ext4: fix potential race between online resizing and write operations
[linux-block.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
1d0c3924
TT
20struct ext4_rcu_ptr {
21 struct rcu_head rcu;
22 void *ptr;
23};
24
25static void ext4_rcu_ptr_callback(struct rcu_head *head)
26{
27 struct ext4_rcu_ptr *ptr;
28
29 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
30 kvfree(ptr->ptr);
31 kfree(ptr);
32}
33
34void ext4_kvfree_array_rcu(void *to_free)
35{
36 struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
37
38 if (ptr) {
39 ptr->ptr = to_free;
40 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
41 return;
42 }
43 synchronize_rcu();
44 kvfree(to_free);
45}
46
8f82f840
YY
47int ext4_resize_begin(struct super_block *sb)
48{
5f8c1093 49 struct ext4_sb_info *sbi = EXT4_SB(sb);
8f82f840
YY
50 int ret = 0;
51
52 if (!capable(CAP_SYS_RESOURCE))
53 return -EPERM;
54
011fa994
TT
55 /*
56 * If we are not using the primary superblock/GDT copy don't resize,
57 * because the user tools have no way of handling this. Probably a
58 * bad time to do it anyways.
59 */
5f8c1093 60 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
011fa994
TT
61 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
62 ext4_warning(sb, "won't resize using backup superblock at %llu",
63 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
64 return -EPERM;
65 }
66
ce723c31
YY
67 /*
68 * We are not allowed to do online-resizing on a filesystem mounted
69 * with error, because it can destroy the filesystem easily.
70 */
71 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
72 ext4_warning(sb, "There are errors in the filesystem, "
8d2ae1cb 73 "so online resizing is not allowed");
ce723c31
YY
74 return -EPERM;
75 }
76
9549a168
TT
77 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
78 &EXT4_SB(sb)->s_ext4_flags))
8f82f840
YY
79 ret = -EBUSY;
80
81 return ret;
82}
83
84void ext4_resize_end(struct super_block *sb)
85{
9549a168 86 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
4e857c58 87 smp_mb__after_atomic();
8f82f840
YY
88}
89
01f795f9
YY
90static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
91 ext4_group_t group) {
92 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
93 EXT4_DESC_PER_BLOCK_BITS(sb);
94}
95
96static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
97 ext4_group_t group) {
98 group = ext4_meta_bg_first_group(sb, group);
99 return ext4_group_first_block_no(sb, group);
100}
101
102static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
103 ext4_group_t group) {
104 ext4_grpblk_t overhead;
105 overhead = ext4_bg_num_gdb(sb, group);
106 if (ext4_bg_has_super(sb, group))
107 overhead += 1 +
108 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
109 return overhead;
110}
111
ac27a0ec
DK
112#define outside(b, first, last) ((b) < (first) || (b) >= (last))
113#define inside(b, first, last) ((b) >= (first) && (b) < (last))
114
115static int verify_group_input(struct super_block *sb,
617ba13b 116 struct ext4_new_group_data *input)
ac27a0ec 117{
617ba13b
MC
118 struct ext4_sb_info *sbi = EXT4_SB(sb);
119 struct ext4_super_block *es = sbi->s_es;
bd81d8ee 120 ext4_fsblk_t start = ext4_blocks_count(es);
617ba13b 121 ext4_fsblk_t end = start + input->blocks_count;
fd2d4291 122 ext4_group_t group = input->group;
617ba13b 123 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
b302ef2d
TT
124 unsigned overhead;
125 ext4_fsblk_t metaend;
ac27a0ec 126 struct buffer_head *bh = NULL;
3a5b2ecd 127 ext4_grpblk_t free_blocks_count, offset;
ac27a0ec
DK
128 int err = -EINVAL;
129
b302ef2d
TT
130 if (group != sbi->s_groups_count) {
131 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
132 input->group, sbi->s_groups_count);
133 return -EINVAL;
134 }
135
136 overhead = ext4_group_overhead_blocks(sb, group);
137 metaend = start + overhead;
d77147ff 138 input->free_clusters_count = free_blocks_count =
ac27a0ec
DK
139 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
140
141 if (test_opt(sb, DEBUG))
617ba13b 142 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
ac27a0ec 143 "(%d free, %u reserved)\n",
617ba13b 144 ext4_bg_has_super(sb, input->group) ? "normal" :
ac27a0ec
DK
145 "no-super", input->group, input->blocks_count,
146 free_blocks_count, input->reserved_blocks);
147
3a5b2ecd 148 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
b302ef2d 149 if (offset != 0)
12062ddd 150 ext4_warning(sb, "Last group not full");
ac27a0ec 151 else if (input->reserved_blocks > input->blocks_count / 5)
12062ddd 152 ext4_warning(sb, "Reserved blocks too high (%u)",
ac27a0ec
DK
153 input->reserved_blocks);
154 else if (free_blocks_count < 0)
12062ddd 155 ext4_warning(sb, "Bad blocks count %u",
ac27a0ec 156 input->blocks_count);
fb265c9c
TT
157 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
158 err = PTR_ERR(bh);
159 bh = NULL;
12062ddd 160 ext4_warning(sb, "Cannot read last block (%llu)",
ac27a0ec 161 end - 1);
fb265c9c 162 } else if (outside(input->block_bitmap, start, end))
12062ddd 163 ext4_warning(sb, "Block bitmap not in group (block %llu)",
1939e49a 164 (unsigned long long)input->block_bitmap);
ac27a0ec 165 else if (outside(input->inode_bitmap, start, end))
12062ddd 166 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
1939e49a 167 (unsigned long long)input->inode_bitmap);
ac27a0ec 168 else if (outside(input->inode_table, start, end) ||
2b2d6d01 169 outside(itend - 1, start, end))
12062ddd 170 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
1939e49a 171 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 172 else if (input->inode_bitmap == input->block_bitmap)
12062ddd 173 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
1939e49a 174 (unsigned long long)input->block_bitmap);
ac27a0ec 175 else if (inside(input->block_bitmap, input->inode_table, itend))
12062ddd
ES
176 ext4_warning(sb, "Block bitmap (%llu) in inode table "
177 "(%llu-%llu)",
1939e49a
RD
178 (unsigned long long)input->block_bitmap,
179 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 180 else if (inside(input->inode_bitmap, input->inode_table, itend))
12062ddd
ES
181 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
182 "(%llu-%llu)",
1939e49a
RD
183 (unsigned long long)input->inode_bitmap,
184 (unsigned long long)input->inode_table, itend - 1);
ac27a0ec 185 else if (inside(input->block_bitmap, start, metaend))
12062ddd 186 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
1939e49a
RD
187 (unsigned long long)input->block_bitmap,
188 start, metaend - 1);
ac27a0ec 189 else if (inside(input->inode_bitmap, start, metaend))
12062ddd 190 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
1939e49a
RD
191 (unsigned long long)input->inode_bitmap,
192 start, metaend - 1);
ac27a0ec 193 else if (inside(input->inode_table, start, metaend) ||
2b2d6d01 194 inside(itend - 1, start, metaend))
12062ddd
ES
195 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
196 "(%llu-%llu)",
1939e49a
RD
197 (unsigned long long)input->inode_table,
198 itend - 1, start, metaend - 1);
ac27a0ec
DK
199 else
200 err = 0;
201 brelse(bh);
202
203 return err;
204}
205
28c7bac0
YY
206/*
207 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
208 * group each time.
