]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ext4/resize.c
UBUNTU: Ubuntu-4.15.0-96.97
[mirror_ubuntu-bionic-kernel.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{
ac495f0c 22 struct ext4_sb_info *sbi = EXT4_SB(sb);
8f82f840
YY
23 int ret = 0;
24
78a8da10 25 if (!ns_capable(sb->s_user_ns, CAP_SYS_RESOURCE))
8f82f840
YY
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 */
ac495f0c 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);
cf143064
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);
cf143064 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
210 flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
211 flexbg_size, GFP_NOFS);
212 if (flex_gd->groups == NULL)
213 goto out2;
214
215 flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
216 if (flex_gd->bg_flags == NULL)
217 goto out1;
218
219 return flex_gd;
220
221out1:
222 kfree(flex_gd->groups);
223out2:
224 kfree(flex_gd);
225out3:
226 return NULL;
227}
228
229static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
230{
231 kfree(flex_gd->bg_flags);
232 kfree(flex_gd->groups);
233 kfree(flex_gd);
234}
235
3fbea4b3
YY
236/*
237 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
238 * and inode tables for a flex group.
239 *
240 * This function is used by 64bit-resize. Note that this function allocates
241 * group tables from the 1st group of groups contained by @flexgd, which may
242 * be a partial of a flex group.
243 *
244 * @sb: super block of fs to which the groups belongs
03c1c290
YY
245 *
246 * Returns 0 on a successful allocation of the metadata blocks in the
247 * block group.
3fbea4b3 248 */
03c1c290 249static int ext4_alloc_group_tables(struct super_block *sb,
3fbea4b3
YY
250 struct ext4_new_flex_group_data *flex_gd,
251 int flexbg_size)
252{
253 struct ext4_new_group_data *group_data = flex_gd->groups;
3fbea4b3
YY
254 ext4_fsblk_t start_blk;
255 ext4_fsblk_t last_blk;
256 ext4_group_t src_group;
257 ext4_group_t bb_index = 0;
258 ext4_group_t ib_index = 0;
259 ext4_group_t it_index = 0;
260 ext4_group_t group;
261 ext4_group_t last_group;
262 unsigned overhead;
b93c9535 263 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
d77147ff 264 int i;
3fbea4b3
YY
265
266 BUG_ON(flex_gd->count == 0 || group_data == NULL);
267
268 src_group = group_data[0].group;
269 last_group = src_group + flex_gd->count - 1;
270
271 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
272 (last_group & ~(flexbg_size - 1))));
273next_group:
274 group = group_data[0].group;
03c1c290
YY
275 if (src_group >= group_data[0].group + flex_gd->count)
276 return -ENOSPC;
3fbea4b3
YY
277 start_blk = ext4_group_first_block_no(sb, src_group);
278 last_blk = start_blk + group_data[src_group - group].blocks_count;
279
01f795f9 280 overhead = ext4_group_overhead_blocks(sb, src_group);
3fbea4b3
YY
281
282 start_blk += overhead;
283
3fbea4b3
YY
284 /* We collect contiguous blocks as much as possible. */
285 src_group++;
01f795f9
YY
286 for (; src_group <= last_group; src_group++) {
287 overhead = ext4_group_overhead_blocks(sb, src_group);
b93c9535 288 if (overhead == 0)
3fbea4b3
YY
289 last_blk += group_data[src_group - group].blocks_count;
290 else
291 break;
01f795f9 292 }
3fbea4b3
YY
293
294 /* Allocate block bitmaps */
295 for (; bb_index < flex_gd->count; bb_index++) {
296 if (start_blk >= last_blk)
297 goto next_group;
298 group_data[bb_index].block_bitmap = start_blk++;
bd86298e 299 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 300 group -= group_data[0].group;
d77147ff 301 group_data[group].mdata_blocks++;
b93c9535 302 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
303 }
304
305 /* Allocate inode bitmaps */
306 for (; ib_index < flex_gd->count; ib_index++) {
307 if (start_blk >= last_blk)
308 goto next_group;
309 group_data[ib_index].inode_bitmap = start_blk++;
bd86298e 310 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 311 group -= group_data[0].group;
d77147ff 312 group_data[group].mdata_blocks++;
b93c9535 313 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
314 }
315
316 /* Allocate inode tables */
317 for (; it_index < flex_gd->count; it_index++) {
b93c9535
TT
318 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
319 ext4_fsblk_t next_group_start;
320
321 if (start_blk + itb > last_blk)
3fbea4b3
YY
322 goto next_group;
323 group_data[it_index].inode_table = start_blk;
b93c9535
TT
324 group = ext4_get_group_number(sb, start_blk);
325 next_group_start = ext4_group_first_block_no(sb, group + 1);
3fbea4b3 326 group -= group_data[0].group;
3fbea4b3 327
b93c9535
TT
328 if (start_blk + itb > next_group_start) {
329 flex_gd->bg_flags[group + 1] &= uninit_mask;
330 overhead = start_blk + itb - next_group_start;
d77147ff 331 group_data[group + 1].mdata_blocks += overhead;
b93c9535
TT
332 itb -= overhead;
333 }
334
d77147ff 335 group_data[group].mdata_blocks += itb;
b93c9535 336 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
337 start_blk += EXT4_SB(sb)->s_itb_per_group;
338 }
339
d77147ff 340 /* Update free clusters count to exclude metadata blocks */
341 for (i = 0; i < flex_gd->count; i++) {
342 group_data[i].free_clusters_count -=
343 EXT4_NUM_B2C(EXT4_SB(sb),
344 group_data[i].mdata_blocks);
345 }
346
3fbea4b3
YY
347 if (test_opt(sb, DEBUG)) {
348 int i;
349 group = group_data[0].group;
350
351 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
352 "%d groups, flexbg size is %d:\n", flex_gd->count,
353 flexbg_size);
354
355 for (i = 0; i < flex_gd->count; i++) {
d77147ff 356 ext4_debug(
357 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
3fbea4b3
YY
358 ext4_bg_has_super(sb, group + i) ? "normal" :
359 "no-super", group + i,
360 group_data[i].blocks_count,
d77147ff 361 group_data[i].free_clusters_count,
362 group_data[i].mdata_blocks);
3fbea4b3
YY
363 }
364 }
03c1c290 365 return 0;
3fbea4b3
YY
366}
367
ac27a0ec 368static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
617ba13b 369 ext4_fsblk_t blk)
ac27a0ec
DK
370{
371 struct buffer_head *bh;
372 int err;
373
374 bh = sb_getblk(sb, blk);
aebf0243 375 if (unlikely(!bh))
860d21e2 376 return ERR_PTR(-ENOMEM);
5d601255 377 BUFFER_TRACE(bh, "get_write_access");
617ba13b 378 if ((err = ext4_journal_get_write_access(handle, bh))) {
ac27a0ec
DK
379 brelse(bh);
380 bh = ERR_PTR(err);
381 } else {
ac27a0ec
DK
382 memset(bh->b_data, 0, sb->s_blocksize);
383 set_buffer_uptodate(bh);
ac27a0ec
DK
384 }
385
386 return bh;
387}
388
14904107
ES
389/*
390 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
391 * If that fails, restart the transaction & regain write access for the
392 * buffer head which is used for block_bitmap modifications.
393 */
6d40bc5a 394static int extend_or_restart_transaction(handle_t *handle, int thresh)
14904107
ES
395{
396 int err;
397
0390131b 398 if (ext4_handle_has_enough_credits(handle, thresh))
14904107
ES
399 return 0;
400
401 err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
402 if (err < 0)
403 return err;
404 if (err) {
6d40bc5a
YY
405 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
406 if (err)
14904107 407 return err;
2b2d6d01 408 }
14904107
ES
409
410 return 0;
411}
412
33afdcc5 413/*
d77147ff 414 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
33afdcc5
YY
415 *
416 * Helper function for ext4_setup_new_group_blocks() which set .
