669 if (err)
670 goto out;
671 count = group_table_count[j];
b93c9535 672 start = (&group_data[i].block_bitmap)[j];
33afdcc5
YY
673 block = start;
674 }
675
676 if (count) {
677 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 678 flex_gd,
679 EXT4_B2C(sbi, start),
680 EXT4_B2C(sbi,
681 start + count
682 - 1));
33afdcc5
YY
683 if (err)
684 goto out;
685 }
686 }
687
688out:
33afdcc5
YY
689 err2 = ext4_journal_stop(handle);
690 if (err2 && !err)
691 err = err2;
692
693 return err;
694}
695
ac27a0ec
DK
696/*
697 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
617ba13b 698 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
ac27a0ec
DK
699 * calling this for the first time. In a sparse filesystem it will be the
700 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
701 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
702 */
617ba13b 703static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
ac27a0ec
DK
704 unsigned *five, unsigned *seven)
705{
706 unsigned *min = three;
707 int mult = 3;
708 unsigned ret;
709
e2b911c5 710 if (!ext4_has_feature_sparse_super(sb)) {
ac27a0ec
DK
711 ret = *min;
712 *min += 1;
713 return ret;
714 }
715
716 if (*five < *min) {
717 min = five;
718 mult = 5;
719 }
720 if (*seven < *min) {
721 min = seven;
722 mult = 7;
723 }
724
725 ret = *min;
726 *min *= mult;
727
728 return ret;
729}
730
731/*
732 * Check that all of the backup GDT blocks are held in the primary GDT block.
733 * It is assumed that they are stored in group order. Returns the number of
734 * groups in current filesystem that have BACKUPS, or -ve error code.
735 */
736static int verify_reserved_gdb(struct super_block *sb,
c72df9f9 737 ext4_group_t end,
ac27a0ec
DK
738 struct buffer_head *primary)
739{
617ba13b 740 const ext4_fsblk_t blk = primary->b_blocknr;
ac27a0ec
DK
741 unsigned three = 1;
742 unsigned five = 5;
743 unsigned seven = 7;
744 unsigned grp;
745 __le32 *p = (__le32 *)primary->b_data;
746 int gdbackups = 0;
747
617ba13b 748 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
bd81d8ee
LV
749 if (le32_to_cpu(*p++) !=
750 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
12062ddd 751 ext4_warning(sb, "reserved GDT %llu"
2ae02107 752 " missing grp %d (%llu)",
ac27a0ec 753 blk, grp,
bd81d8ee
LV
754 grp *
755 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
756 blk);
ac27a0ec
DK
757 return -EINVAL;
758 }
617ba13b 759 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
ac27a0ec
DK
760 return -EFBIG;
761 }
762
763 return gdbackups;
764}
765
766/*
767 * Called when we need to bring a reserved group descriptor table block into
768 * use from the resize inode. The primary copy of the new GDT block currently
769 * is an indirect block (under the double indirect block in the resize inode).
770 * The new backup GDT blocks will be stored as leaf blocks in this indirect
771 * block, in group order. Even though we know all the block numbers we need,
772 * we check to ensure that the resize inode has actually reserved these blocks.
773 *
774 * Don't need to update the block bitmaps because the blocks are still in use.
775 *
776 * We get all of the error cases out of the way, so that we are sure to not
777 * fail once we start modifying the data on disk, because JBD has no rollback.
778 */
779static int add_new_gdb(handle_t *handle, struct inode *inode,
2f919710 780 ext4_group_t group)
ac27a0ec
DK
781{
782 struct super_block *sb = inode->i_sb;
617ba13b 783 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
2f919710 784 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
617ba13b 785 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
fb265c9c
TT
786 struct buffer_head **o_group_desc, **n_group_desc = NULL;
787 struct buffer_head *dind = NULL;
788 struct buffer_head *gdb_bh = NULL;
ac27a0ec 789 int gdbackups;
fb265c9c 790 struct ext4_iloc iloc = { .bh = NULL };
ac27a0ec
DK
791 __le32 *data;
792 int err;
793
794 if (test_opt(sb, DEBUG))
795 printk(KERN_DEBUG
617ba13b 796 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
ac27a0ec
DK
797 gdb_num);
798
fb265c9c
TT
799 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
800 if (IS_ERR(gdb_bh))
801 return PTR_ERR(gdb_bh);
ac27a0ec 802
c72df9f9 803 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
2f919710 804 if (gdbackups < 0) {
ac27a0ec 805 err = gdbackups;
fb265c9c 806 goto errout;
ac27a0ec
DK
807 }
808
617ba13b 809 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
fb265c9c
TT
810 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
811 if (IS_ERR(dind)) {
812 err = PTR_ERR(dind);
813 dind = NULL;
814 goto errout;
ac27a0ec
DK
815 }
816
817 data = (__le32 *)dind->b_data;
617ba13b 818 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
12062ddd 819 ext4_warning(sb, "new group %u GDT block %llu not reserved",
2f919710 820 group, gdblock);
ac27a0ec 821 err = -EINVAL;
fb265c9c 822 goto errout;
ac27a0ec
DK
823 }
824
5d601255 825 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
b4097142
TT
826 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
827 if (unlikely(err))
fb265c9c 828 goto errout;
ac27a0ec 829
5d601255 830 BUFFER_TRACE(gdb_bh, "get_write_access");
2f919710 831 err = ext4_journal_get_write_access(handle, gdb_bh);
b4097142 832 if (unlikely(err))
fb265c9c 833 goto errout;
ac27a0ec 834
5d601255 835 BUFFER_TRACE(dind, "get_write_access");
b4097142
TT
836 err = ext4_journal_get_write_access(handle, dind);
837 if (unlikely(err))
838 ext4_std_error(sb, err);
ac27a0ec 839
617ba13b 840 /* ext4_reserve_inode_write() gets a reference on the iloc */
b4097142
TT
841 err = ext4_reserve_inode_write(handle, inode, &iloc);
842 if (unlikely(err))
fb265c9c 843 goto errout;
ac27a0ec 844
f18a5f21
TT
845 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
846 sizeof(struct buffer_head *),
847 GFP_NOFS);
ac27a0ec
DK
848 if (!n_group_desc) {
849 err = -ENOMEM;
f18a5f21
TT
850 ext4_warning(sb, "not enough memory for %lu groups",
851 gdb_num + 1);
fb265c9c 852 goto errout;
ac27a0ec
DK
853 }
854
855 /*
856 * Finally, we have all of the possible failures behind us...
857 *
858 * Remove new GDT block from inode double-indirect block and clear out
859 * the new GDT block for use (which also "frees" the backup GDT blocks
860 * from the reserved inode). We don't need to change the bitmaps for
861 * these blocks, because they are marked as in-use from being in the
862 * reserved inode, and will become GDT blocks (primary and backup).