209 */
210struct ext4_new_flex_group_data {
211 struct ext4_new_group_data *groups; /* new_group_data for groups
212 in the flex group */
213 __u16 *bg_flags; /* block group flags of groups
214 in @groups */
215 ext4_group_t count; /* number of groups in @groups
216 */
217};
218
219/*
220 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
221 * @flexbg_size.
222 *
223 * Returns NULL on failure otherwise address of the allocated structure.
224 */
225static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
226{
227 struct ext4_new_flex_group_data *flex_gd;
228
229 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
230 if (flex_gd == NULL)
231 goto out3;
232
46901760 233 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
967ac8af 234 goto out2;
28c7bac0
YY
235 flex_gd->count = flexbg_size;
236
6da2ec56
KC
237 flex_gd->groups = kmalloc_array(flexbg_size,
238 sizeof(struct ext4_new_group_data),
239 GFP_NOFS);
28c7bac0
YY
240 if (flex_gd->groups == NULL)
241 goto out2;
242
6da2ec56
KC
243 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
244 GFP_NOFS);
28c7bac0
YY
245 if (flex_gd->bg_flags == NULL)
246 goto out1;
247
248 return flex_gd;
249
250out1:
251 kfree(flex_gd->groups);
252out2:
253 kfree(flex_gd);
254out3:
255 return NULL;
256}
257
258static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
259{
260 kfree(flex_gd->bg_flags);
261 kfree(flex_gd->groups);
262 kfree(flex_gd);
263}
264
3fbea4b3
YY
265/*
266 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
267 * and inode tables for a flex group.
268 *
269 * This function is used by 64bit-resize. Note that this function allocates
270 * group tables from the 1st group of groups contained by @flexgd, which may
271 * be a partial of a flex group.
272 *
273 * @sb: super block of fs to which the groups belongs
03c1c290
YY
274 *
275 * Returns 0 on a successful allocation of the metadata blocks in the
276 * block group.
3fbea4b3 277 */
03c1c290 278static int ext4_alloc_group_tables(struct super_block *sb,
3fbea4b3
YY
279 struct ext4_new_flex_group_data *flex_gd,
280 int flexbg_size)
281{
282 struct ext4_new_group_data *group_data = flex_gd->groups;
3fbea4b3
YY
283 ext4_fsblk_t start_blk;
284 ext4_fsblk_t last_blk;
285 ext4_group_t src_group;
286 ext4_group_t bb_index = 0;
287 ext4_group_t ib_index = 0;
288 ext4_group_t it_index = 0;
289 ext4_group_t group;
290 ext4_group_t last_group;
291 unsigned overhead;
b93c9535 292 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
d77147ff 293 int i;
3fbea4b3
YY
294
295 BUG_ON(flex_gd->count == 0 || group_data == NULL);
296
297 src_group = group_data[0].group;
298 last_group = src_group + flex_gd->count - 1;
299
300 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
301 (last_group & ~(flexbg_size - 1))));
302next_group:
303 group = group_data[0].group;
03c1c290
YY
304 if (src_group >= group_data[0].group + flex_gd->count)
305 return -ENOSPC;
3fbea4b3
YY
306 start_blk = ext4_group_first_block_no(sb, src_group);
307 last_blk = start_blk + group_data[src_group - group].blocks_count;
308
01f795f9 309 overhead = ext4_group_overhead_blocks(sb, src_group);
3fbea4b3
YY
310
311 start_blk += overhead;
312
3fbea4b3
YY
313 /* We collect contiguous blocks as much as possible. */
314 src_group++;
01f795f9
YY
315 for (; src_group <= last_group; src_group++) {
316 overhead = ext4_group_overhead_blocks(sb, src_group);
b93c9535 317 if (overhead == 0)
3fbea4b3
YY
318 last_blk += group_data[src_group - group].blocks_count;
319 else
320 break;
01f795f9 321 }
3fbea4b3
YY
322
323 /* Allocate block bitmaps */
324 for (; bb_index < flex_gd->count; bb_index++) {
325 if (start_blk >= last_blk)
326 goto next_group;
327 group_data[bb_index].block_bitmap = start_blk++;
bd86298e 328 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 329 group -= group_data[0].group;
d77147ff 330 group_data[group].mdata_blocks++;
b93c9535 331 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
332 }
333
334 /* Allocate inode bitmaps */
335 for (; ib_index < flex_gd->count; ib_index++) {
336 if (start_blk >= last_blk)
337 goto next_group;
338 group_data[ib_index].inode_bitmap = start_blk++;
bd86298e 339 group = ext4_get_group_number(sb, start_blk - 1);
3fbea4b3 340 group -= group_data[0].group;
d77147ff 341 group_data[group].mdata_blocks++;
b93c9535 342 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
343 }
344
345 /* Allocate inode tables */
346 for (; it_index < flex_gd->count; it_index++) {
b93c9535
TT
347 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
348 ext4_fsblk_t next_group_start;
349
350 if (start_blk + itb > last_blk)
3fbea4b3
YY
351 goto next_group;
352 group_data[it_index].inode_table = start_blk;
b93c9535
TT
353 group = ext4_get_group_number(sb, start_blk);
354 next_group_start = ext4_group_first_block_no(sb, group + 1);
3fbea4b3 355 group -= group_data[0].group;
3fbea4b3 356
b93c9535
TT
357 if (start_blk + itb > next_group_start) {