417 *
418 * @sb: super block
419 * @handle: journal handle
420 * @flex_gd: flex group data
421 */
422static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
423 struct ext4_new_flex_group_data *flex_gd,
d77147ff 424 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
33afdcc5 425{
d77147ff 426 struct ext4_sb_info *sbi = EXT4_SB(sb);
427 ext4_group_t count = last_cluster - first_cluster + 1;
33afdcc5
YY
428 ext4_group_t count2;
429
d77147ff 430 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
431 last_cluster);
432 for (count2 = count; count > 0;
433 count -= count2, first_cluster += count2) {
33afdcc5
YY
434 ext4_fsblk_t start;
435 struct buffer_head *bh;
436 ext4_group_t group;
437 int err;
438
d77147ff 439 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
440 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
33afdcc5
YY
441 group -= flex_gd->groups[0].group;
442
d77147ff 443 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
33afdcc5
YY
444 if (count2 > count)
445 count2 = count;
446
447 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
448 BUG_ON(flex_gd->count > 1);
449 continue;
450 }
451
452 err = extend_or_restart_transaction(handle, 1);
453 if (err)
454 return err;
455
456 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
aebf0243 457 if (unlikely(!bh))
860d21e2 458 return -ENOMEM;
33afdcc5 459
5d601255 460 BUFFER_TRACE(bh, "get_write_access");
33afdcc5 461 err = ext4_journal_get_write_access(handle, bh);
35fede91
VA
462 if (err) {
463 brelse(bh);
464 return err;
465 }
d77147ff 466 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
467 first_cluster, first_cluster - start, count2);
468 ext4_set_bits(bh->b_data, first_cluster - start, count2);
33afdcc5
YY
469
470 err = ext4_handle_dirty_metadata(handle, NULL, bh);
35fede91 471 brelse(bh);
33afdcc5
YY
472 if (unlikely(err))
473 return err;
33afdcc5
YY
474 }
475
476 return 0;
477}
478
479/*
480 * Set up the block and inode bitmaps, and the inode table for the new groups.
481 * This doesn't need to be part of the main transaction, since we are only
482 * changing blocks outside the actual filesystem. We still do journaling to
483 * ensure the recovery is correct in case of a failure just after resize.
484 * If any part of this fails, we simply abort the resize.
485 *
486 * setup_new_flex_group_blocks handles a flex group as follow:
487 * 1. copy super block and GDT, and initialize group tables if necessary.
488 * In this step, we only set bits in blocks bitmaps for blocks taken by
489 * super block and GDT.
490 * 2. allocate group tables in block bitmaps, that is, set bits in block
491 * bitmap for blocks taken by group tables.
492 */
493static int setup_new_flex_group_blocks(struct super_block *sb,
494 struct ext4_new_flex_group_data *flex_gd)
495{
496 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
497 ext4_fsblk_t start;
498 ext4_fsblk_t block;
499 struct ext4_sb_info *sbi = EXT4_SB(sb);
500 struct ext4_super_block *es = sbi->s_es;
501 struct ext4_new_group_data *group_data = flex_gd->groups;
502 __u16 *bg_flags = flex_gd->bg_flags;
503 handle_t *handle;
504 ext4_group_t group, count;
505 struct buffer_head *bh = NULL;
506 int reserved_gdb, i, j, err = 0, err2;
01f795f9 507 int meta_bg;
33afdcc5
YY
508
509 BUG_ON(!flex_gd->count || !group_data ||
510 group_data[0].group != sbi->s_groups_count);
511
512 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
e2b911c5 513 meta_bg = ext4_has_feature_meta_bg(sb);
33afdcc5
YY
514
515 /* This transaction may be extended/restarted along the way */
9924a92a 516 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
33afdcc5
YY
517 if (IS_ERR(handle))
518 return PTR_ERR(handle);
519
520 group = group_data[0].group;
521 for (i = 0; i < flex_gd->count; i++, group++) {
522 unsigned long gdblocks;
01f795f9 523 ext4_grpblk_t overhead;
33afdcc5
YY
524
525 gdblocks = ext4_bg_num_gdb(sb, group);
526 start = ext4_group_first_block_no(sb, group);
527
01f795f9 528 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
6df935ad
YY
529 goto handle_itb;
530
01f795f9
YY
531 if (meta_bg == 1) {
532 ext4_group_t first_group;
533 first_group = ext4_meta_bg_first_group(sb, group);
534 if (first_group != group + 1 &&
535 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
536 goto handle_itb;
537 }
538
539 block = start + ext4_bg_has_super(sb, group);
33afdcc5 540 /* Copy all of the GDT blocks into the backup in this group */
01f795f9 541 for (j = 0; j < gdblocks; j++, block++) {
33afdcc5
YY
542 struct buffer_head *gdb;
543
544 ext4_debug("update backup group %#04llx\n", block);
545 err = extend_or_restart_transaction(handle, 1);
546 if (err)
547 goto out;
548
549 gdb = sb_getblk(sb, block);
aebf0243 550 if (unlikely(!gdb)) {
860d21e2 551 err = -ENOMEM;
33afdcc5
YY
552 goto out;
553 }
554
5d601255 555 BUFFER_TRACE(gdb, "get_write_access");
33afdcc5
YY
556 err = ext4_journal_get_write_access(handle, gdb);
557 if (err) {
558 brelse(gdb);
559 goto out;
560 }
561 memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
562 gdb->b_size);
563 set_buffer_uptodate(gdb);
564
565 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
566 if (unlikely(err)) {
567 brelse(gdb);
568 goto out;
569 }
570 brelse(gdb);
571 }
572
573 /* Zero out all of the reserved backup group descriptor
574 * table blocks
575 */
576 if (ext4_bg_has_super(sb, group)) {
577 err = sb_issue_zeroout(sb, gdblocks + start + 1,
578 reserved_gdb, GFP_NOFS);
579 if (err)
580 goto out;
581 }
582
6df935ad 583handle_itb:
33afdcc5
YY
584 /* Initialize group tables of the grop @group */
585 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
586 goto handle_bb;
587
588 /* Zero out all of the inode table blocks */
589 block = group_data[i].inode_table;
590 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
591 block, sbi->s_itb_per_group);
592 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
593 GFP_NOFS);
594 if (err)
595 goto out;
596
597handle_bb:
598 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
599 goto handle_ib;
600
601 /* Initialize block bitmap of the @group */
602 block = group_data[i].block_bitmap;
603 err = extend_or_restart_transaction(handle, 1);
604 if (err)
605 goto out;
606
607 bh = bclean(handle, sb, block);
608 if (IS_ERR(bh)) {
609 err = PTR_ERR(bh);
610 goto out;
611 }
01f795f9
YY
612 overhead = ext4_group_overhead_blocks(sb, group);
613 if (overhead != 0) {
33afdcc5
YY
614 ext4_debug("mark backup superblock %#04llx (+0)\n",
615 start);
d77147ff 616 ext4_set_bits(bh->b_data, 0,
617 EXT4_NUM_B2C(sbi, overhead));
33afdcc5 618 }
d77147ff 619 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
33afdcc5
YY
620 sb->s_blocksize * 8, bh->b_data);
621 err = ext4_handle_dirty_metadata(handle, NULL, bh);
ea2ffe75 622 brelse(bh);
33afdcc5
YY
623 if (err)
624 goto out;
33afdcc5
YY
625
626handle_ib:
627 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
628 continue;
629
630 /* Initialize inode bitmap of the @group */
631 block = group_data[i].inode_bitmap;
632 err = extend_or_restart_transaction(handle, 1);
633 if (err)
634 goto out;
635 /* Mark unused entries in inode bitmap used */
636 bh = bclean(handle, sb, block);
637 if (IS_ERR(bh)) {
638 err = PTR_ERR(bh);
639 goto out;
640 }
641
642 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
643 sb->s_blocksize * 8, bh->b_data);
644 err = ext4_handle_dirty_metadata(handle, NULL, bh);
ea2ffe75 645 brelse(bh);
33afdcc5
YY
646 if (err)
647 goto out;
33afdcc5 648 }
33afdcc5
YY
649
650 /* Mark group tables in block bitmap */
651 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
652 count = group_table_count[j];
653 start = (&group_data[0].block_bitmap)[j];
654 block = start;
655 for (i = 1; i < flex_gd->count; i++) {
656 block += group_table_count[j];
657 if (block == (&group_data[i].block_bitmap)[j]) {
658 count += group_table_count[j];
659 continue;
660 }
661 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 662 flex_gd,
663 EXT4_B2C(sbi, start),
664 EXT4_B2C(sbi,
665 start + count
666 - 1));
33afdcc5
YY
667 if (err)
668 goto out;
669 count = group_table_count[j];
b93c9535 670 start = (&group_data[i].block_bitmap)[j];
33afdcc5
YY
671 block = start;
672 }
673
674 if (count) {
675 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 676 flex_gd,
677 EXT4_B2C(sbi, start),
678 EXT4_B2C(sbi,
679 start + count
680 - 1));
33afdcc5
YY
681 if (err)
682 goto out;
683 }
684 }
685
686out:
33afdcc5
YY
687 err2 = ext4_journal_stop(handle);
688 if (err2 && !err)
689 err = err2;
690
691 return err;
692}
693
ac27a0ec
DK
694/*
695 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
617ba13b 696 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
ac27a0ec
DK
697 * calling this for the first time. In a sparse filesystem it will be the
698 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
699 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
700 */
617ba13b 701static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
ac27a0ec
DK
702 unsigned *five, unsigned *seven)
703{
704 unsigned *min = three;
705 int mult = 3;
706 unsigned ret;
707
e2b911c5 708 if (!ext4_has_feature_sparse_super(sb)) {
ac27a0ec
DK
709 ret = *min;
710 *min += 1;
711 return ret;
712 }
713
714 if (*five < *min) {
715 min = five;
716 mult = 5;
717 }
718 if (*seven < *min) {
719 min = seven;
720 mult = 7;
721 }
722
723 ret = *min;
724 *min *= mult;
725
726 return ret;
727}
728
729/*
730 * Check that all of the backup GDT blocks are held in the primary GDT block.