863 */
617ba13b 864 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
b4097142
TT
865 err = ext4_handle_dirty_metadata(handle, NULL, dind);
866 if (unlikely(err)) {
867 ext4_std_error(sb, err);
fb265c9c 868 goto errout;
b4097142 869 }
d77147ff 870 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
871 (9 - EXT4_SB(sb)->s_cluster_bits);
617ba13b 872 ext4_mark_iloc_dirty(handle, inode, &iloc);
2f919710
YY
873 memset(gdb_bh->b_data, 0, sb->s_blocksize);
874 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
b4097142
TT
875 if (unlikely(err)) {
876 ext4_std_error(sb, err);
fb265c9c 877 goto errout;
b4097142
TT
878 }
879 brelse(dind);
ac27a0ec 880
617ba13b 881 o_group_desc = EXT4_SB(sb)->s_group_desc;
ac27a0ec 882 memcpy(n_group_desc, o_group_desc,
617ba13b 883 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
2f919710 884 n_group_desc[gdb_num] = gdb_bh;
617ba13b
MC
885 EXT4_SB(sb)->s_group_desc = n_group_desc;
886 EXT4_SB(sb)->s_gdb_count++;
b93b41d4 887 kvfree(o_group_desc);
ac27a0ec 888
e8546d06 889 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
b50924c2 890 err = ext4_handle_dirty_super(handle, sb);
b4097142
TT
891 if (err)
892 ext4_std_error(sb, err);
b4097142 893 return err;
fb265c9c 894errout:
b93b41d4 895 kvfree(n_group_desc);
ac27a0ec 896 brelse(iloc.bh);
ac27a0ec 897 brelse(dind);
2f919710 898 brelse(gdb_bh);
ac27a0ec 899
617ba13b 900 ext4_debug("leaving with error %d\n", err);
ac27a0ec
DK
901 return err;
902}
903
01f795f9
YY
904/*
905 * add_new_gdb_meta_bg is the sister of add_new_gdb.
906 */
907static int add_new_gdb_meta_bg(struct super_block *sb,
908 handle_t *handle, ext4_group_t group) {
909 ext4_fsblk_t gdblock;
910 struct buffer_head *gdb_bh;
911 struct buffer_head **o_group_desc, **n_group_desc;
912 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
913 int err;
914
915 gdblock = ext4_meta_bg_first_block_no(sb, group) +
916 ext4_bg_has_super(sb, group);
fb265c9c
TT
917 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
918 if (IS_ERR(gdb_bh))
919 return PTR_ERR(gdb_bh);
01f795f9
YY
920 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
921 sizeof(struct buffer_head *),
922 GFP_NOFS);
923 if (!n_group_desc) {
61a9c11e 924 brelse(gdb_bh);
01f795f9
YY
925 err = -ENOMEM;
926 ext4_warning(sb, "not enough memory for %lu groups",
927 gdb_num + 1);
928 return err;
929 }
930
931 o_group_desc = EXT4_SB(sb)->s_group_desc;
932 memcpy(n_group_desc, o_group_desc,
933 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
934 n_group_desc[gdb_num] = gdb_bh;
935 EXT4_SB(sb)->s_group_desc = n_group_desc;
936 EXT4_SB(sb)->s_gdb_count++;
b93b41d4 937 kvfree(o_group_desc);
5d601255 938 BUFFER_TRACE(gdb_bh, "get_write_access");
01f795f9 939 err = ext4_journal_get_write_access(handle, gdb_bh);
01f795f9
YY
940 return err;
941}
942
ac27a0ec
DK
943/*
944 * Called when we are adding a new group which has a backup copy of each of
945 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
946 * We need to add these reserved backup GDT blocks to the resize inode, so
947 * that they are kept for future resizing and not allocated to files.
948 *
949 * Each reserved backup GDT block will go into a different indirect block.
950 * The indirect blocks are actually the primary reserved GDT blocks,
951 * so we know in advance what their block numbers are. We only get the
952 * double-indirect block to verify it is pointing to the primary reserved
953 * GDT blocks so we don't overwrite a data block by accident. The reserved
954 * backup GDT blocks are stored in their reserved primary GDT block.
955 */
956static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
668f4dc5 957 ext4_group_t group)
ac27a0ec
DK
958{
959 struct super_block *sb = inode->i_sb;
617ba13b 960 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
d77147ff 961 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
ac27a0ec
DK
962 struct buffer_head **primary;
963 struct buffer_head *dind;
617ba13b
MC
964 struct ext4_iloc iloc;
965 ext4_fsblk_t blk;
ac27a0ec
DK
966 __le32 *data, *end;
967 int gdbackups = 0;
968 int res, i;
969 int err;
970
6da2ec56 971 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
ac27a0ec
DK
972 if (!primary)
973 return -ENOMEM;
974
617ba13b 975 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
fb265c9c
TT
976 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
977 if (IS_ERR(dind)) {
978 err = PTR_ERR(dind);
979 dind = NULL;
ac27a0ec
DK
980 goto exit_free;
981 }
982
617ba13b 983 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
94460093
JB
984 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
985 EXT4_ADDR_PER_BLOCK(sb));
617ba13b 986 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
ac27a0ec
DK
987
988 /* Get each reserved primary GDT block and verify it holds backups */
989 for (res = 0; res < reserved_gdb; res++, blk++) {
990 if (le32_to_cpu(*data) != blk) {
12062ddd 991 ext4_warning(sb, "reserved block %llu"
ac27a0ec
DK
992 " not at offset %ld",
993 blk,
994 (long)(data - (__le32 *)dind->b_data));
995 err = -EINVAL;
996 goto exit_bh;
997 }
fb265c9c
TT
998 primary[res] = ext4_sb_bread(sb, blk, 0);
999 if (IS_ERR(primary[res])) {
1000 err = PTR_ERR(primary[res]);
1001 primary[res] = NULL;
ac27a0ec
DK
1002 goto exit_bh;
1003 }
c72df9f9
YY
1004 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1005 if (gdbackups < 0) {
ac27a0ec
DK
1006 brelse(primary[res]);
1007 err = gdbackups;
1008 goto exit_bh;
1009 }
1010 if (++data >= end)
1011 data = (__le32 *)dind->b_data;
1012 }
1013
1014 for (i = 0; i < reserved_gdb; i++) {
5d601255 1015 BUFFER_TRACE(primary[i], "get_write_access");
37be2f59 1016 if ((err = ext4_journal_get_write_access(handle, primary[i])))
ac27a0ec 1017 goto exit_bh;
ac27a0ec
DK
1018 }
1019
617ba13b 1020 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
ac27a0ec
DK
1021 goto exit_bh;
1022
1023 /*
1024 * Finally we can add each of the reserved backup GDT blocks from
1025 * the new group to its reserved primary GDT block.