358 flex_gd->bg_flags[group + 1] &= uninit_mask;
359 overhead = start_blk + itb - next_group_start;
d77147ff 360 group_data[group + 1].mdata_blocks += overhead;
b93c9535
TT
361 itb -= overhead;
362 }
363
d77147ff 364 group_data[group].mdata_blocks += itb;
b93c9535 365 flex_gd->bg_flags[group] &= uninit_mask;
3fbea4b3
YY
366 start_blk += EXT4_SB(sb)->s_itb_per_group;
367 }
368
d77147ff 369 /* Update free clusters count to exclude metadata blocks */
370 for (i = 0; i < flex_gd->count; i++) {
371 group_data[i].free_clusters_count -=
372 EXT4_NUM_B2C(EXT4_SB(sb),
373 group_data[i].mdata_blocks);
374 }
375
3fbea4b3
YY
376 if (test_opt(sb, DEBUG)) {
377 int i;
378 group = group_data[0].group;
379
380 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
381 "%d groups, flexbg size is %d:\n", flex_gd->count,
382 flexbg_size);
383
384 for (i = 0; i < flex_gd->count; i++) {
d77147ff 385 ext4_debug(
386 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
3fbea4b3
YY
387 ext4_bg_has_super(sb, group + i) ? "normal" :
388 "no-super", group + i,
389 group_data[i].blocks_count,
d77147ff 390 group_data[i].free_clusters_count,
391 group_data[i].mdata_blocks);
3fbea4b3
YY
392 }
393 }
03c1c290 394 return 0;
3fbea4b3
YY
395}
396
ac27a0ec 397static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
617ba13b 398 ext4_fsblk_t blk)
ac27a0ec
DK
399{
400 struct buffer_head *bh;
401 int err;
402
403 bh = sb_getblk(sb, blk);
aebf0243 404 if (unlikely(!bh))
860d21e2 405 return ERR_PTR(-ENOMEM);
5d601255 406 BUFFER_TRACE(bh, "get_write_access");
617ba13b 407 if ((err = ext4_journal_get_write_access(handle, bh))) {
ac27a0ec
DK
408 brelse(bh);
409 bh = ERR_PTR(err);
410 } else {
ac27a0ec
DK
411 memset(bh->b_data, 0, sb->s_blocksize);
412 set_buffer_uptodate(bh);
ac27a0ec
DK
413 }
414
415 return bh;
416}
417
83448bdf
JK
418static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
419{
420 return ext4_journal_ensure_credits_fn(handle, credits,
421 EXT4_MAX_TRANS_DATA, 0, 0);
422}
423
33afdcc5 424/*
d77147ff 425 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
33afdcc5
YY
426 *
427 * Helper function for ext4_setup_new_group_blocks() which set .
428 *
429 * @sb: super block
430 * @handle: journal handle
431 * @flex_gd: flex group data
432 */
433static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
434 struct ext4_new_flex_group_data *flex_gd,
d77147ff 435 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
33afdcc5 436{
d77147ff 437 struct ext4_sb_info *sbi = EXT4_SB(sb);
438 ext4_group_t count = last_cluster - first_cluster + 1;
33afdcc5
YY
439 ext4_group_t count2;
440
d77147ff 441 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
442 last_cluster);
443 for (count2 = count; count > 0;
444 count -= count2, first_cluster += count2) {
33afdcc5
YY
445 ext4_fsblk_t start;
446 struct buffer_head *bh;
447 ext4_group_t group;
448 int err;
449
d77147ff 450 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
451 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
33afdcc5
YY
452 group -= flex_gd->groups[0].group;
453
d77147ff 454 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
33afdcc5
YY
455 if (count2 > count)
456 count2 = count;
457
458 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
459 BUG_ON(flex_gd->count > 1);
460 continue;
461 }
462
83448bdf 463 err = ext4_resize_ensure_credits_batch(handle, 1);
a4130367 464 if (err < 0)
33afdcc5
YY
465 return err;
466
467 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
aebf0243 468 if (unlikely(!bh))
860d21e2 469 return -ENOMEM;
33afdcc5 470
5d601255 471 BUFFER_TRACE(bh, "get_write_access");
33afdcc5 472 err = ext4_journal_get_write_access(handle, bh);
cea57941
VA
473 if (err) {
474 brelse(bh);
33afdcc5 475 return err;
cea57941 476 }
d77147ff 477 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
478 first_cluster, first_cluster - start, count2);
479 ext4_set_bits(bh->b_data, first_cluster - start, count2);
33afdcc5
YY
480
481 err = ext4_handle_dirty_metadata(handle, NULL, bh);
cea57941 482 brelse(bh);
33afdcc5
YY
483 if (unlikely(err))
484 return err;
33afdcc5
YY
485 }
486
487 return 0;
488}
489
490/*
491 * Set up the block and inode bitmaps, and the inode table for the new groups.
492 * This doesn't need to be part of the main transaction, since we are only
493 * changing blocks outside the actual filesystem. We still do journaling to
494 * ensure the recovery is correct in case of a failure just after resize.
495 * If any part of this fails, we simply abort the resize.
496 *
497 * setup_new_flex_group_blocks handles a flex group as follow:
498 * 1. copy super block and GDT, and initialize group tables if necessary.
499 * In this step, we only set bits in blocks bitmaps for blocks taken by
500 * super block and GDT.
501 * 2. allocate group tables in block bitmaps, that is, set bits in block
502 * bitmap for blocks taken by group tables.