731 * It is assumed that they are stored in group order. Returns the number of
732 * groups in current filesystem that have BACKUPS, or -ve error code.
733 */
734static int verify_reserved_gdb(struct super_block *sb,
c72df9f9 735 ext4_group_t end,
ac27a0ec
DK
736 struct buffer_head *primary)
737{
617ba13b 738 const ext4_fsblk_t blk = primary->b_blocknr;
ac27a0ec
DK
739 unsigned three = 1;
740 unsigned five = 5;
741 unsigned seven = 7;
742 unsigned grp;
743 __le32 *p = (__le32 *)primary->b_data;
744 int gdbackups = 0;
745
617ba13b 746 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
bd81d8ee
LV
747 if (le32_to_cpu(*p++) !=
748 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
12062ddd 749 ext4_warning(sb, "reserved GDT %llu"
2ae02107 750 " missing grp %d (%llu)",
ac27a0ec 751 blk, grp,
bd81d8ee
LV
752 grp *
753 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
754 blk);
ac27a0ec
DK
755 return -EINVAL;
756 }
617ba13b 757 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
ac27a0ec
DK
758 return -EFBIG;
759 }
760
761 return gdbackups;
762}
763
764/*
765 * Called when we need to bring a reserved group descriptor table block into
766 * use from the resize inode. The primary copy of the new GDT block currently
767 * is an indirect block (under the double indirect block in the resize inode).
768 * The new backup GDT blocks will be stored as leaf blocks in this indirect
769 * block, in group order. Even though we know all the block numbers we need,
770 * we check to ensure that the resize inode has actually reserved these blocks.
771 *
772 * Don't need to update the block bitmaps because the blocks are still in use.
773 *
774 * We get all of the error cases out of the way, so that we are sure to not
775 * fail once we start modifying the data on disk, because JBD has no rollback.
776 */
777static int add_new_gdb(handle_t *handle, struct inode *inode,
2f919710 778 ext4_group_t group)
ac27a0ec
DK
779{
780 struct super_block *sb = inode->i_sb;
617ba13b 781 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
2f919710 782 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
617ba13b 783 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
cf143064
TT
784 struct buffer_head **o_group_desc, **n_group_desc = NULL;
785 struct buffer_head *dind = NULL;
786 struct buffer_head *gdb_bh = NULL;
ac27a0ec 787 int gdbackups;
cf143064 788 struct ext4_iloc iloc = { .bh = NULL };
ac27a0ec
DK
789 __le32 *data;
790 int err;
791
792 if (test_opt(sb, DEBUG))
793 printk(KERN_DEBUG
617ba13b 794 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
ac27a0ec
DK
795 gdb_num);
796
cf143064
TT
797 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
798 if (IS_ERR(gdb_bh))
799 return PTR_ERR(gdb_bh);
ac27a0ec 800
c72df9f9 801 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
2f919710 802 if (gdbackups < 0) {
ac27a0ec 803 err = gdbackups;
cf143064 804 goto errout;
ac27a0ec
DK
805 }
806
617ba13b 807 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
cf143064
TT
808 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
809 if (IS_ERR(dind)) {
810 err = PTR_ERR(dind);
811 dind = NULL;
812 goto errout;
ac27a0ec
DK
813 }
814
815 data = (__le32 *)dind->b_data;
617ba13b 816 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
12062ddd 817 ext4_warning(sb, "new group %u GDT block %llu not reserved",
2f919710 818 group, gdblock);
ac27a0ec 819 err = -EINVAL;
cf143064 820 goto errout;
ac27a0ec
DK
821 }
822
5d601255 823 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
b4097142
TT
824 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
825 if (unlikely(err))
cf143064 826 goto errout;
ac27a0ec 827
5d601255 828 BUFFER_TRACE(gdb_bh, "get_write_access");
2f919710 829 err = ext4_journal_get_write_access(handle, gdb_bh);
b4097142 830 if (unlikely(err))
cf143064 831 goto errout;
ac27a0ec 832
5d601255 833 BUFFER_TRACE(dind, "get_write_access");
b4097142
TT
834 err = ext4_journal_get_write_access(handle, dind);
835 if (unlikely(err))
836 ext4_std_error(sb, err);
ac27a0ec 837
617ba13b 838 /* ext4_reserve_inode_write() gets a reference on the iloc */
b4097142
TT
839 err = ext4_reserve_inode_write(handle, inode, &iloc);
840 if (unlikely(err))
cf143064 841 goto errout;
ac27a0ec 842
f18a5f21
TT
843 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
844 sizeof(struct buffer_head *),
845 GFP_NOFS);
ac27a0ec
DK
846 if (!n_group_desc) {
847 err = -ENOMEM;
f18a5f21
TT
848 ext4_warning(sb, "not enough memory for %lu groups",
849 gdb_num + 1);
cf143064 850 goto errout;
ac27a0ec
DK
851 }
852
853 /*
854 * Finally, we have all of the possible failures behind us...
855 *
856 * Remove new GDT block from inode double-indirect block and clear out
857 * the new GDT block for use (which also "frees" the backup GDT blocks
858 * from the reserved inode). We don't need to change the bitmaps for
859 * these blocks, because they are marked as in-use from being in the
860 * reserved inode, and will become GDT blocks (primary and backup).
861 */
617ba13b 862 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
b4097142
TT
863 err = ext4_handle_dirty_metadata(handle, NULL, dind);
864 if (unlikely(err)) {
865 ext4_std_error(sb, err);
cf143064 866 goto errout;
b4097142 867 }
d77147ff 868 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
869 (9 - EXT4_SB(sb)->s_cluster_bits);
617ba13b 870 ext4_mark_iloc_dirty(handle, inode, &iloc);
2f919710
YY
871 memset(gdb_bh->b_data, 0, sb->s_blocksize);
872 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
b4097142
TT
873 if (unlikely(err)) {
874 ext4_std_error(sb, err);
9abe1e3e 875 iloc.bh = NULL;
cf143064 876 goto errout;
b4097142
TT
877 }
878 brelse(dind);
ac27a0ec 879
617ba13b 880 o_group_desc = EXT4_SB(sb)->s_group_desc;
ac27a0ec 881 memcpy(n_group_desc, o_group_desc,
617ba13b 882 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
2f919710 883 n_group_desc[gdb_num] = gdb_bh;
617ba13b
MC
884 EXT4_SB(sb)->s_group_desc = n_group_desc;
885 EXT4_SB(sb)->s_gdb_count++;
b93b41d4 886 kvfree(o_group_desc);
ac27a0ec 887
e8546d06 888 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
b50924c2 889 err = ext4_handle_dirty_super(handle, sb);
b4097142
TT
890 if (err)
891 ext4_std_error(sb, err);
b4097142 892 return err;
cf143064 893errout:
b93b41d4 894 kvfree(n_group_desc);
ac27a0ec 895 brelse(iloc.bh);
ac27a0ec 896 brelse(dind);
2f919710 897 brelse(gdb_bh);
ac27a0ec 898
617ba13b 899 ext4_debug("leaving with error %d\n", err);
ac27a0ec
DK
900 return err;
901}
902
01f795f9
YY
903/*
904 * add_new_gdb_meta_bg is the sister of add_new_gdb.