1026 */
668f4dc5 1027 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1028 for (i = 0; i < reserved_gdb; i++) {
1029 int err2;
1030 data = (__le32 *)primary[i]->b_data;
1031 /* printk("reserving backup %lu[%u] = %lu\n",
1032 primary[i]->b_blocknr, gdbackups,
1033 blk + primary[i]->b_blocknr); */
1034 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
0390131b 1035 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
ac27a0ec
DK
1036 if (!err)
1037 err = err2;
1038 }
d77147ff 1039
1040 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
617ba13b 1041 ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec
DK
1042
1043exit_bh:
1044 while (--res >= 0)
1045 brelse(primary[res]);
1046 brelse(dind);
1047
1048exit_free:
1049 kfree(primary);
1050
1051 return err;
1052}
1053
1054/*
617ba13b 1055 * Update the backup copies of the ext4 metadata. These don't need to be part
ac27a0ec
DK
1056 * of the main resize transaction, because e2fsck will re-write them if there
1057 * is a problem (basically only OOM will cause a problem). However, we
1058 * _should_ update the backups if possible, in case the primary gets trashed
1059 * for some reason and we need to run e2fsck from a backup superblock. The
1060 * important part is that the new block and inode counts are in the backup
1061 * superblocks, and the location of the new group metadata in the GDT backups.
1062 *
32ed5058
TT
1063 * We do not need take the s_resize_lock for this, because these
1064 * blocks are not otherwise touched by the filesystem code when it is
1065 * mounted. We don't need to worry about last changing from
1066 * sbi->s_groups_count, because the worst that can happen is that we
1067 * do not copy the full number of backups at this time. The resize
1068 * which changed s_groups_count will backup again.
ac27a0ec 1069 */
904dad47 1070static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
01f795f9 1071 int size, int meta_bg)
ac27a0ec 1072{
617ba13b 1073 struct ext4_sb_info *sbi = EXT4_SB(sb);
01f795f9 1074 ext4_group_t last;
617ba13b 1075 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1076 unsigned three = 1;
1077 unsigned five = 5;
1078 unsigned seven = 7;
01f795f9 1079 ext4_group_t group = 0;
ac27a0ec
DK
1080 int rest = sb->s_blocksize - size;
1081 handle_t *handle;
1082 int err = 0, err2;
1083
9924a92a 1084 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
ac27a0ec
DK
1085 if (IS_ERR(handle)) {
1086 group = 1;
1087 err = PTR_ERR(handle);
1088 goto exit_err;
1089 }
1090
01f795f9
YY
1091 if (meta_bg == 0) {
1092 group = ext4_list_backups(sb, &three, &five, &seven);
1093 last = sbi->s_groups_count;
1094 } else {
904dad47 1095 group = ext4_get_group_number(sb, blk_off) + 1;
01f795f9
YY
1096 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1097 }
1098
1099 while (group < sbi->s_groups_count) {
ac27a0ec 1100 struct buffer_head *bh;
01f795f9 1101 ext4_fsblk_t backup_block;
ac27a0ec
DK
1102
1103 /* Out of journal space, and can't get more - abort - so sad */
0390131b
FM
1104 if (ext4_handle_valid(handle) &&
1105 handle->h_buffer_credits == 0 &&
617ba13b
MC
1106 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1107 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
ac27a0ec
DK
1108 break;
1109
01f795f9 1110 if (meta_bg == 0)
9378c676 1111 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
01f795f9
YY
1112 else
1113 backup_block = (ext4_group_first_block_no(sb, group) +
1114 ext4_bg_has_super(sb, group));
1115
1116 bh = sb_getblk(sb, backup_block);
aebf0243 1117 if (unlikely(!bh)) {
860d21e2 1118 err = -ENOMEM;
ac27a0ec
DK
1119 break;
1120 }
01f795f9
YY
1121 ext4_debug("update metadata backup %llu(+%llu)\n",
1122 backup_block, backup_block -
1123 ext4_group_first_block_no(sb, group));
5d601255 1124 BUFFER_TRACE(bh, "get_write_access");
ea0abbb6
VA
1125 if ((err = ext4_journal_get_write_access(handle, bh))) {
1126 brelse(bh);
ac27a0ec 1127 break;
ea0abbb6 1128 }
ac27a0ec
DK
1129 lock_buffer(bh);
1130 memcpy(bh->b_data, data, size);
1131 if (rest)
1132 memset(bh->b_data + size, 0, rest);
1133 set_buffer_uptodate(bh);
1134 unlock_buffer(bh);
b4097142
TT
1135 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1136 if (unlikely(err))
1137 ext4_std_error(sb, err);
ac27a0ec 1138 brelse(bh);
01f795f9
YY
1139
1140 if (meta_bg == 0)
1141 group = ext4_list_backups(sb, &three, &five, &seven);
1142 else if (group == last)
1143 break;
1144 else
1145 group = last;
ac27a0ec 1146 }
617ba13b 1147 if ((err2 = ext4_journal_stop(handle)) && !err)
ac27a0ec
DK
1148 err = err2;
1149
1150 /*
1151 * Ugh! Need to have e2fsck write the backup copies. It is too
1152 * late to revert the resize, we shouldn't fail just because of
1153 * the backup copies (they are only needed in case of corruption).
1154 *
1155 * However, if we got here we have a journal problem too, so we
1156 * can't really start a transaction to mark the superblock.
1157 * Chicken out and just set the flag on the hope it will be written
1158 * to disk, and if not - we will simply wait until next fsck.
1159 */
1160exit_err:
1161 if (err) {
12062ddd 1162 ext4_warning(sb, "can't update backup for group %u (err %d), "
ac27a0ec 1163 "forcing fsck on next reboot", group, err);
617ba13b
MC
1164 sbi->s_mount_state &= ~EXT4_VALID_FS;
1165 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
ac27a0ec
DK
1166 mark_buffer_dirty(sbi->s_sbh);
1167 }
1168}
1169
bb08c1e7
YY
1170/*
1171 * ext4_add_new_descs() adds @count group descriptor of groups
1172 * starting at @group
1173 *
1174 * @handle: journal handle
1175 * @sb: super block
1176 * @group: the group no. of the first group desc to be added
1177 * @resize_inode: the resize inode
1178 * @count: number of group descriptors to be added
1179 */
1180static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1181 ext4_group_t group, struct inode *resize_inode,
1182 ext4_group_t count)
1183{
1184 struct ext4_sb_info *sbi = EXT4_SB(sb);
1185 struct ext4_super_block *es = sbi->s_es;
1186 struct buffer_head *gdb_bh;
1187 int i, gdb_off, gdb_num, err = 0;
01f795f9 1188 int meta_bg;
bb08c1e7 1189
e2b911c5 1190 meta_bg = ext4_has_feature_meta_bg(sb);
bb08c1e7
YY
1191 for (i = 0; i < count; i++, group++) {
1192 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1193 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1194
1195 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1196 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1197
1198 /*
1199 * We will only either add reserved group blocks to a backup group
1200 * or remove reserved blocks for the first group in a new group block.