503 */
504static int setup_new_flex_group_blocks(struct super_block *sb,
505 struct ext4_new_flex_group_data *flex_gd)
506{
507 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
508 ext4_fsblk_t start;
509 ext4_fsblk_t block;
510 struct ext4_sb_info *sbi = EXT4_SB(sb);
511 struct ext4_super_block *es = sbi->s_es;
512 struct ext4_new_group_data *group_data = flex_gd->groups;
513 __u16 *bg_flags = flex_gd->bg_flags;
514 handle_t *handle;
515 ext4_group_t group, count;
516 struct buffer_head *bh = NULL;
517 int reserved_gdb, i, j, err = 0, err2;
01f795f9 518 int meta_bg;
33afdcc5
YY
519
520 BUG_ON(!flex_gd->count || !group_data ||
521 group_data[0].group != sbi->s_groups_count);
522
523 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
e2b911c5 524 meta_bg = ext4_has_feature_meta_bg(sb);
33afdcc5
YY
525
526 /* This transaction may be extended/restarted along the way */
9924a92a 527 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
33afdcc5
YY
528 if (IS_ERR(handle))
529 return PTR_ERR(handle);
530
531 group = group_data[0].group;
532 for (i = 0; i < flex_gd->count; i++, group++) {
533 unsigned long gdblocks;
01f795f9 534 ext4_grpblk_t overhead;
33afdcc5
YY
535
536 gdblocks = ext4_bg_num_gdb(sb, group);
537 start = ext4_group_first_block_no(sb, group);
538
01f795f9 539 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
6df935ad
YY
540 goto handle_itb;
541
01f795f9
YY
542 if (meta_bg == 1) {
543 ext4_group_t first_group;
544 first_group = ext4_meta_bg_first_group(sb, group);
545 if (first_group != group + 1 &&
546 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
547 goto handle_itb;
548 }
549
550 block = start + ext4_bg_has_super(sb, group);
33afdcc5 551 /* Copy all of the GDT blocks into the backup in this group */
01f795f9 552 for (j = 0; j < gdblocks; j++, block++) {
33afdcc5
YY
553 struct buffer_head *gdb;
554
555 ext4_debug("update backup group %#04llx\n", block);
83448bdf 556 err = ext4_resize_ensure_credits_batch(handle, 1);
a4130367 557 if (err < 0)
33afdcc5
YY
558 goto out;
559
560 gdb = sb_getblk(sb, block);
aebf0243 561 if (unlikely(!gdb)) {
860d21e2 562 err = -ENOMEM;
33afdcc5
YY
563 goto out;
564 }
565
5d601255 566 BUFFER_TRACE(gdb, "get_write_access");
33afdcc5
YY
567 err = ext4_journal_get_write_access(handle, gdb);
568 if (err) {
569 brelse(gdb);
570 goto out;
571 }
1d0c3924
TT
572 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
573 s_group_desc, j)->b_data, gdb->b_size);
33afdcc5
YY
574 set_buffer_uptodate(gdb);
575
576 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
577 if (unlikely(err)) {
578 brelse(gdb);
579 goto out;
580 }
581 brelse(gdb);
582 }
583
584 /* Zero out all of the reserved backup group descriptor
585 * table blocks
586 */
587 if (ext4_bg_has_super(sb, group)) {
588 err = sb_issue_zeroout(sb, gdblocks + start + 1,
589 reserved_gdb, GFP_NOFS);
590 if (err)
591 goto out;
592 }
593
6df935ad 594handle_itb:
33afdcc5
YY
595 /* Initialize group tables of the grop @group */
596 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
597 goto handle_bb;
598
599 /* Zero out all of the inode table blocks */
600 block = group_data[i].inode_table;
601 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
602 block, sbi->s_itb_per_group);
603 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
604 GFP_NOFS);
605 if (err)
606 goto out;
607
608handle_bb:
609 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
610 goto handle_ib;
611
612 /* Initialize block bitmap of the @group */
613 block = group_data[i].block_bitmap;
83448bdf 614 err = ext4_resize_ensure_credits_batch(handle, 1);
a4130367 615 if (err < 0)
33afdcc5
YY
616 goto out;
617
618 bh = bclean(handle, sb, block);
619 if (IS_ERR(bh)) {
620 err = PTR_ERR(bh);
621 goto out;
622 }
01f795f9
YY
623 overhead = ext4_group_overhead_blocks(sb, group);
624 if (overhead != 0) {
33afdcc5
YY
625 ext4_debug("mark backup superblock %#04llx (+0)\n",
626 start);
d77147ff 627 ext4_set_bits(bh->b_data, 0,
628 EXT4_NUM_B2C(sbi, overhead));
33afdcc5 629 }
d77147ff 630 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
33afdcc5
YY
631 sb->s_blocksize * 8, bh->b_data);
632 err = ext4_handle_dirty_metadata(handle, NULL, bh);
9e402893 633 brelse(bh);
33afdcc5
YY
634 if (err)
635 goto out;
33afdcc5
YY
636
637handle_ib:
638 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
639 continue;
640
641 /* Initialize inode bitmap of the @group */
642 block = group_data[i].inode_bitmap;
83448bdf 643 err = ext4_resize_ensure_credits_batch(handle, 1);
a4130367 644 if (err < 0)
33afdcc5
YY
645 goto out;
646 /* Mark unused entries in inode bitmap used */
647 bh = bclean(handle, sb, block);
648 if (IS_ERR(bh)) {
649 err = PTR_ERR(bh);
650 goto out;
651 }
652
653 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
654 sb->s_blocksize * 8, bh->b_data);
655 err = ext4_handle_dirty_metadata(handle, NULL, bh);
9e402893 656 brelse(bh);
33afdcc5
YY
657 if (err)
658 goto out;
33afdcc5 659 }
33afdcc5
YY
660
661 /* Mark group tables in block bitmap */
662 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
663 count = group_table_count[j];
664 start = (&group_data[0].block_bitmap)[j];
665 block = start;
666 for (i = 1; i < flex_gd->count; i++) {
667 block += group_table_count[j];
668 if (block == (&group_data[i].block_bitmap)[j]) {
669 count += group_table_count[j];
670 continue;
671 }
672 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 673 flex_gd,
674 EXT4_B2C(sbi, start),
675 EXT4_B2C(sbi,
676 start + count
677 - 1));
33afdcc5
YY
678 if (err)
679 goto out;
680 count = group_table_count[j];
b93c9535 681 start = (&group_data[i].block_bitmap)[j];
33afdcc5
YY
682 block = start;
683 }
684
685 if (count) {
686 err = set_flexbg_block_bitmap(sb, handle,
d77147ff 687 flex_gd,
688 EXT4_B2C(sbi, start),
689 EXT4_B2C(sbi,
690 start + count
691 - 1));
33afdcc5
YY
692 if (err)
693 goto out;
694 }
695 }
696
697out:
33afdcc5
YY
698 err2 = ext4_journal_stop(handle);
699 if (err2 && !err)
700 err = err2;
701
702 return err;
703}
704
ac27a0ec
DK
705/*
706 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
617ba13b 707 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
ac27a0ec
DK
708 * calling this for the first time. In a sparse filesystem it will be the
709 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
710 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
711 */
617ba13b 712static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
ac27a0ec
DK
713 unsigned *five, unsigned *seven)
714{
715 unsigned *min = three;
716 int mult = 3;
717 unsigned ret;
718
e2b911c5 719 if (!ext4_has_feature_sparse_super(sb)) {
ac27a0ec
DK
720 ret = *min;
721 *min += 1;
722 return ret;
723 }
724
725 if (*five < *min) {
726 min = five;
727 mult = 5;
728 }
729 if (*seven < *min) {
730 min = seven;
731 mult = 7;
732 }
733
734 ret = *min;
735 *min *= mult;
736
737 return ret;
738}
739
740/*
741 * Check that all of the backup GDT blocks are held in the primary GDT block.