905 */
906static int add_new_gdb_meta_bg(struct super_block *sb,
907 handle_t *handle, ext4_group_t group) {
908 ext4_fsblk_t gdblock;
909 struct buffer_head *gdb_bh;
910 struct buffer_head **o_group_desc, **n_group_desc;
911 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
912 int err;
913
914 gdblock = ext4_meta_bg_first_block_no(sb, group) +
915 ext4_bg_has_super(sb, group);
cf143064
TT
916 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
917 if (IS_ERR(gdb_bh))
918 return PTR_ERR(gdb_bh);
01f795f9
YY
919 n_group_desc = ext4_kvmalloc((gdb_num + 1) *
920 sizeof(struct buffer_head *),
921 GFP_NOFS);
922 if (!n_group_desc) {
8e69ddcf 923 brelse(gdb_bh);
01f795f9
YY
924 err = -ENOMEM;
925 ext4_warning(sb, "not enough memory for %lu groups",
926 gdb_num + 1);
927 return err;
928 }
929
930 o_group_desc = EXT4_SB(sb)->s_group_desc;
931 memcpy(n_group_desc, o_group_desc,
932 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
933 n_group_desc[gdb_num] = gdb_bh;
ea8e7af1
LC
934
935 BUFFER_TRACE(gdb_bh, "get_write_access");
936 err = ext4_journal_get_write_access(handle, gdb_bh);
937 if (err) {
938 kvfree(n_group_desc);
939 brelse(gdb_bh);
940 return err;
941 }
942
01f795f9
YY
943 EXT4_SB(sb)->s_group_desc = n_group_desc;
944 EXT4_SB(sb)->s_gdb_count++;
b93b41d4 945 kvfree(o_group_desc);
01f795f9
YY
946 return err;
947}
948
ac27a0ec
DK
949/*
950 * Called when we are adding a new group which has a backup copy of each of
951 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
952 * We need to add these reserved backup GDT blocks to the resize inode, so
953 * that they are kept for future resizing and not allocated to files.
954 *
955 * Each reserved backup GDT block will go into a different indirect block.
956 * The indirect blocks are actually the primary reserved GDT blocks,
957 * so we know in advance what their block numbers are. We only get the
958 * double-indirect block to verify it is pointing to the primary reserved
959 * GDT blocks so we don't overwrite a data block by accident. The reserved
960 * backup GDT blocks are stored in their reserved primary GDT block.
961 */
962static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
668f4dc5 963 ext4_group_t group)
ac27a0ec
DK
964{
965 struct super_block *sb = inode->i_sb;
617ba13b 966 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
d77147ff 967 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
ac27a0ec
DK
968 struct buffer_head **primary;
969 struct buffer_head *dind;
617ba13b
MC
970 struct ext4_iloc iloc;
971 ext4_fsblk_t blk;
ac27a0ec
DK
972 __le32 *data, *end;
973 int gdbackups = 0;
974 int res, i;
975 int err;
976
216553c4 977 primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
ac27a0ec
DK
978 if (!primary)
979 return -ENOMEM;
980
617ba13b 981 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
cf143064
TT
982 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
983 if (IS_ERR(dind)) {
984 err = PTR_ERR(dind);
985 dind = NULL;
ac27a0ec
DK
986 goto exit_free;
987 }
988
617ba13b 989 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
94460093
JB
990 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
991 EXT4_ADDR_PER_BLOCK(sb));
617ba13b 992 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
ac27a0ec
DK
993
994 /* Get each reserved primary GDT block and verify it holds backups */
995 for (res = 0; res < reserved_gdb; res++, blk++) {
996 if (le32_to_cpu(*data) != blk) {
12062ddd 997 ext4_warning(sb, "reserved block %llu"
ac27a0ec
DK
998 " not at offset %ld",
999 blk,
1000 (long)(data - (__le32 *)dind->b_data));
1001 err = -EINVAL;
1002 goto exit_bh;
1003 }
cf143064
TT
1004 primary[res] = ext4_sb_bread(sb, blk, 0);
1005 if (IS_ERR(primary[res])) {
1006 err = PTR_ERR(primary[res]);
1007 primary[res] = NULL;
ac27a0ec
DK
1008 goto exit_bh;
1009 }
c72df9f9
YY
1010 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1011 if (gdbackups < 0) {
ac27a0ec
DK
1012 brelse(primary[res]);
1013 err = gdbackups;
1014 goto exit_bh;
1015 }
1016 if (++data >= end)
1017 data = (__le32 *)dind->b_data;
1018 }
1019
1020 for (i = 0; i < reserved_gdb; i++) {
5d601255 1021 BUFFER_TRACE(primary[i], "get_write_access");
37be2f59 1022 if ((err = ext4_journal_get_write_access(handle, primary[i])))
ac27a0ec 1023 goto exit_bh;
ac27a0ec
DK
1024 }
1025
617ba13b 1026 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
ac27a0ec
DK
1027 goto exit_bh;
1028
1029 /*
1030 * Finally we can add each of the reserved backup GDT blocks from
1031 * the new group to its reserved primary GDT block.
1032 */
668f4dc5 1033 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1034 for (i = 0; i < reserved_gdb; i++) {
1035 int err2;
1036 data = (__le32 *)primary[i]->b_data;
1037 /* printk("reserving backup %lu[%u] = %lu\n",
1038 primary[i]->b_blocknr, gdbackups,
1039 blk + primary[i]->b_blocknr); */
1040 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
0390131b 1041 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
ac27a0ec
DK
1042 if (!err)
1043 err = err2;
1044 }
d77147ff 1045
1046 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
617ba13b 1047 ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec
DK
1048
1049exit_bh:
1050 while (--res >= 0)
1051 brelse(primary[res]);
1052 brelse(dind);
1053
1054exit_free:
1055 kfree(primary);
1056
1057 return err;
1058}
1059
1060/*
617ba13b 1061 * Update the backup copies of the ext4 metadata. These don't need to be part
ac27a0ec
DK
1062 * of the main resize transaction, because e2fsck will re-write them if there
1063 * is a problem (basically only OOM will cause a problem). However, we
1064 * _should_ update the backups if possible, in case the primary gets trashed
1065 * for some reason and we need to run e2fsck from a backup superblock. The
1066 * important part is that the new block and inode counts are in the backup
1067 * superblocks, and the location of the new group metadata in the GDT backups.
1068 *
32ed5058
TT
1069 * We do not need take the s_resize_lock for this, because these
1070 * blocks are not otherwise touched by the filesystem code when it is
1071 * mounted. We don't need to worry about last changing from
1072 * sbi->s_groups_count, because the worst that can happen is that we
1073 * do not copy the full number of backups at this time. The resize
1074 * which changed s_groups_count will backup again.
ac27a0ec 1075 */
904dad47 1076static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
01f795f9 1077 int size, int meta_bg)
ac27a0ec 1078{
617ba13b 1079 struct ext4_sb_info *sbi = EXT4_SB(sb);
01f795f9 1080 ext4_group_t last;
617ba13b 1081 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1082 unsigned three = 1;
1083 unsigned five = 5;
1084 unsigned seven = 7;
01f795f9 1085 ext4_group_t group = 0;
ac27a0ec
DK
1086 int rest = sb->s_blocksize - size;
1087 handle_t *handle;
1088 int err = 0, err2;
1089
9924a92a 1090 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
ac27a0ec
DK
1091 if (IS_ERR(handle)) {
1092 group = 1;
1093 err = PTR_ERR(handle);
1094 goto exit_err;
1095 }
1096
01f795f9
YY
1097 if (meta_bg == 0) {
1098 group = ext4_list_backups(sb, &three, &five, &seven);
1099 last = sbi->s_groups_count;
1100 } else {
904dad47 1101 group = ext4_get_group_number(sb, blk_off) + 1;
01f795f9
YY
1102 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1103 }
1104
1105 while (group < sbi->s_groups_count) {
ac27a0ec 1106 struct buffer_head *bh;
01f795f9 1107 ext4_fsblk_t backup_block;
ac27a0ec
DK
1108
1109 /* Out of journal space, and can't get more - abort - so sad */
0390131b
FM
1110 if (ext4_handle_valid(handle) &&
1111 handle->h_buffer_credits == 0 &&
617ba13b
MC
1112 ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1113 (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
ac27a0ec
DK
1114 break;
1115
01f795f9 1116 if (meta_bg == 0)
9378c676 1117 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
01f795f9
YY
1118 else
1119 backup_block = (ext4_group_first_block_no(sb, group) +
1120 ext4_bg_has_super(sb, group));
1121
1122 bh = sb_getblk(sb, backup_block);
aebf0243 1123 if (unlikely(!bh)) {
860d21e2 1124 err = -ENOMEM;
ac27a0ec
DK
1125 break;
1126 }
01f795f9
YY
1127 ext4_debug("update metadata backup %llu(+%llu)\n",
1128 backup_block, backup_block -
1129 ext4_group_first_block_no(sb, group));
5d601255 1130 BUFFER_TRACE(bh, "get_write_access");
9341c1b0
VA
1131 if ((err = ext4_journal_get_write_access(handle, bh))) {
1132 brelse(bh);
ac27a0ec 1133 break;
9341c1b0 1134 }
ac27a0ec
DK
1135 lock_buffer(bh);
1136 memcpy(bh->b_data, data, size);
1137 if (rest)
1138 memset(bh->b_data + size, 0, rest);
1139 set_buffer_uptodate(bh);
1140 unlock_buffer(bh);
b4097142
TT
1141 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1142 if (unlikely(err))
1143 ext4_std_error(sb, err);
ac27a0ec 1144 brelse(bh);
01f795f9
YY
1145
1146 if (meta_bg == 0)
1147 group = ext4_list_backups(sb, &three, &five, &seven);
1148 else if (group == last)
1149 break;
1150 else
1151 group = last;
ac27a0ec 1152 }
617ba13b 1153 if ((err2 = ext4_journal_stop(handle)) && !err)
ac27a0ec
DK
1154 err = err2;
1155
1156 /*
1157 * Ugh! Need to have e2fsck write the backup copies. It is too
1158 * late to revert the resize, we shouldn't fail just because of
1159 * the backup copies (they are only needed in case of corruption).