1201 * Doing both would be mean more complex code, and sane people don't
1202 * use non-sparse filesystems anymore. This is already checked above.
1203 */
1204 if (gdb_off) {
1205 gdb_bh = sbi->s_group_desc[gdb_num];
5d601255 1206 BUFFER_TRACE(gdb_bh, "get_write_access");
bb08c1e7
YY
1207 err = ext4_journal_get_write_access(handle, gdb_bh);
1208
1209 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1210 err = reserve_backup_gdb(handle, resize_inode, group);
01f795f9
YY
1211 } else if (meta_bg != 0) {
1212 err = add_new_gdb_meta_bg(sb, handle, group);
1213 } else {
bb08c1e7 1214 err = add_new_gdb(handle, resize_inode, group);
01f795f9 1215 }
bb08c1e7
YY
1216 if (err)
1217 break;
1218 }
1219 return err;
1220}
1221
41a246d1
DW
1222static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1223{
1224 struct buffer_head *bh = sb_getblk(sb, block);
aebf0243 1225 if (unlikely(!bh))
41a246d1 1226 return NULL;
7f1468d1
DM
1227 if (!bh_uptodate_or_lock(bh)) {
1228 if (bh_submit_read(bh) < 0) {
1229 brelse(bh);
1230 return NULL;
1231 }
41a246d1 1232 }
41a246d1
DW
1233
1234 return bh;
1235}
1236
1237static int ext4_set_bitmap_checksums(struct super_block *sb,
1238 ext4_group_t group,
1239 struct ext4_group_desc *gdp,
1240 struct ext4_new_group_data *group_data)
1241{
1242 struct buffer_head *bh;
1243
9aa5d32b 1244 if (!ext4_has_metadata_csum(sb))
41a246d1
DW
1245 return 0;
1246
1247 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1248 if (!bh)
1249 return -EIO;
1250 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1251 EXT4_INODES_PER_GROUP(sb) / 8);
1252 brelse(bh);
1253
fa77dcfa
DW
1254 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1255 if (!bh)
1256 return -EIO;
79f1ba49 1257 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
fa77dcfa
DW
1258 brelse(bh);
1259
41a246d1
DW
1260 return 0;
1261}
1262
083f5b24
YY
1263/*
1264 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1265 */
1266static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1267 struct ext4_new_flex_group_data *flex_gd)
1268{
1269 struct ext4_new_group_data *group_data = flex_gd->groups;
1270 struct ext4_group_desc *gdp;
1271 struct ext4_sb_info *sbi = EXT4_SB(sb);
1272 struct buffer_head *gdb_bh;
1273 ext4_group_t group;
1274 __u16 *bg_flags = flex_gd->bg_flags;
1275 int i, gdb_off, gdb_num, err = 0;
d77147ff 1276
083f5b24
YY
1277
1278 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1279 group = group_data->group;
1280
1281 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1282 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1283
1284 /*
1285 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1286 */
1287 gdb_bh = sbi->s_group_desc[gdb_num];
1288 /* Update group descriptor block for new group */
2716b802 1289 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
083f5b24
YY
1290 gdb_off * EXT4_DESC_SIZE(sb));
1291
1292 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1293 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1294 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
41a246d1
DW
1295 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1296 if (err) {
1297 ext4_std_error(sb, err);
1298 break;
1299 }
1300
083f5b24
YY
1301 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1302 ext4_free_group_clusters_set(sb, gdp,
d77147ff 1303 group_data->free_clusters_count);
083f5b24 1304 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
93f90526
TT
1305 if (ext4_has_group_desc_csum(sb))
1306 ext4_itable_unused_set(sb, gdp,
1307 EXT4_INODES_PER_GROUP(sb));
083f5b24 1308 gdp->bg_flags = cpu_to_le16(*bg_flags);
feb0ab32 1309 ext4_group_desc_csum_set(sb, group, gdp);
083f5b24
YY
1310
1311 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1312 if (unlikely(err)) {
1313 ext4_std_error(sb, err);
1314 break;
1315 }
1316
1317 /*
1318 * We can allocate memory for mb_alloc based on the new group
1319 * descriptor
1320 */
1321 err = ext4_mb_add_groupinfo(sb, group, gdp);
1322 if (err)
1323 break;
1324 }
1325 return err;
1326}
1327
2e10e2f2
YY
1328/*
1329 * ext4_update_super() updates the super block so that the newly added
1330 * groups can be seen by the filesystem.
1331 *
1332 * @sb: super block
1333 * @flex_gd: new added groups
1334 */
1335static void ext4_update_super(struct super_block *sb,
1336 struct ext4_new_flex_group_data *flex_gd)
1337{
1338 ext4_fsblk_t blocks_count = 0;
1339 ext4_fsblk_t free_blocks = 0;
1340 ext4_fsblk_t reserved_blocks = 0;
1341 struct ext4_new_group_data *group_data = flex_gd->groups;
1342 struct ext4_sb_info *sbi = EXT4_SB(sb);
1343 struct ext4_super_block *es = sbi->s_es;
8a991849 1344 int i;
2e10e2f2
YY
1345
1346 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1347 /*
1348 * Make the new blocks and inodes valid next. We do this before
1349 * increasing the group count so that once the group is enabled,
1350 * all of its blocks and inodes are already valid.
1351 *
1352 * We always allocate group-by-group, then block-by-block or
1353 * inode-by-inode within a group, so enabling these
1354 * blocks/inodes before the group is live won't actually let us
1355 * allocate the new space yet.
1356 */
1357 for (i = 0; i < flex_gd->count; i++) {
1358 blocks_count += group_data[i].blocks_count;
d77147ff 1359 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
2e10e2f2
YY
1360 }
1361
1362 reserved_blocks = ext4_r_blocks_count(es) * 100;
01f795f9 1363 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
2e10e2f2
YY
1364 reserved_blocks *= blocks_count;
1365 do_div(reserved_blocks, 100);
1366
1367 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
636d7e2e 1368 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
2e10e2f2
YY
1369 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1370 flex_gd->count);
636d7e2e
DW
1371 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1372 flex_gd->count);
2e10e2f2 1373
01f795f9 1374 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
2e10e2f2
YY
1375 /*
1376 * We need to protect s_groups_count against other CPUs seeing
1377 * inconsistent state in the superblock.