742 * It is assumed that they are stored in group order. Returns the number of
743 * groups in current filesystem that have BACKUPS, or -ve error code.
744 */
745static int verify_reserved_gdb(struct super_block *sb,
c72df9f9 746 ext4_group_t end,
ac27a0ec
DK
747 struct buffer_head *primary)
748{
617ba13b 749 const ext4_fsblk_t blk = primary->b_blocknr;
ac27a0ec
DK
750 unsigned three = 1;
751 unsigned five = 5;
752 unsigned seven = 7;
753 unsigned grp;
754 __le32 *p = (__le32 *)primary->b_data;
755 int gdbackups = 0;
756
617ba13b 757 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
bd81d8ee
LV
758 if (le32_to_cpu(*p++) !=
759 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
12062ddd 760 ext4_warning(sb, "reserved GDT %llu"
2ae02107 761 " missing grp %d (%llu)",
ac27a0ec 762 blk, grp,
bd81d8ee
LV
763 grp *
764 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
765 blk);
ac27a0ec
DK
766 return -EINVAL;
767 }
617ba13b 768 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
ac27a0ec
DK
769 return -EFBIG;
770 }
771
772 return gdbackups;
773}
774
775/*
776 * Called when we need to bring a reserved group descriptor table block into
777 * use from the resize inode. The primary copy of the new GDT block currently
778 * is an indirect block (under the double indirect block in the resize inode).
779 * The new backup GDT blocks will be stored as leaf blocks in this indirect
780 * block, in group order. Even though we know all the block numbers we need,
781 * we check to ensure that the resize inode has actually reserved these blocks.
782 *
783 * Don't need to update the block bitmaps because the blocks are still in use.
784 *
785 * We get all of the error cases out of the way, so that we are sure to not
786 * fail once we start modifying the data on disk, because JBD has no rollback.
787 */
788static int add_new_gdb(handle_t *handle, struct inode *inode,
2f919710 789 ext4_group_t group)
ac27a0ec
DK
790{
791 struct super_block *sb = inode->i_sb;
617ba13b 792 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
2f919710 793 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
617ba13b 794 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
fb265c9c
TT
795 struct buffer_head **o_group_desc, **n_group_desc = NULL;
796 struct buffer_head *dind = NULL;
797 struct buffer_head *gdb_bh = NULL;
ac27a0ec 798 int gdbackups;
fb265c9c 799 struct ext4_iloc iloc = { .bh = NULL };
ac27a0ec
DK
800 __le32 *data;
801 int err;
802
803 if (test_opt(sb, DEBUG))
804 printk(KERN_DEBUG
617ba13b 805 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
ac27a0ec
DK
806 gdb_num);
807
fb265c9c
TT
808 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
809 if (IS_ERR(gdb_bh))
810 return PTR_ERR(gdb_bh);
ac27a0ec 811
c72df9f9 812 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
2f919710 813 if (gdbackups < 0) {
ac27a0ec 814 err = gdbackups;
fb265c9c 815 goto errout;
ac27a0ec
DK
816 }
817
617ba13b 818 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
fb265c9c
TT
819 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
820 if (IS_ERR(dind)) {
821 err = PTR_ERR(dind);
822 dind = NULL;
823 goto errout;
ac27a0ec
DK
824 }
825
826 data = (__le32 *)dind->b_data;
617ba13b 827 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
12062ddd 828 ext4_warning(sb, "new group %u GDT block %llu not reserved",
2f919710 829 group, gdblock);
ac27a0ec 830 err = -EINVAL;
fb265c9c 831 goto errout;
ac27a0ec
DK
832 }
833
5d601255 834 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
b4097142
TT
835 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
836 if (unlikely(err))
fb265c9c 837 goto errout;
ac27a0ec 838
5d601255 839 BUFFER_TRACE(gdb_bh, "get_write_access");
2f919710 840 err = ext4_journal_get_write_access(handle, gdb_bh);
b4097142 841 if (unlikely(err))
fb265c9c 842 goto errout;
ac27a0ec 843
5d601255 844 BUFFER_TRACE(dind, "get_write_access");
b4097142
TT
845 err = ext4_journal_get_write_access(handle, dind);
846 if (unlikely(err))
847 ext4_std_error(sb, err);
ac27a0ec 848
617ba13b 849 /* ext4_reserve_inode_write() gets a reference on the iloc */
b4097142
TT
850 err = ext4_reserve_inode_write(handle, inode, &iloc);
851 if (unlikely(err))
fb265c9c 852 goto errout;
ac27a0ec 853
71b565ce
TT
854 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
855 GFP_KERNEL);
ac27a0ec
DK
856 if (!n_group_desc) {
857 err = -ENOMEM;
f18a5f21
TT
858 ext4_warning(sb, "not enough memory for %lu groups",
859 gdb_num + 1);
fb265c9c 860 goto errout;
ac27a0ec
DK
861 }
862
863 /*
864 * Finally, we have all of the possible failures behind us...
865 *
866 * Remove new GDT block from inode double-indirect block and clear out
867 * the new GDT block for use (which also "frees" the backup GDT blocks
868 * from the reserved inode). We don't need to change the bitmaps for
869 * these blocks, because they are marked as in-use from being in the
870 * reserved inode, and will become GDT blocks (primary and backup).