1160 *
1161 * However, if we got here we have a journal problem too, so we
1162 * can't really start a transaction to mark the superblock.
1163 * Chicken out and just set the flag on the hope it will be written
1164 * to disk, and if not - we will simply wait until next fsck.
1165 */
1166exit_err:
1167 if (err) {
12062ddd 1168 ext4_warning(sb, "can't update backup for group %u (err %d), "
ac27a0ec 1169 "forcing fsck on next reboot", group, err);
617ba13b
MC
1170 sbi->s_mount_state &= ~EXT4_VALID_FS;
1171 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
ac27a0ec
DK
1172 mark_buffer_dirty(sbi->s_sbh);
1173 }
1174}
1175
bb08c1e7
YY
1176/*
1177 * ext4_add_new_descs() adds @count group descriptor of groups
1178 * starting at @group
1179 *
1180 * @handle: journal handle
1181 * @sb: super block
1182 * @group: the group no. of the first group desc to be added
1183 * @resize_inode: the resize inode
1184 * @count: number of group descriptors to be added
1185 */
1186static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1187 ext4_group_t group, struct inode *resize_inode,
1188 ext4_group_t count)
1189{
1190 struct ext4_sb_info *sbi = EXT4_SB(sb);
1191 struct ext4_super_block *es = sbi->s_es;
1192 struct buffer_head *gdb_bh;
1193 int i, gdb_off, gdb_num, err = 0;
01f795f9 1194 int meta_bg;
bb08c1e7 1195
e2b911c5 1196 meta_bg = ext4_has_feature_meta_bg(sb);
bb08c1e7
YY
1197 for (i = 0; i < count; i++, group++) {
1198 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1199 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1200
1201 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1202 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1203
1204 /*
1205 * We will only either add reserved group blocks to a backup group
1206 * or remove reserved blocks for the first group in a new group block.
1207 * Doing both would be mean more complex code, and sane people don't
1208 * use non-sparse filesystems anymore. This is already checked above.
1209 */
1210 if (gdb_off) {
1211 gdb_bh = sbi->s_group_desc[gdb_num];
5d601255 1212 BUFFER_TRACE(gdb_bh, "get_write_access");
bb08c1e7
YY
1213 err = ext4_journal_get_write_access(handle, gdb_bh);
1214
1215 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1216 err = reserve_backup_gdb(handle, resize_inode, group);
01f795f9
YY
1217 } else if (meta_bg != 0) {
1218 err = add_new_gdb_meta_bg(sb, handle, group);
1219 } else {
bb08c1e7 1220 err = add_new_gdb(handle, resize_inode, group);
01f795f9 1221 }
bb08c1e7
YY
1222 if (err)
1223 break;
1224 }
1225 return err;
1226}
1227
41a246d1
DW
1228static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1229{
1230 struct buffer_head *bh = sb_getblk(sb, block);
aebf0243 1231 if (unlikely(!bh))
41a246d1 1232 return NULL;
7f1468d1
DM
1233 if (!bh_uptodate_or_lock(bh)) {
1234 if (bh_submit_read(bh) < 0) {
1235 brelse(bh);
1236 return NULL;
1237 }
41a246d1 1238 }
41a246d1
DW
1239
1240 return bh;
1241}
1242
1243static int ext4_set_bitmap_checksums(struct super_block *sb,
1244 ext4_group_t group,
1245 struct ext4_group_desc *gdp,
1246 struct ext4_new_group_data *group_data)
1247{
1248 struct buffer_head *bh;
1249
9aa5d32b 1250 if (!ext4_has_metadata_csum(sb))
41a246d1
DW
1251 return 0;
1252
1253 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1254 if (!bh)
1255 return -EIO;
1256 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1257 EXT4_INODES_PER_GROUP(sb) / 8);
1258 brelse(bh);
1259
fa77dcfa
DW
1260 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1261 if (!bh)
1262 return -EIO;
79f1ba49 1263 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
fa77dcfa
DW
1264 brelse(bh);
1265
41a246d1
DW
1266 return 0;
1267}
1268
083f5b24
YY
1269/*
1270 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1271 */
1272static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1273 struct ext4_new_flex_group_data *flex_gd)
1274{
1275 struct ext4_new_group_data *group_data = flex_gd->groups;
1276 struct ext4_group_desc *gdp;
1277 struct ext4_sb_info *sbi = EXT4_SB(sb);
1278 struct buffer_head *gdb_bh;
1279 ext4_group_t group;
1280 __u16 *bg_flags = flex_gd->bg_flags;
1281 int i, gdb_off, gdb_num, err = 0;
d77147ff 1282
083f5b24
YY
1283
1284 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1285 group = group_data->group;
1286
1287 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1288 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1289
1290 /*
1291 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1292 */
1293 gdb_bh = sbi->s_group_desc[gdb_num];
1294 /* Update group descriptor block for new group */
2716b802 1295 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
083f5b24
YY
1296 gdb_off * EXT4_DESC_SIZE(sb));
1297
1298 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1299 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1300 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
41a246d1
DW
1301 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1302 if (err) {
1303 ext4_std_error(sb, err);
1304 break;
1305 }
1306
083f5b24
YY
1307 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1308 ext4_free_group_clusters_set(sb, gdp,
d77147ff 1309 group_data->free_clusters_count);
083f5b24 1310 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
93f90526
TT
1311 if (ext4_has_group_desc_csum(sb))
1312 ext4_itable_unused_set(sb, gdp,
1313 EXT4_INODES_PER_GROUP(sb));
083f5b24 1314 gdp->bg_flags = cpu_to_le16(*bg_flags);
feb0ab32 1315 ext4_group_desc_csum_set(sb, group, gdp);
083f5b24
YY
1316
1317 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1318 if (unlikely(err)) {
1319 ext4_std_error(sb, err);
1320 break;
1321 }
1322
1323 /*
1324 * We can allocate memory for mb_alloc based on the new group
1325 * descriptor
1326 */
1327 err = ext4_mb_add_groupinfo(sb, group, gdp);
1328 if (err)
1329 break;
1330 }
1331 return err;
1332}
1333
2e10e2f2
YY
1334/*
1335 * ext4_update_super() updates the super block so that the newly added
1336 * groups can be seen by the filesystem.
1337 *
1338 * @sb: super block
1339 * @flex_gd: new added groups
1340 */
1341static void ext4_update_super(struct super_block *sb,
1342 struct ext4_new_flex_group_data *flex_gd)
1343{
1344 ext4_fsblk_t blocks_count = 0;
1345 ext4_fsblk_t free_blocks = 0;
1346 ext4_fsblk_t reserved_blocks = 0;
1347 struct ext4_new_group_data *group_data = flex_gd->groups;
1348 struct ext4_sb_info *sbi = EXT4_SB(sb);
1349 struct ext4_super_block *es = sbi->s_es;
8a991849 1350 int i;
2e10e2f2
YY
1351
1352 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1353 /*
1354 * Make the new blocks and inodes valid next. We do this before
1355 * increasing the group count so that once the group is enabled,
1356 * all of its blocks and inodes are already valid.
1357 *
1358 * We always allocate group-by-group, then block-by-block or
1359 * inode-by-inode within a group, so enabling these
1360 * blocks/inodes before the group is live won't actually let us
1361 * allocate the new space yet.
1362 */
1363 for (i = 0; i < flex_gd->count; i++) {
1364 blocks_count += group_data[i].blocks_count;
d77147ff 1365 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
2e10e2f2
YY
1366 }
1367
1368 reserved_blocks = ext4_r_blocks_count(es) * 100;
01f795f9 1369 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
2e10e2f2
YY
1370 reserved_blocks *= blocks_count;
1371 do_div(reserved_blocks, 100);
1372
1373 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
636d7e2e 1374 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
2e10e2f2
YY
1375 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1376 flex_gd->count);
636d7e2e
DW
1377 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1378 flex_gd->count);
2e10e2f2 1379
01f795f9 1380 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
2e10e2f2
YY
1381 /*
1382 * We need to protect s_groups_count against other CPUs seeing
1383 * inconsistent state in the superblock.