1378 *
1379 * The precise rules we use are:
1380 *
1381 * * Writers must perform a smp_wmb() after updating all
1382 * dependent data and before modifying the groups count
1383 *
1384 * * Readers must perform an smp_rmb() after reading the groups
1385 * count and before reading any dependent data.
1386 *
1387 * NB. These rules can be relaxed when checking the group count
1388 * while freeing data, as we can only allocate from a block
1389 * group after serialising against the group count, and we can
1390 * only then free after serialising in turn against that
1391 * allocation.
1392 */
1393 smp_wmb();
1394
1395 /* Update the global fs size fields */
1396 sbi->s_groups_count += flex_gd->count;
c5c72d81
TT
1397 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1398 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2e10e2f2
YY
1399
1400 /* Update the reserved block counts only once the new group is
1401 * active. */
1402 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1403 reserved_blocks);
1404
1405 /* Update the free space counts */
1406 percpu_counter_add(&sbi->s_freeclusters_counter,
810da240 1407 EXT4_NUM_B2C(sbi, free_blocks));
2e10e2f2
YY
1408 percpu_counter_add(&sbi->s_freeinodes_counter,
1409 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1410
01f795f9
YY
1411 ext4_debug("free blocks count %llu",
1412 percpu_counter_read(&sbi->s_freeclusters_counter));
e2b911c5 1413 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
2e10e2f2
YY
1414 ext4_group_t flex_group;
1415 flex_group = ext4_flex_group(sbi, group_data[0].group);
90ba983f
TT
1416 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1417 &sbi->s_flex_groups[flex_group].free_clusters);
2e10e2f2
YY
1418 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1419 &sbi->s_flex_groups[flex_group].free_inodes);
1420 }
1421
952fc18e
TT
1422 /*
1423 * Update the fs overhead information
1424 */
1425 ext4_calculate_overhead(sb);
1426
2e10e2f2
YY
1427 if (test_opt(sb, DEBUG))
1428 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1429 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1430 blocks_count, free_blocks, reserved_blocks);
1431}
1432
4bac1f8c
YY
1433/* Add a flex group to an fs. Ensure we handle all possible error conditions
1434 * _before_ we start modifying the filesystem, because we cannot abort the
1435 * transaction and not have it write the data to disk.
1436 */
1437static int ext4_flex_group_add(struct super_block *sb,
1438 struct inode *resize_inode,
1439 struct ext4_new_flex_group_data *flex_gd)
1440{
1441 struct ext4_sb_info *sbi = EXT4_SB(sb);
1442 struct ext4_super_block *es = sbi->s_es;
1443 ext4_fsblk_t o_blocks_count;
1444 ext4_grpblk_t last;
1445 ext4_group_t group;
1446 handle_t *handle;
1447 unsigned reserved_gdb;
1448 int err = 0, err2 = 0, credit;
1449
1450 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1451
1452 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1453 o_blocks_count = ext4_blocks_count(es);
1454 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1455 BUG_ON(last);
1456
1457 err = setup_new_flex_group_blocks(sb, flex_gd);
1458 if (err)
1459 goto exit;
1460 /*
1461 * We will always be modifying at least the superblock and GDT
2c869b26 1462 * blocks. If we are adding a group past the last current GDT block,
4bac1f8c
YY
1463 * we will also modify the inode and the dindirect block. If we
1464 * are adding a group with superblock/GDT backups we will also
1465 * modify each of the reserved GDT dindirect blocks.
1466 */
2c869b26
JK
1467 credit = 3; /* sb, resize inode, resize inode dindirect */
1468 /* GDT blocks */
1469 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1470 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
9924a92a 1471 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
4bac1f8c
YY
1472 if (IS_ERR(handle)) {
1473 err = PTR_ERR(handle);
1474 goto exit;
1475 }
1476
5d601255 1477 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
4bac1f8c
YY
1478 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1479 if (err)
1480 goto exit_journal;
1481
1482 group = flex_gd->groups[0].group;
49598e04 1483 BUG_ON(group != sbi->s_groups_count);
4bac1f8c
YY
1484 err = ext4_add_new_descs(handle, sb, group,
1485 resize_inode, flex_gd->count);
1486 if (err)
1487 goto exit_journal;
1488
1489 err = ext4_setup_new_descs(handle, sb, flex_gd);
1490 if (err)
1491 goto exit_journal;
1492
1493 ext4_update_super(sb, flex_gd);
1494
1495 err = ext4_handle_dirty_super(handle, sb);
1496
1497exit_journal:
1498 err2 = ext4_journal_stop(handle);
1499 if (!err)
1500 err = err2;
1501
1502 if (!err) {
2ebd1704
YY
1503 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1504 int gdb_num_end = ((group + flex_gd->count - 1) /
1505 EXT4_DESC_PER_BLOCK(sb));
e2b911c5 1506 int meta_bg = ext4_has_feature_meta_bg(sb);
0acdb887 1507 sector_t old_gdb = 0;
2ebd1704 1508
4bac1f8c 1509 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
01f795f9 1510 sizeof(struct ext4_super_block), 0);
2ebd1704 1511 for (; gdb_num <= gdb_num_end; gdb_num++) {
4bac1f8c 1512 struct buffer_head *gdb_bh;
2ebd1704 1513
4bac1f8c 1514 gdb_bh = sbi->s_group_desc[gdb_num];
0acdb887
TM
1515 if (old_gdb == gdb_bh->b_blocknr)
1516 continue;
4bac1f8c 1517 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
01f795f9 1518 gdb_bh->b_size, meta_bg);
0acdb887 1519 old_gdb = gdb_bh->b_blocknr;
4bac1f8c
YY
1520 }
1521 }
1522exit:
1523 return err;
1524}
1525
19c5246d
YY
1526static int ext4_setup_next_flex_gd(struct super_block *sb,
1527 struct ext4_new_flex_group_data *flex_gd,
1528 ext4_fsblk_t n_blocks_count,
1529 unsigned long flexbg_size)
1530{
d77147ff 1531 struct ext4_sb_info *sbi = EXT4_SB(sb);
1532 struct ext4_super_block *es = sbi->s_es;
19c5246d
YY
1533 struct ext4_new_group_data *group_data = flex_gd->groups;
1534 ext4_fsblk_t o_blocks_count;
1535 ext4_group_t n_group;
1536 ext4_group_t group;
1537 ext4_group_t last_group;
1538 ext4_grpblk_t last;
d77147ff 1539 ext4_grpblk_t clusters_per_group;
19c5246d
YY
1540 unsigned long i;
1541
d77147ff 1542 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
19c5246d
YY
1543
1544 o_blocks_count = ext4_blocks_count(es);
1545
1546 if (o_blocks_count == n_blocks_count)
1547 return 0;
1548
1549 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1550 BUG_ON(last);
1551 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1552
1553 last_group = group | (flexbg_size - 1);
1554 if (last_group > n_group)
1555 last_group = n_group;
1556
1557 flex_gd->count = last_group - group + 1;
1558
1559 for (i = 0; i < flex_gd->count; i++) {
1560 int overhead;
1561
1562 group_data[i].group = group + i;
d77147ff 1563 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
01f795f9 1564 overhead = ext4_group_overhead_blocks(sb, group + i);
d77147ff 1565 group_data[i].mdata_blocks = overhead;
1566 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
7f511862 1567 if (ext4_has_group_desc_csum(sb)) {
19c5246d
YY
1568 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1569 EXT4_BG_INODE_UNINIT;
7f511862
TT
1570 if (!test_opt(sb, INIT_INODE_TABLE))
1571 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1572 } else
19c5246d
YY
1573 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1574 }
1575
feb0ab32 1576 if (last_group == n_group && ext4_has_group_desc_csum(sb))
19c5246d
YY
1577 /* We need to initialize block bitmap of last group. */
1578 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1579
d77147ff 1580 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1581 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1582 group_data[i - 1].free_clusters_count -= clusters_per_group -
1583 last - 1;
19c5246d
YY
1584 }
1585
1586 return 1;
1587}
1588
ac27a0ec
DK
1589/* Add group descriptor data to an existing or new group descriptor block.