871 */
617ba13b 872 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
b4097142
TT
873 err = ext4_handle_dirty_metadata(handle, NULL, dind);
874 if (unlikely(err)) {
875 ext4_std_error(sb, err);
fb265c9c 876 goto errout;
b4097142 877 }
d77147ff 878 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
879 (9 - EXT4_SB(sb)->s_cluster_bits);
617ba13b 880 ext4_mark_iloc_dirty(handle, inode, &iloc);
2f919710
YY
881 memset(gdb_bh->b_data, 0, sb->s_blocksize);
882 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
b4097142
TT
883 if (unlikely(err)) {
884 ext4_std_error(sb, err);
8c380ab4 885 iloc.bh = NULL;
fb265c9c 886 goto errout;
b4097142
TT
887 }
888 brelse(dind);
ac27a0ec 889
1d0c3924
TT
890 rcu_read_lock();
891 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
ac27a0ec 892 memcpy(n_group_desc, o_group_desc,
617ba13b 893 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
1d0c3924 894 rcu_read_unlock();
2f919710 895 n_group_desc[gdb_num] = gdb_bh;
1d0c3924 896 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
617ba13b 897 EXT4_SB(sb)->s_gdb_count++;
1d0c3924 898 ext4_kvfree_array_rcu(o_group_desc);
ac27a0ec 899
e8546d06 900 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
b50924c2 901 err = ext4_handle_dirty_super(handle, sb);
b4097142
TT
902 if (err)
903 ext4_std_error(sb, err);
b4097142 904 return err;
fb265c9c 905errout:
b93b41d4 906 kvfree(n_group_desc);
ac27a0ec 907 brelse(iloc.bh);
ac27a0ec 908 brelse(dind);
2f919710 909 brelse(gdb_bh);
ac27a0ec 910
617ba13b 911 ext4_debug("leaving with error %d\n", err);
ac27a0ec
DK
912 return err;
913}
914
01f795f9
YY
915/*
916 * add_new_gdb_meta_bg is the sister of add_new_gdb.
917 */
918static int add_new_gdb_meta_bg(struct super_block *sb,
919 handle_t *handle, ext4_group_t group) {
920 ext4_fsblk_t gdblock;
921 struct buffer_head *gdb_bh;
922 struct buffer_head **o_group_desc, **n_group_desc;
923 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
924 int err;
925
926 gdblock = ext4_meta_bg_first_block_no(sb, group) +
927 ext4_bg_has_super(sb, group);
fb265c9c
TT
928 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
929 if (IS_ERR(gdb_bh))
930 return PTR_ERR(gdb_bh);
71b565ce
TT
931 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
932 GFP_KERNEL);
01f795f9 933 if (!n_group_desc) {
61a9c11e 934 brelse(gdb_bh);
01f795f9
YY
935 err = -ENOMEM;
936 ext4_warning(sb, "not enough memory for %lu groups",
937 gdb_num + 1);
938 return err;
939 }
940
1d0c3924
TT
941 rcu_read_lock();
942 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
01f795f9
YY
943 memcpy(n_group_desc, o_group_desc,
944 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
1d0c3924 945 rcu_read_unlock();
01f795f9 946 n_group_desc[gdb_num] = gdb_bh;
d64264d6
LC
947
948 BUFFER_TRACE(gdb_bh, "get_write_access");
949 err = ext4_journal_get_write_access(handle, gdb_bh);
950 if (err) {
951 kvfree(n_group_desc);
952 brelse(gdb_bh);
953 return err;
954 }
955
1d0c3924 956 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
01f795f9 957 EXT4_SB(sb)->s_gdb_count++;
1d0c3924 958 ext4_kvfree_array_rcu(o_group_desc);
01f795f9
YY
959 return err;
960}
961
ac27a0ec
DK
962/*
963 * Called when we are adding a new group which has a backup copy of each of
964 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
965 * We need to add these reserved backup GDT blocks to the resize inode, so
966 * that they are kept for future resizing and not allocated to files.
967 *
968 * Each reserved backup GDT block will go into a different indirect block.
969 * The indirect blocks are actually the primary reserved GDT blocks,
970 * so we know in advance what their block numbers are. We only get the
971 * double-indirect block to verify it is pointing to the primary reserved
972 * GDT blocks so we don't overwrite a data block by accident. The reserved
973 * backup GDT blocks are stored in their reserved primary GDT block.
974 */
975static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
668f4dc5 976 ext4_group_t group)
ac27a0ec
DK
977{
978 struct super_block *sb = inode->i_sb;
617ba13b 979 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
d77147ff 980 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
ac27a0ec
DK
981 struct buffer_head **primary;
982 struct buffer_head *dind;
617ba13b
MC
983 struct ext4_iloc iloc;
984 ext4_fsblk_t blk;
ac27a0ec
DK
985 __le32 *data, *end;
986 int gdbackups = 0;
987 int res, i;
988 int err;
989
6da2ec56 990 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
ac27a0ec
DK
991 if (!primary)
992 return -ENOMEM;
993
617ba13b 994 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
fb265c9c
TT
995 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
996 if (IS_ERR(dind)) {
997 err = PTR_ERR(dind);
998 dind = NULL;
ac27a0ec
DK
999 goto exit_free;
1000 }
1001
617ba13b 1002 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
94460093
JB
1003 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1004 EXT4_ADDR_PER_BLOCK(sb));
617ba13b 1005 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
ac27a0ec
DK
1006
1007 /* Get each reserved primary GDT block and verify it holds backups */
1008 for (res = 0; res < reserved_gdb; res++, blk++) {
1009 if (le32_to_cpu(*data) != blk) {
12062ddd 1010 ext4_warning(sb, "reserved block %llu"
ac27a0ec
DK
1011 " not at offset %ld",
1012 blk,
1013 (long)(data - (__le32 *)dind->b_data));
1014 err = -EINVAL;
1015 goto exit_bh;
1016 }
fb265c9c
TT
1017 primary[res] = ext4_sb_bread(sb, blk, 0);
1018 if (IS_ERR(primary[res])) {
1019 err = PTR_ERR(primary[res]);
1020 primary[res] = NULL;
ac27a0ec
DK
1021 goto exit_bh;
1022 }
c72df9f9
YY
1023 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1024 if (gdbackups < 0) {
ac27a0ec
DK
1025 brelse(primary[res]);
1026 err = gdbackups;
1027 goto exit_bh;
1028 }
1029 if (++data >= end)
1030 data = (__le32 *)dind->b_data;
1031 }
1032
1033 for (i = 0; i < reserved_gdb; i++) {
5d601255 1034 BUFFER_TRACE(primary[i], "get_write_access");
37be2f59 1035 if ((err = ext4_journal_get_write_access(handle, primary[i])))
ac27a0ec 1036 goto exit_bh;
ac27a0ec
DK
1037 }
1038
617ba13b 1039 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
ac27a0ec
DK
1040 goto exit_bh;
1041
1042 /*
1043 * Finally we can add each of the reserved backup GDT blocks from
1044 * the new group to its reserved primary GDT block.