1384 *
1385 * The precise rules we use are:
1386 *
1387 * * Writers must perform a smp_wmb() after updating all
1388 * dependent data and before modifying the groups count
1389 *
1390 * * Readers must perform an smp_rmb() after reading the groups
1391 * count and before reading any dependent data.
1392 *
1393 * NB. These rules can be relaxed when checking the group count
1394 * while freeing data, as we can only allocate from a block
1395 * group after serialising against the group count, and we can
1396 * only then free after serialising in turn against that
1397 * allocation.
1398 */
1399 smp_wmb();
1400
1401 /* Update the global fs size fields */
1402 sbi->s_groups_count += flex_gd->count;
c5c72d81
TT
1403 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1404 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2e10e2f2
YY
1405
1406 /* Update the reserved block counts only once the new group is
1407 * active. */
1408 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1409 reserved_blocks);
1410
1411 /* Update the free space counts */
1412 percpu_counter_add(&sbi->s_freeclusters_counter,
810da240 1413 EXT4_NUM_B2C(sbi, free_blocks));
2e10e2f2
YY
1414 percpu_counter_add(&sbi->s_freeinodes_counter,
1415 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1416
01f795f9
YY
1417 ext4_debug("free blocks count %llu",
1418 percpu_counter_read(&sbi->s_freeclusters_counter));
e2b911c5 1419 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
2e10e2f2
YY
1420 ext4_group_t flex_group;
1421 flex_group = ext4_flex_group(sbi, group_data[0].group);
90ba983f
TT
1422 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1423 &sbi->s_flex_groups[flex_group].free_clusters);
2e10e2f2
YY
1424 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1425 &sbi->s_flex_groups[flex_group].free_inodes);
1426 }
1427
952fc18e
TT
1428 /*
1429 * Update the fs overhead information
1430 */
1431 ext4_calculate_overhead(sb);
1432
2e10e2f2
YY
1433 if (test_opt(sb, DEBUG))
1434 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1435 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1436 blocks_count, free_blocks, reserved_blocks);
1437}
1438
4bac1f8c
YY
1439/* Add a flex group to an fs. Ensure we handle all possible error conditions
1440 * _before_ we start modifying the filesystem, because we cannot abort the
1441 * transaction and not have it write the data to disk.
1442 */
1443static int ext4_flex_group_add(struct super_block *sb,
1444 struct inode *resize_inode,
1445 struct ext4_new_flex_group_data *flex_gd)
1446{
1447 struct ext4_sb_info *sbi = EXT4_SB(sb);
1448 struct ext4_super_block *es = sbi->s_es;
1449 ext4_fsblk_t o_blocks_count;
1450 ext4_grpblk_t last;
1451 ext4_group_t group;
1452 handle_t *handle;
1453 unsigned reserved_gdb;
1454 int err = 0, err2 = 0, credit;
1455
1456 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1457
1458 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1459 o_blocks_count = ext4_blocks_count(es);
1460 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1461 BUG_ON(last);
1462
1463 err = setup_new_flex_group_blocks(sb, flex_gd);
1464 if (err)
1465 goto exit;
1466 /*
1467 * We will always be modifying at least the superblock and GDT
2c869b26 1468 * blocks. If we are adding a group past the last current GDT block,
4bac1f8c
YY
1469 * we will also modify the inode and the dindirect block. If we
1470 * are adding a group with superblock/GDT backups we will also
1471 * modify each of the reserved GDT dindirect blocks.
1472 */
2c869b26
JK
1473 credit = 3; /* sb, resize inode, resize inode dindirect */
1474 /* GDT blocks */
1475 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1476 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
9924a92a 1477 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
4bac1f8c
YY
1478 if (IS_ERR(handle)) {
1479 err = PTR_ERR(handle);
1480 goto exit;
1481 }
1482
5d601255 1483 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
4bac1f8c
YY
1484 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1485 if (err)
1486 goto exit_journal;
1487
1488 group = flex_gd->groups[0].group;
1489 BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1490 err = ext4_add_new_descs(handle, sb, group,
1491 resize_inode, flex_gd->count);
1492 if (err)
1493 goto exit_journal;
1494
1495 err = ext4_setup_new_descs(handle, sb, flex_gd);
1496 if (err)
1497 goto exit_journal;
1498
1499 ext4_update_super(sb, flex_gd);
1500
1501 err = ext4_handle_dirty_super(handle, sb);
1502
1503exit_journal:
1504 err2 = ext4_journal_stop(handle);
1505 if (!err)
1506 err = err2;
1507
1508 if (!err) {
2ebd1704
YY
1509 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1510 int gdb_num_end = ((group + flex_gd->count - 1) /
1511 EXT4_DESC_PER_BLOCK(sb));
e2b911c5 1512 int meta_bg = ext4_has_feature_meta_bg(sb);
0acdb887 1513 sector_t old_gdb = 0;
2ebd1704 1514
4bac1f8c 1515 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
01f795f9 1516 sizeof(struct ext4_super_block), 0);
2ebd1704 1517 for (; gdb_num <= gdb_num_end; gdb_num++) {
4bac1f8c 1518 struct buffer_head *gdb_bh;
2ebd1704 1519
4bac1f8c 1520 gdb_bh = sbi->s_group_desc[gdb_num];
0acdb887
TM
1521 if (old_gdb == gdb_bh->b_blocknr)
1522 continue;
4bac1f8c 1523 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
01f795f9 1524 gdb_bh->b_size, meta_bg);
0acdb887 1525 old_gdb = gdb_bh->b_blocknr;
4bac1f8c
YY
1526 }
1527 }
1528exit:
1529 return err;
1530}
1531
19c5246d
YY
1532static int ext4_setup_next_flex_gd(struct super_block *sb,
1533 struct ext4_new_flex_group_data *flex_gd,
1534 ext4_fsblk_t n_blocks_count,
1535 unsigned long flexbg_size)
1536{
d77147ff 1537 struct ext4_sb_info *sbi = EXT4_SB(sb);
1538 struct ext4_super_block *es = sbi->s_es;
19c5246d
YY
1539 struct ext4_new_group_data *group_data = flex_gd->groups;
1540 ext4_fsblk_t o_blocks_count;
1541 ext4_group_t n_group;
1542 ext4_group_t group;
1543 ext4_group_t last_group;
1544 ext4_grpblk_t last;
d77147ff 1545 ext4_grpblk_t clusters_per_group;
19c5246d
YY
1546 unsigned long i;
1547
d77147ff 1548 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
19c5246d
YY
1549
1550 o_blocks_count = ext4_blocks_count(es);
1551
1552 if (o_blocks_count == n_blocks_count)
1553 return 0;
1554
1555 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1556 BUG_ON(last);
1557 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1558
1559 last_group = group | (flexbg_size - 1);
1560 if (last_group > n_group)
1561 last_group = n_group;
1562
1563 flex_gd->count = last_group - group + 1;
1564
1565 for (i = 0; i < flex_gd->count; i++) {
1566 int overhead;
1567
1568 group_data[i].group = group + i;
d77147ff 1569 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
01f795f9 1570 overhead = ext4_group_overhead_blocks(sb, group + i);
d77147ff 1571 group_data[i].mdata_blocks = overhead;
1572 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
7f511862 1573 if (ext4_has_group_desc_csum(sb)) {
19c5246d
YY
1574 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1575 EXT4_BG_INODE_UNINIT;
7f511862
TT
1576 if (!test_opt(sb, INIT_INODE_TABLE))
1577 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1578 } else
19c5246d
YY
1579 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1580 }
1581
feb0ab32 1582 if (last_group == n_group && ext4_has_group_desc_csum(sb))
19c5246d
YY
1583 /* We need to initialize block bitmap of last group. */
1584 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1585
d77147ff 1586 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1587 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1588 group_data[i - 1].free_clusters_count -= clusters_per_group -
1589 last - 1;
19c5246d
YY
1590 }
1591
1592 return 1;
1593}
1594
ac27a0ec
DK
1595/* Add group descriptor data to an existing or new group descriptor block.
1596 * Ensure we handle all possible error conditions _before_ we start modifying
1597 * the filesystem, because we cannot abort the transaction and not have it
1598 * write the data to disk.
1599 *
1600 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1601 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1602 *
1603 * We only need to hold the superblock lock while we are actually adding
1604 * in the new group's counts to the superblock. Prior to that we have
1605 * not really "added" the group at all. We re-check that we are still
1606 * adding in the last group in case things have changed since verifying.