1590 * Ensure we handle all possible error conditions _before_ we start modifying
1591 * the filesystem, because we cannot abort the transaction and not have it
1592 * write the data to disk.
1593 *
1594 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1595 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1596 *
1597 * We only need to hold the superblock lock while we are actually adding
1598 * in the new group's counts to the superblock. Prior to that we have
1599 * not really "added" the group at all. We re-check that we are still
1600 * adding in the last group in case things have changed since verifying.
1601 */
617ba13b 1602int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
ac27a0ec 1603{
61f296cc 1604 struct ext4_new_flex_group_data flex_gd;
617ba13b
MC
1605 struct ext4_sb_info *sbi = EXT4_SB(sb);
1606 struct ext4_super_block *es = sbi->s_es;
1607 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
ac27a0ec 1608 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
ac27a0ec 1609 struct inode *inode = NULL;
03b40e34 1610 int gdb_off;
61f296cc
YY
1611 int err;
1612 __u16 bg_flags = 0;
ac27a0ec 1613
617ba13b 1614 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 1615
e2b911c5 1616 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
12062ddd 1617 ext4_warning(sb, "Can't resize non-sparse filesystem further");
ac27a0ec
DK
1618 return -EPERM;
1619 }
1620
bd81d8ee
LV
1621 if (ext4_blocks_count(es) + input->blocks_count <
1622 ext4_blocks_count(es)) {
12062ddd 1623 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1624 return -EINVAL;
1625 }
1626
617ba13b 1627 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
ac27a0ec 1628 le32_to_cpu(es->s_inodes_count)) {
12062ddd 1629 ext4_warning(sb, "inodes_count overflow");
ac27a0ec
DK
1630 return -EINVAL;
1631 }
1632
1633 if (reserved_gdb || gdb_off == 0) {
e647e291 1634 if (!ext4_has_feature_resize_inode(sb) ||
e2b911c5 1635 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
12062ddd 1636 ext4_warning(sb,
ac27a0ec
DK
1637 "No reserved GDT blocks, can't resize");
1638 return -EPERM;
1639 }
8a363970 1640 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1d1fe1ee 1641 if (IS_ERR(inode)) {
12062ddd 1642 ext4_warning(sb, "Error opening resize inode");
1d1fe1ee 1643 return PTR_ERR(inode);
ac27a0ec
DK
1644 }
1645 }
1646
920313a7 1647
61f296cc 1648 err = verify_group_input(sb, input);
08c3a813 1649 if (err)
61f296cc 1650 goto out;
ac27a0ec 1651
117fff10
TT
1652 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1653 if (err)
7f511862 1654 goto out;
117fff10 1655
28623c2f
TT
1656 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1657 if (err)
1658 goto out;
1659
61f296cc
YY
1660 flex_gd.count = 1;
1661 flex_gd.groups = input;
1662 flex_gd.bg_flags = &bg_flags;
1663 err = ext4_flex_group_add(sb, inode, &flex_gd);
1664out:
ac27a0ec
DK
1665 iput(inode);
1666 return err;
617ba13b 1667} /* ext4_group_add */
ac27a0ec 1668
18e31438
YY
1669/*
1670 * extend a group without checking assuming that checking has been done.
1671 */
1672static int ext4_group_extend_no_check(struct super_block *sb,
1673 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1674{
1675 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1676 handle_t *handle;
1677 int err = 0, err2;
1678
1679 /* We will update the superblock, one block bitmap, and
1680 * one group descriptor via ext4_group_add_blocks().
1681 */
9924a92a 1682 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
18e31438
YY
1683 if (IS_ERR(handle)) {
1684 err = PTR_ERR(handle);
1685 ext4_warning(sb, "error %d on journal start", err);
1686 return err;
1687 }
1688
5d601255 1689 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
18e31438
YY
1690 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1691 if (err) {
1692 ext4_warning(sb, "error %d on journal write access", err);
1693 goto errout;
1694 }
1695
1696 ext4_blocks_count_set(es, o_blocks_count + add);
636d7e2e 1697 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
18e31438
YY
1698 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1699 o_blocks_count + add);
1700 /* We add the blocks to the bitmap and set the group need init bit */
1701 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1702 if (err)
1703 goto errout;
1704 ext4_handle_dirty_super(handle, sb);
1705 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1706 o_blocks_count + add);
1707errout:
1708 err2 = ext4_journal_stop(handle);
1709 if (err2 && !err)
1710 err = err2;
1711
1712 if (!err) {
1713 if (test_opt(sb, DEBUG))
1714 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1715 "blocks\n", ext4_blocks_count(es));
6ca792ed 1716 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
01f795f9 1717 (char *)es, sizeof(struct ext4_super_block), 0);
18e31438
YY
1718 }
1719 return err;
1720}
1721
2b2d6d01
TT
1722/*
1723 * Extend the filesystem to the new number of blocks specified. This entry
ac27a0ec
DK
1724 * point is only used to extend the current filesystem to the end of the last
1725 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1726 * for emergencies (because it has no dependencies on reserved blocks).