1045 */
668f4dc5 1046 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1047 for (i = 0; i < reserved_gdb; i++) {
1048 int err2;
1049 data = (__le32 *)primary[i]->b_data;
1050 /* printk("reserving backup %lu[%u] = %lu\n",
1051 primary[i]->b_blocknr, gdbackups,
1052 blk + primary[i]->b_blocknr); */
1053 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
0390131b 1054 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
ac27a0ec
DK
1055 if (!err)
1056 err = err2;
1057 }
d77147ff 1058
1059 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
617ba13b 1060 ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec
DK
1061
1062exit_bh:
1063 while (--res >= 0)
1064 brelse(primary[res]);
1065 brelse(dind);
1066
1067exit_free:
1068 kfree(primary);
1069
1070 return err;
1071}
1072
1073/*
617ba13b 1074 * Update the backup copies of the ext4 metadata. These don't need to be part
ac27a0ec
DK
1075 * of the main resize transaction, because e2fsck will re-write them if there
1076 * is a problem (basically only OOM will cause a problem). However, we
1077 * _should_ update the backups if possible, in case the primary gets trashed
1078 * for some reason and we need to run e2fsck from a backup superblock. The
1079 * important part is that the new block and inode counts are in the backup
1080 * superblocks, and the location of the new group metadata in the GDT backups.
1081 *
32ed5058
TT
1082 * We do not need take the s_resize_lock for this, because these
1083 * blocks are not otherwise touched by the filesystem code when it is
1084 * mounted. We don't need to worry about last changing from
1085 * sbi->s_groups_count, because the worst that can happen is that we
1086 * do not copy the full number of backups at this time. The resize
1087 * which changed s_groups_count will backup again.
ac27a0ec 1088 */
904dad47 1089static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
01f795f9 1090 int size, int meta_bg)
ac27a0ec 1091{
617ba13b 1092 struct ext4_sb_info *sbi = EXT4_SB(sb);
01f795f9 1093 ext4_group_t last;
617ba13b 1094 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
ac27a0ec
DK
1095 unsigned three = 1;
1096 unsigned five = 5;
1097 unsigned seven = 7;
01f795f9 1098 ext4_group_t group = 0;
ac27a0ec
DK
1099 int rest = sb->s_blocksize - size;
1100 handle_t *handle;
1101 int err = 0, err2;
1102
9924a92a 1103 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
ac27a0ec
DK
1104 if (IS_ERR(handle)) {
1105 group = 1;
1106 err = PTR_ERR(handle);
1107 goto exit_err;
1108 }
1109
01f795f9
YY
1110 if (meta_bg == 0) {
1111 group = ext4_list_backups(sb, &three, &five, &seven);
1112 last = sbi->s_groups_count;
1113 } else {
904dad47 1114 group = ext4_get_group_number(sb, blk_off) + 1;
01f795f9
YY
1115 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1116 }
1117
1118 while (group < sbi->s_groups_count) {
ac27a0ec 1119 struct buffer_head *bh;
01f795f9 1120 ext4_fsblk_t backup_block;
ac27a0ec
DK
1121
1122 /* Out of journal space, and can't get more - abort - so sad */
83448bdf 1123 err = ext4_resize_ensure_credits_batch(handle, 1);
a4130367 1124 if (err < 0)
ac27a0ec
DK
1125 break;
1126
01f795f9 1127 if (meta_bg == 0)
9378c676 1128 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
01f795f9
YY
1129 else
1130 backup_block = (ext4_group_first_block_no(sb, group) +
1131 ext4_bg_has_super(sb, group));
1132
1133 bh = sb_getblk(sb, backup_block);
aebf0243 1134 if (unlikely(!bh)) {
860d21e2 1135 err = -ENOMEM;
ac27a0ec
DK
1136 break;
1137 }
01f795f9
YY
1138 ext4_debug("update metadata backup %llu(+%llu)\n",
1139 backup_block, backup_block -
1140 ext4_group_first_block_no(sb, group));
5d601255 1141 BUFFER_TRACE(bh, "get_write_access");
ea0abbb6
VA
1142 if ((err = ext4_journal_get_write_access(handle, bh))) {
1143 brelse(bh);
ac27a0ec 1144 break;
ea0abbb6 1145 }
ac27a0ec
DK
1146 lock_buffer(bh);
1147 memcpy(bh->b_data, data, size);
1148 if (rest)
1149 memset(bh->b_data + size, 0, rest);
1150 set_buffer_uptodate(bh);
1151 unlock_buffer(bh);
b4097142
TT
1152 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1153 if (unlikely(err))
1154 ext4_std_error(sb, err);
ac27a0ec 1155 brelse(bh);
01f795f9
YY
1156
1157 if (meta_bg == 0)
1158 group = ext4_list_backups(sb, &three, &five, &seven);
1159 else if (group == last)
1160 break;
1161 else
1162 group = last;
ac27a0ec 1163 }
617ba13b 1164 if ((err2 = ext4_journal_stop(handle)) && !err)
ac27a0ec
DK
1165 err = err2;
1166
1167 /*
1168 * Ugh! Need to have e2fsck write the backup copies. It is too
1169 * late to revert the resize, we shouldn't fail just because of
1170 * the backup copies (they are only needed in case of corruption).
1171 *
1172 * However, if we got here we have a journal problem too, so we
1173 * can't really start a transaction to mark the superblock.
1174 * Chicken out and just set the flag on the hope it will be written
1175 * to disk, and if not - we will simply wait until next fsck.
1176 */
1177exit_err:
1178 if (err) {
12062ddd 1179 ext4_warning(sb, "can't update backup for group %u (err %d), "
ac27a0ec 1180 "forcing fsck on next reboot", group, err);
617ba13b
MC
1181 sbi->s_mount_state &= ~EXT4_VALID_FS;
1182 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
ac27a0ec
DK
1183 mark_buffer_dirty(sbi->s_sbh);
1184 }
1185}
1186
bb08c1e7
YY
1187/*
1188 * ext4_add_new_descs() adds @count group descriptor of groups
1189 * starting at @group
1190 *
1191 * @handle: journal handle
1192 * @sb: super block
1193 * @group: the group no. of the first group desc to be added
1194 * @resize_inode: the resize inode
1195 * @count: number of group descriptors to be added
1196 */
1197static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1198 ext4_group_t group, struct inode *resize_inode,
1199 ext4_group_t count)
1200{
1201 struct ext4_sb_info *sbi = EXT4_SB(sb);
1202 struct ext4_super_block *es = sbi->s_es;
1203 struct buffer_head *gdb_bh;
1204 int i, gdb_off, gdb_num, err = 0;
01f795f9 1205 int meta_bg;
bb08c1e7 1206
e2b911c5 1207 meta_bg = ext4_has_feature_meta_bg(sb);
bb08c1e7
YY
1208 for (i = 0; i < count; i++, group++) {
1209 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1210 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1211
1212 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1213 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1214
1215 /*
1216 * We will only either add reserved group blocks to a backup group
1217 * or remove reserved blocks for the first group in a new group block.