1607 */
617ba13b 1608int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
ac27a0ec 1609{
61f296cc 1610 struct ext4_new_flex_group_data flex_gd;
617ba13b
MC
1611 struct ext4_sb_info *sbi = EXT4_SB(sb);
1612 struct ext4_super_block *es = sbi->s_es;
1613 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
ac27a0ec 1614 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
ac27a0ec 1615 struct inode *inode = NULL;
03b40e34 1616 int gdb_off;
61f296cc
YY
1617 int err;
1618 __u16 bg_flags = 0;
ac27a0ec 1619
617ba13b 1620 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 1621
e2b911c5 1622 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
12062ddd 1623 ext4_warning(sb, "Can't resize non-sparse filesystem further");
ac27a0ec
DK
1624 return -EPERM;
1625 }
1626
bd81d8ee
LV
1627 if (ext4_blocks_count(es) + input->blocks_count <
1628 ext4_blocks_count(es)) {
12062ddd 1629 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1630 return -EINVAL;
1631 }
1632
617ba13b 1633 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
ac27a0ec 1634 le32_to_cpu(es->s_inodes_count)) {
12062ddd 1635 ext4_warning(sb, "inodes_count overflow");
ac27a0ec
DK
1636 return -EINVAL;
1637 }
1638
1639 if (reserved_gdb || gdb_off == 0) {
9a060855 1640 if (!ext4_has_feature_resize_inode(sb) ||
e2b911c5 1641 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
12062ddd 1642 ext4_warning(sb,
ac27a0ec
DK
1643 "No reserved GDT blocks, can't resize");
1644 return -EPERM;
1645 }
23b9aab1 1646 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1d1fe1ee 1647 if (IS_ERR(inode)) {
12062ddd 1648 ext4_warning(sb, "Error opening resize inode");
1d1fe1ee 1649 return PTR_ERR(inode);
ac27a0ec
DK
1650 }
1651 }
1652
920313a7 1653
61f296cc 1654 err = verify_group_input(sb, input);
08c3a813 1655 if (err)
61f296cc 1656 goto out;
ac27a0ec 1657
117fff10
TT
1658 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1659 if (err)
7f511862 1660 goto out;
117fff10 1661
28623c2f
TT
1662 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1663 if (err)
1664 goto out;
1665
61f296cc
YY
1666 flex_gd.count = 1;
1667 flex_gd.groups = input;
1668 flex_gd.bg_flags = &bg_flags;
1669 err = ext4_flex_group_add(sb, inode, &flex_gd);
1670out:
ac27a0ec
DK
1671 iput(inode);
1672 return err;
617ba13b 1673} /* ext4_group_add */
ac27a0ec 1674
18e31438
YY
1675/*
1676 * extend a group without checking assuming that checking has been done.
1677 */
1678static int ext4_group_extend_no_check(struct super_block *sb,
1679 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1680{
1681 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1682 handle_t *handle;
1683 int err = 0, err2;
1684
1685 /* We will update the superblock, one block bitmap, and
1686 * one group descriptor via ext4_group_add_blocks().
1687 */
9924a92a 1688 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
18e31438
YY
1689 if (IS_ERR(handle)) {
1690 err = PTR_ERR(handle);
1691 ext4_warning(sb, "error %d on journal start", err);
1692 return err;
1693 }
1694
5d601255 1695 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
18e31438
YY
1696 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1697 if (err) {
1698 ext4_warning(sb, "error %d on journal write access", err);
1699 goto errout;
1700 }
1701
1702 ext4_blocks_count_set(es, o_blocks_count + add);
636d7e2e 1703 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
18e31438
YY
1704 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1705 o_blocks_count + add);
1706 /* We add the blocks to the bitmap and set the group need init bit */
1707 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1708 if (err)
1709 goto errout;
1710 ext4_handle_dirty_super(handle, sb);
1711 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1712 o_blocks_count + add);
1713errout:
1714 err2 = ext4_journal_stop(handle);
1715 if (err2 && !err)
1716 err = err2;
1717
1718 if (!err) {
1719 if (test_opt(sb, DEBUG))
1720 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1721 "blocks\n", ext4_blocks_count(es));
6ca792ed 1722 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
01f795f9 1723 (char *)es, sizeof(struct ext4_super_block), 0);
18e31438
YY
1724 }
1725 return err;
1726}
1727
2b2d6d01
TT
1728/*
1729 * Extend the filesystem to the new number of blocks specified. This entry
ac27a0ec
DK
1730 * point is only used to extend the current filesystem to the end of the last
1731 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1732 * for emergencies (because it has no dependencies on reserved blocks).
1733 *
617ba13b 1734 * If we _really_ wanted, we could use default values to call ext4_group_add()
ac27a0ec
DK
1735 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1736 * GDT blocks are reserved to grow to the desired size.
1737 */
617ba13b
MC
1738int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1739 ext4_fsblk_t n_blocks_count)
ac27a0ec 1740{
617ba13b 1741 ext4_fsblk_t o_blocks_count;
617ba13b
MC
1742 ext4_grpblk_t last;
1743 ext4_grpblk_t add;
af5bc92d 1744 struct buffer_head *bh;
d89651c8 1745 int err;
5f21b0e6 1746 ext4_group_t group;
ac27a0ec 1747
bd81d8ee 1748 o_blocks_count = ext4_blocks_count(es);
ac27a0ec
DK
1749
1750 if (test_opt(sb, DEBUG))
92b97816
TT
1751 ext4_msg(sb, KERN_DEBUG,
1752 "extending last group from %llu to %llu blocks",
1753 o_blocks_count, n_blocks_count);
ac27a0ec
DK
1754
1755 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1756 return 0;
1757
1758 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
92b97816
TT
1759 ext4_msg(sb, KERN_ERR,
1760 "filesystem too large to resize to %llu blocks safely",
1761 n_blocks_count);
ac27a0ec 1762 if (sizeof(sector_t) < 8)
12062ddd 1763 ext4_warning(sb, "CONFIG_LBDAF not enabled");
ac27a0ec
DK
1764 return -EINVAL;
1765 }
1766
1767 if (n_blocks_count < o_blocks_count) {
12062ddd 1768 ext4_warning(sb, "can't shrink FS - resize aborted");
8f82f840 1769 return -EINVAL;
ac27a0ec
DK
1770 }
1771
1772 /* Handle the remaining blocks in the last group only. */
5f21b0e6 1773 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
ac27a0ec
DK
1774
1775 if (last == 0) {
12062ddd 1776 ext4_warning(sb, "need to use ext2online to resize further");
ac27a0ec
DK
1777 return -EPERM;
1778 }
1779
617ba13b 1780 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
ac27a0ec
DK
1781
1782 if (o_blocks_count + add < o_blocks_count) {
12062ddd 1783 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1784 return -EINVAL;
1785 }
1786
1787 if (o_blocks_count + add > n_blocks_count)
1788 add = n_blocks_count - o_blocks_count;
1789
1790 if (o_blocks_count + add < n_blocks_count)
12062ddd 1791 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
ac27a0ec
DK
1792 o_blocks_count + add, add);
1793
1794 /* See if the device is actually as big as what was requested */
2b2d6d01 1795 bh = sb_bread(sb, o_blocks_count + add - 1);
ac27a0ec 1796 if (!bh) {
12062ddd 1797 ext4_warning(sb, "can't read last block, resize aborted");
ac27a0ec
DK
1798 return -ENOSPC;
1799 }
1800 brelse(bh);
1801
d89651c8 1802 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
ac27a0ec 1803 return err;
617ba13b 1804} /* ext4_group_extend */
19c5246d 1805
1c6bd717
TT
1806
1807static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1808{
1809 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1810}
1811
1812/*
1813 * Release the resize inode and drop the resize_inode feature if there
1814 * are no more reserved gdt blocks, and then convert the file system
1815 * to enable meta_bg
1816 */
1817static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1818{
1819 handle_t *handle;
1820 struct ext4_sb_info *sbi = EXT4_SB(sb);
1821 struct ext4_super_block *es = sbi->s_es;
59e31c15 1822 struct ext4_inode_info *ei = EXT4_I(inode);
1c6bd717
TT
1823 ext4_fsblk_t nr;
1824 int i, ret, err = 0;
1825 int credits = 1;
1826
1827 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
59e31c15 1828 if (inode) {
1c6bd717
TT
1829 if (es->s_reserved_gdt_blocks) {
1830 ext4_error(sb, "Unexpected non-zero "
1831 "s_reserved_gdt_blocks");
1832 return -EPERM;
1833 }
1c6bd717
TT
1834
1835 /* Do a quick sanity check of the resize inode */