1727 *
617ba13b 1728 * If we _really_ wanted, we could use default values to call ext4_group_add()
ac27a0ec
DK
1729 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1730 * GDT blocks are reserved to grow to the desired size.
1731 */
617ba13b
MC
1732int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1733 ext4_fsblk_t n_blocks_count)
ac27a0ec 1734{
617ba13b 1735 ext4_fsblk_t o_blocks_count;
617ba13b
MC
1736 ext4_grpblk_t last;
1737 ext4_grpblk_t add;
af5bc92d 1738 struct buffer_head *bh;
d89651c8 1739 int err;
5f21b0e6 1740 ext4_group_t group;
ac27a0ec 1741
bd81d8ee 1742 o_blocks_count = ext4_blocks_count(es);
ac27a0ec
DK
1743
1744 if (test_opt(sb, DEBUG))
92b97816
TT
1745 ext4_msg(sb, KERN_DEBUG,
1746 "extending last group from %llu to %llu blocks",
1747 o_blocks_count, n_blocks_count);
ac27a0ec
DK
1748
1749 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1750 return 0;
1751
1752 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
92b97816
TT
1753 ext4_msg(sb, KERN_ERR,
1754 "filesystem too large to resize to %llu blocks safely",
1755 n_blocks_count);
ac27a0ec 1756 if (sizeof(sector_t) < 8)
12062ddd 1757 ext4_warning(sb, "CONFIG_LBDAF not enabled");
ac27a0ec
DK
1758 return -EINVAL;
1759 }
1760
1761 if (n_blocks_count < o_blocks_count) {
12062ddd 1762 ext4_warning(sb, "can't shrink FS - resize aborted");
8f82f840 1763 return -EINVAL;
ac27a0ec
DK
1764 }
1765
1766 /* Handle the remaining blocks in the last group only. */
5f21b0e6 1767 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
ac27a0ec
DK
1768
1769 if (last == 0) {
12062ddd 1770 ext4_warning(sb, "need to use ext2online to resize further");
ac27a0ec
DK
1771 return -EPERM;
1772 }
1773
617ba13b 1774 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
ac27a0ec
DK
1775
1776 if (o_blocks_count + add < o_blocks_count) {
12062ddd 1777 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1778 return -EINVAL;
1779 }
1780
1781 if (o_blocks_count + add > n_blocks_count)
1782 add = n_blocks_count - o_blocks_count;
1783
1784 if (o_blocks_count + add < n_blocks_count)
12062ddd 1785 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
ac27a0ec
DK
1786 o_blocks_count + add, add);
1787
1788 /* See if the device is actually as big as what was requested */
2b2d6d01 1789 bh = sb_bread(sb, o_blocks_count + add - 1);
ac27a0ec 1790 if (!bh) {
12062ddd 1791 ext4_warning(sb, "can't read last block, resize aborted");
ac27a0ec
DK
1792 return -ENOSPC;
1793 }
1794 brelse(bh);
1795
d89651c8 1796 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
ac27a0ec 1797 return err;
617ba13b 1798} /* ext4_group_extend */
19c5246d 1799
1c6bd717
TT
1800
1801static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1802{
1803 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1804}
1805
1806/*
1807 * Release the resize inode and drop the resize_inode feature if there
1808 * are no more reserved gdt blocks, and then convert the file system
1809 * to enable meta_bg
1810 */
1811static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1812{
1813 handle_t *handle;
1814 struct ext4_sb_info *sbi = EXT4_SB(sb);
1815 struct ext4_super_block *es = sbi->s_es;
59e31c15 1816 struct ext4_inode_info *ei = EXT4_I(inode);
1c6bd717
TT
1817 ext4_fsblk_t nr;
1818 int i, ret, err = 0;
1819 int credits = 1;
1820
1821 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
59e31c15 1822 if (inode) {
1c6bd717
TT
1823 if (es->s_reserved_gdt_blocks) {
1824 ext4_error(sb, "Unexpected non-zero "
1825 "s_reserved_gdt_blocks");
1826 return -EPERM;
1827 }
1c6bd717
TT
1828
1829 /* Do a quick sanity check of the resize inode */
d77147ff 1830 if (inode->i_blocks != 1 << (inode->i_blkbits -
1831 (9 - sbi->s_cluster_bits)))
1c6bd717
TT
1832 goto invalid_resize_inode;
1833 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1834 if (i == EXT4_DIND_BLOCK) {
1835 if (ei->i_data[i])
1836 continue;
1837 else
1838 goto invalid_resize_inode;
1839 }
1840 if (ei->i_data[i])
1841 goto invalid_resize_inode;
1842 }
1843 credits += 3; /* block bitmap, bg descriptor, resize inode */
1844 }
1845
9924a92a 1846 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1c6bd717
TT
1847 if (IS_ERR(handle))
1848 return PTR_ERR(handle);
1849
5d601255 1850 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1c6bd717
TT
1851 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1852 if (err)
1853 goto errout;
1854
e2b911c5
DW
1855 ext4_clear_feature_resize_inode(sb);
1856 ext4_set_feature_meta_bg(sb);
1c6bd717
TT
1857 sbi->s_es->s_first_meta_bg =
1858 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1859
1860 err = ext4_handle_dirty_super(handle, sb);
1861 if (err) {
1862 ext4_std_error(sb, err);
1863 goto errout;
1864 }
1865
1866 if (inode) {
1867 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1868 ext4_free_blocks(handle, inode, NULL, nr, 1,
1869 EXT4_FREE_BLOCKS_METADATA |
1870 EXT4_FREE_BLOCKS_FORGET);
1871 ei->i_data[EXT4_DIND_BLOCK] = 0;
1872 inode->i_blocks = 0;
1873
1874 err = ext4_mark_inode_dirty(handle, inode);
1875 if (err)
1876 ext4_std_error(sb, err);
1877 }
1878
1879errout:
1880 ret = ext4_journal_stop(handle);
1881 if (!err)
1882 err = ret;
1883 return ret;
1884
1885invalid_resize_inode:
1886 ext4_error(sb, "corrupted/inconsistent resize inode");
1887 return -EINVAL;
1888}
1889
19c5246d
YY
1890/*
1891 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1892 *
1893 * @sb: super block of the fs to be resized
1894 * @n_blocks_count: the number of blocks resides in the resized fs
1895 */
1896int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1897{
1898 struct ext4_new_flex_group_data *flex_gd = NULL;
1899 struct ext4_sb_info *sbi = EXT4_SB(sb);
1900 struct ext4_super_block *es = sbi->s_es;
1901 struct buffer_head *bh;
01f795f9
YY
1902 struct inode *resize_inode = NULL;
1903 ext4_grpblk_t add, offset;
19c5246d
YY
1904 unsigned long n_desc_blocks;
1905 unsigned long o_desc_blocks;
01f795f9
YY
1906 ext4_group_t o_group;
1907 ext4_group_t n_group;
1908 ext4_fsblk_t o_blocks_count;
1c6bd717 1909 ext4_fsblk_t n_blocks_count_retry = 0;
4da4a56e 1910 unsigned long last_update_time = 0;
117fff10 1911 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
01f795f9 1912 int meta_bg;
19c5246d 1913
59e31c15
TT
1914 /* See if the device is actually as big as what was requested */
1915 bh = sb_bread(sb, n_blocks_count - 1);
1916 if (!bh) {
1917 ext4_warning(sb, "can't read last block, resize aborted");
1918 return -ENOSPC;
1919 }
1920 brelse(bh);
1921
1c6bd717 1922retry:
19c5246d
YY
1923 o_blocks_count = ext4_blocks_count(es);
1924
59e31c15
TT
1925 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1926 "to %llu blocks", o_blocks_count, n_blocks_count);
19c5246d
YY
1927
1928 if (n_blocks_count < o_blocks_count) {
1929 /* On-line shrinking not supported */
1930 ext4_warning(sb, "can't shrink FS - resize aborted");
1931 return -EINVAL;
1932 }
1933
1934 if (n_blocks_count == o_blocks_count)
1935 /* Nothing need to do */
1936 return 0;
1937
bd86298e 1938 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
4f2f76f7 1939 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
3f8a6411
TT
1940 ext4_warning(sb, "resize would cause inodes_count overflow");
1941 return -EINVAL;
1942 }
a0ade1de 1943 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
19c5246d 1944
1c6bd717
TT
1945 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1946 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
19c5246d 1947
e2b911c5 1948 meta_bg = ext4_has_feature_meta_bg(sb);
19c5246d 1949
e2b911c5 1950 if (ext4_has_feature_resize_inode(sb)) {
01f795f9
YY
1951 if (meta_bg) {
1952 ext4_error(sb, "resize_inode and meta_bg enabled "
1953 "simultaneously");
1954 return -EINVAL;
1955 }
1c6bd717
TT
1956 if (n_desc_blocks > o_desc_blocks +
1957 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1958 n_blocks_count_retry = n_blocks_count;
1959 n_desc_blocks = o_desc_blocks +
1960 le16_to_cpu(es->s_reserved_gdt_blocks);
1961 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
aec51758 1962 n_blocks_count = (ext4_fsblk_t)n_group *
f96c3ac8
JK
1963 EXT4_BLOCKS_PER_GROUP(sb) +
1964 le32_to_cpu(es->s_first_data_block);
1c6bd717 1965 n_group--; /* set to last group number */
01f795f9 1966 }
1c6bd717
TT
1967
1968 if (!resize_inode)
8a363970
TT
1969 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1970 EXT4_IGET_SPECIAL);
01f795f9
YY
1971 if (IS_ERR(resize_inode)) {
1972 ext4_warning(sb, "Error opening resize inode");
1973 return PTR_ERR(resize_inode);
1974 }
1c6bd717
TT
1975 }
1976
59e31c15 1977 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1c6bd717
TT
1978 err = ext4_convert_meta_bg(sb, resize_inode);
1979 if (err)
1980 goto out;
1981 if (resize_inode) {
1982 iput(resize_inode);
1983 resize_inode = NULL;
1984 }
1985 if (n_blocks_count_retry) {
1986 n_blocks_count = n_blocks_count_retry;
1987 n_blocks_count_retry = 0;
1988 goto retry;
1989 }
19c5246d
YY
1990 }
1991
f0a459de
TT
1992 /*
1993 * Make sure the last group has enough space so that it's
1994 * guaranteed to have enough space for all metadata blocks
1995 * that it might need to hold. (We might not need to store
1996 * the inode table blocks in the last block group, but there
1997 * will be cases where this might be needed.)
1998 */
1999 if ((ext4_group_first_block_no(sb, n_group) +
2000 ext4_group_overhead_blocks(sb, n_group) + 2 +
2001 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2002 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2003 n_group--;
2004 n_blocks_count_retry = 0;
2005 if (resize_inode) {
2006 iput(resize_inode);
2007 resize_inode = NULL;
2008 }
2009 goto retry;
2010 }
2011
a0ade1de
LC
2012 /* extend the last group */
2013 if (n_group == o_group)
2014 add = n_blocks_count - o_blocks_count;
2015 else
d77147ff 2016 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
a0ade1de 2017 if (add > 0) {
19c5246d
YY
2018 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2019 if (err)
2020 goto out;
2021 }
2022
d7574ad0 2023 if (ext4_blocks_count(es) == n_blocks_count)
19c5246d
YY
2024 goto out;
2025
117fff10
TT
2026 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2027 if (err)
f348e224 2028 goto out;
117fff10 2029
28623c2f
TT
2030 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2031 if (err)
2032 goto out;
2033
19c5246d
YY
2034 flex_gd = alloc_flex_gd(flexbg_size);
2035 if (flex_gd == NULL) {
2036 err = -ENOMEM;
2037 goto out;
2038 }
2039
2040 /* Add flex groups. Note that a regular group is a
2041 * flex group with 1 group.
2042 */
2043 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2044 flexbg_size)) {
4da4a56e
TT
2045 if (jiffies - last_update_time > HZ * 10) {
2046 if (last_update_time)
2047 ext4_msg(sb, KERN_INFO,
2048 "resized to %llu blocks",
2049 ext4_blocks_count(es));
2050 last_update_time = jiffies;
2051 }
03c1c290
YY
2052 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2053 break;
19c5246d
YY
2054 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2055 if (unlikely(err))
2056 break;
2057 }
2058
1c6bd717
TT
2059 if (!err && n_blocks_count_retry) {
2060 n_blocks_count = n_blocks_count_retry;
2061 n_blocks_count_retry = 0;
2062 free_flex_gd(flex_gd);
2063 flex_gd = NULL;
db6aee62
VA
2064 if (resize_inode) {
2065 iput(resize_inode);
2066 resize_inode = NULL;
2067 }
1c6bd717
TT
2068 goto retry;
2069 }
2070
19c5246d
YY
2071out:
2072 if (flex_gd)
2073 free_flex_gd(flex_gd);
01f795f9
YY
2074 if (resize_inode != NULL)
2075 iput(resize_inode);
59e31c15 2076 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
19c5246d
YY
2077 return err;
2078}