d77147ff 1836 if (inode->i_blocks != 1 << (inode->i_blkbits -
1837 (9 - sbi->s_cluster_bits)))
1c6bd717
TT
1838 goto invalid_resize_inode;
1839 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1840 if (i == EXT4_DIND_BLOCK) {
1841 if (ei->i_data[i])
1842 continue;
1843 else
1844 goto invalid_resize_inode;
1845 }
1846 if (ei->i_data[i])
1847 goto invalid_resize_inode;
1848 }
1849 credits += 3; /* block bitmap, bg descriptor, resize inode */
1850 }
1851
9924a92a 1852 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1c6bd717
TT
1853 if (IS_ERR(handle))
1854 return PTR_ERR(handle);
1855
5d601255 1856 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1c6bd717
TT
1857 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1858 if (err)
1859 goto errout;
1860
e2b911c5
DW
1861 ext4_clear_feature_resize_inode(sb);
1862 ext4_set_feature_meta_bg(sb);
1c6bd717
TT
1863 sbi->s_es->s_first_meta_bg =
1864 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1865
1866 err = ext4_handle_dirty_super(handle, sb);
1867 if (err) {
1868 ext4_std_error(sb, err);
1869 goto errout;
1870 }
1871
1872 if (inode) {
1873 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1874 ext4_free_blocks(handle, inode, NULL, nr, 1,
1875 EXT4_FREE_BLOCKS_METADATA |
1876 EXT4_FREE_BLOCKS_FORGET);
1877 ei->i_data[EXT4_DIND_BLOCK] = 0;
1878 inode->i_blocks = 0;
1879
1880 err = ext4_mark_inode_dirty(handle, inode);
1881 if (err)
1882 ext4_std_error(sb, err);
1883 }
1884
1885errout:
1886 ret = ext4_journal_stop(handle);
1887 if (!err)
1888 err = ret;
1889 return ret;
1890
1891invalid_resize_inode:
1892 ext4_error(sb, "corrupted/inconsistent resize inode");
1893 return -EINVAL;
1894}
1895
19c5246d
YY
1896/*
1897 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1898 *
1899 * @sb: super block of the fs to be resized
1900 * @n_blocks_count: the number of blocks resides in the resized fs
1901 */
1902int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1903{
1904 struct ext4_new_flex_group_data *flex_gd = NULL;
1905 struct ext4_sb_info *sbi = EXT4_SB(sb);
1906 struct ext4_super_block *es = sbi->s_es;
1907 struct buffer_head *bh;
01f795f9
YY
1908 struct inode *resize_inode = NULL;
1909 ext4_grpblk_t add, offset;
19c5246d
YY
1910 unsigned long n_desc_blocks;
1911 unsigned long o_desc_blocks;
01f795f9
YY
1912 ext4_group_t o_group;
1913 ext4_group_t n_group;
1914 ext4_fsblk_t o_blocks_count;
1c6bd717 1915 ext4_fsblk_t n_blocks_count_retry = 0;
4da4a56e 1916 unsigned long last_update_time = 0;
117fff10 1917 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
01f795f9 1918 int meta_bg;
19c5246d 1919
59e31c15
TT
1920 /* See if the device is actually as big as what was requested */
1921 bh = sb_bread(sb, n_blocks_count - 1);
1922 if (!bh) {
1923 ext4_warning(sb, "can't read last block, resize aborted");
1924 return -ENOSPC;
1925 }
1926 brelse(bh);
1927
1c6bd717 1928retry:
19c5246d
YY
1929 o_blocks_count = ext4_blocks_count(es);
1930
59e31c15
TT
1931 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1932 "to %llu blocks", o_blocks_count, n_blocks_count);
19c5246d
YY
1933
1934 if (n_blocks_count < o_blocks_count) {
1935 /* On-line shrinking not supported */
1936 ext4_warning(sb, "can't shrink FS - resize aborted");
1937 return -EINVAL;
1938 }
1939
1940 if (n_blocks_count == o_blocks_count)
1941 /* Nothing need to do */
1942 return 0;
1943
bd86298e 1944 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
668dda3b 1945 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
3f8a6411
TT
1946 ext4_warning(sb, "resize would cause inodes_count overflow");
1947 return -EINVAL;
1948 }
a0ade1de 1949 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
19c5246d 1950
1c6bd717
TT
1951 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1952 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
19c5246d 1953
e2b911c5 1954 meta_bg = ext4_has_feature_meta_bg(sb);
19c5246d 1955
e2b911c5 1956 if (ext4_has_feature_resize_inode(sb)) {
01f795f9
YY
1957 if (meta_bg) {
1958 ext4_error(sb, "resize_inode and meta_bg enabled "
1959 "simultaneously");
1960 return -EINVAL;
1961 }
1c6bd717
TT
1962 if (n_desc_blocks > o_desc_blocks +
1963 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1964 n_blocks_count_retry = n_blocks_count;
1965 n_desc_blocks = o_desc_blocks +
1966 le16_to_cpu(es->s_reserved_gdt_blocks);
1967 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
aec51758 1968 n_blocks_count = (ext4_fsblk_t)n_group *
2d999d90
JK
1969 EXT4_BLOCKS_PER_GROUP(sb) +
1970 le32_to_cpu(es->s_first_data_block);
1c6bd717 1971 n_group--; /* set to last group number */
01f795f9 1972 }
1c6bd717
TT
1973
1974 if (!resize_inode)
23b9aab1
TT
1975 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1976 EXT4_IGET_SPECIAL);
01f795f9
YY
1977 if (IS_ERR(resize_inode)) {
1978 ext4_warning(sb, "Error opening resize inode");
1979 return PTR_ERR(resize_inode);
1980 }
1c6bd717
TT
1981 }
1982
59e31c15 1983 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1c6bd717
TT
1984 err = ext4_convert_meta_bg(sb, resize_inode);
1985 if (err)
1986 goto out;
1987 if (resize_inode) {
1988 iput(resize_inode);
1989 resize_inode = NULL;
1990 }
1991 if (n_blocks_count_retry) {
1992 n_blocks_count = n_blocks_count_retry;
1993 n_blocks_count_retry = 0;
1994 goto retry;
1995 }
19c5246d
YY
1996 }
1997
fb2d1e9c
TT
1998 /*
1999 * Make sure the last group has enough space so that it's
2000 * guaranteed to have enough space for all metadata blocks
2001 * that it might need to hold. (We might not need to store
2002 * the inode table blocks in the last block group, but there
2003 * will be cases where this might be needed.)
2004 */
2005 if ((ext4_group_first_block_no(sb, n_group) +
2006 ext4_group_overhead_blocks(sb, n_group) + 2 +
2007 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2008 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2009 n_group--;
2010 n_blocks_count_retry = 0;
2011 if (resize_inode) {
2012 iput(resize_inode);
2013 resize_inode = NULL;
2014 }
2015 goto retry;
2016 }
2017
a0ade1de
LC
2018 /* extend the last group */
2019 if (n_group == o_group)
2020 add = n_blocks_count - o_blocks_count;
2021 else
d77147ff 2022 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
a0ade1de 2023 if (add > 0) {
19c5246d
YY
2024 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2025 if (err)
2026 goto out;
2027 }
2028
d7574ad0 2029 if (ext4_blocks_count(es) == n_blocks_count)
19c5246d
YY
2030 goto out;
2031
117fff10
TT
2032 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2033 if (err)
11ab4805 2034 goto out;
117fff10 2035
28623c2f
TT
2036 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2037 if (err)
2038 goto out;
2039
19c5246d
YY
2040 flex_gd = alloc_flex_gd(flexbg_size);
2041 if (flex_gd == NULL) {
2042 err = -ENOMEM;
2043 goto out;
2044 }
2045
2046 /* Add flex groups. Note that a regular group is a
2047 * flex group with 1 group.
2048 */
2049 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2050 flexbg_size)) {
4da4a56e
TT
2051 if (jiffies - last_update_time > HZ * 10) {
2052 if (last_update_time)
2053 ext4_msg(sb, KERN_INFO,
2054 "resized to %llu blocks",
2055 ext4_blocks_count(es));
2056 last_update_time = jiffies;
2057 }
03c1c290
YY
2058 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2059 break;
19c5246d
YY
2060 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2061 if (unlikely(err))
2062 break;
2063 }
2064
1c6bd717
TT
2065 if (!err && n_blocks_count_retry) {
2066 n_blocks_count = n_blocks_count_retry;
2067 n_blocks_count_retry = 0;
2068 free_flex_gd(flex_gd);
2069 flex_gd = NULL;
56928b27
VA
2070 if (resize_inode) {
2071 iput(resize_inode);
2072 resize_inode = NULL;
2073 }
1c6bd717
TT
2074 goto retry;
2075 }
2076
19c5246d
YY
2077out:
2078 if (flex_gd)
2079 free_flex_gd(flex_gd);
01f795f9
YY
2080 if (resize_inode != NULL)
2081 iput(resize_inode);
9ed728f2
LC
2082 if (err)
2083 ext4_warning(sb, "error (%d) occurred during "
2084 "file system resize", err);
2085 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2086 ext4_blocks_count(es));
19c5246d
YY
2087 return err;
2088}