1218 * Doing both would be mean more complex code, and sane people don't
1219 * use non-sparse filesystems anymore. This is already checked above.
1220 */
1221 if (gdb_off) {
1d0c3924
TT
1222 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1223 gdb_num);
5d601255 1224 BUFFER_TRACE(gdb_bh, "get_write_access");
bb08c1e7
YY
1225 err = ext4_journal_get_write_access(handle, gdb_bh);
1226
1227 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1228 err = reserve_backup_gdb(handle, resize_inode, group);
01f795f9
YY
1229 } else if (meta_bg != 0) {
1230 err = add_new_gdb_meta_bg(sb, handle, group);
1231 } else {
bb08c1e7 1232 err = add_new_gdb(handle, resize_inode, group);
01f795f9 1233 }
bb08c1e7
YY
1234 if (err)
1235 break;
1236 }
1237 return err;
1238}
1239
41a246d1
DW
1240static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1241{
1242 struct buffer_head *bh = sb_getblk(sb, block);
aebf0243 1243 if (unlikely(!bh))
41a246d1 1244 return NULL;
7f1468d1
DM
1245 if (!bh_uptodate_or_lock(bh)) {
1246 if (bh_submit_read(bh) < 0) {
1247 brelse(bh);
1248 return NULL;
1249 }
41a246d1 1250 }
41a246d1
DW
1251
1252 return bh;
1253}
1254
1255static int ext4_set_bitmap_checksums(struct super_block *sb,
1256 ext4_group_t group,
1257 struct ext4_group_desc *gdp,
1258 struct ext4_new_group_data *group_data)
1259{
1260 struct buffer_head *bh;
1261
9aa5d32b 1262 if (!ext4_has_metadata_csum(sb))
41a246d1
DW
1263 return 0;
1264
1265 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1266 if (!bh)
1267 return -EIO;
1268 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1269 EXT4_INODES_PER_GROUP(sb) / 8);
1270 brelse(bh);
1271
fa77dcfa
DW
1272 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1273 if (!bh)
1274 return -EIO;
79f1ba49 1275 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
fa77dcfa
DW
1276 brelse(bh);
1277
41a246d1
DW
1278 return 0;
1279}
1280
083f5b24
YY
1281/*
1282 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1283 */
1284static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1285 struct ext4_new_flex_group_data *flex_gd)
1286{
1287 struct ext4_new_group_data *group_data = flex_gd->groups;
1288 struct ext4_group_desc *gdp;
1289 struct ext4_sb_info *sbi = EXT4_SB(sb);
1290 struct buffer_head *gdb_bh;
1291 ext4_group_t group;
1292 __u16 *bg_flags = flex_gd->bg_flags;
1293 int i, gdb_off, gdb_num, err = 0;
d77147ff 1294
083f5b24
YY
1295
1296 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1297 group = group_data->group;
1298
1299 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1300 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1301
1302 /*
1303 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1304 */
1d0c3924 1305 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
083f5b24 1306 /* Update group descriptor block for new group */
2716b802 1307 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
083f5b24
YY
1308 gdb_off * EXT4_DESC_SIZE(sb));
1309
1310 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1311 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1312 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
41a246d1
DW
1313 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1314 if (err) {
1315 ext4_std_error(sb, err);
1316 break;
1317 }
1318
083f5b24
YY
1319 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1320 ext4_free_group_clusters_set(sb, gdp,
d77147ff 1321 group_data->free_clusters_count);
083f5b24 1322 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
93f90526
TT
1323 if (ext4_has_group_desc_csum(sb))
1324 ext4_itable_unused_set(sb, gdp,
1325 EXT4_INODES_PER_GROUP(sb));
083f5b24 1326 gdp->bg_flags = cpu_to_le16(*bg_flags);
feb0ab32 1327 ext4_group_desc_csum_set(sb, group, gdp);
083f5b24
YY
1328
1329 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1330 if (unlikely(err)) {
1331 ext4_std_error(sb, err);
1332 break;
1333 }
1334
1335 /*
1336 * We can allocate memory for mb_alloc based on the new group
1337 * descriptor
1338 */
1339 err = ext4_mb_add_groupinfo(sb, group, gdp);
1340 if (err)
1341 break;
1342 }
1343 return err;
1344}
1345
2e10e2f2
YY
1346/*
1347 * ext4_update_super() updates the super block so that the newly added
1348 * groups can be seen by the filesystem.
1349 *
1350 * @sb: super block
1351 * @flex_gd: new added groups
1352 */
1353static void ext4_update_super(struct super_block *sb,
1354 struct ext4_new_flex_group_data *flex_gd)
1355{
1356 ext4_fsblk_t blocks_count = 0;
1357 ext4_fsblk_t free_blocks = 0;
1358 ext4_fsblk_t reserved_blocks = 0;
1359 struct ext4_new_group_data *group_data = flex_gd->groups;
1360 struct ext4_sb_info *sbi = EXT4_SB(sb);
1361 struct ext4_super_block *es = sbi->s_es;
8a991849 1362 int i;
2e10e2f2
YY
1363
1364 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1365 /*
1366 * Make the new blocks and inodes valid next. We do this before
1367 * increasing the group count so that once the group is enabled,
1368 * all of its blocks and inodes are already valid.
1369 *
1370 * We always allocate group-by-group, then block-by-block or
1371 * inode-by-inode within a group, so enabling these
1372 * blocks/inodes before the group is live won't actually let us
1373 * allocate the new space yet.
1374 */
1375 for (i = 0; i < flex_gd->count; i++) {
1376 blocks_count += group_data[i].blocks_count;
d77147ff 1377 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
2e10e2f2
YY
1378 }
1379
1380 reserved_blocks = ext4_r_blocks_count(es) * 100;
01f795f9 1381 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
2e10e2f2
YY
1382 reserved_blocks *= blocks_count;
1383 do_div(reserved_blocks, 100);
1384
1385 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
636d7e2e 1386 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
2e10e2f2
YY
1387 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1388 flex_gd->count);
636d7e2e
DW
1389 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1390 flex_gd->count);
2e10e2f2 1391
01f795f9 1392 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
2e10e2f2
YY
1393 /*
1394 * We need to protect s_groups_count against other CPUs seeing
1395 * inconsistent state in the superblock.
1396 *
1397 * The precise rules we use are:
1398 *
1399 * * Writers must perform a smp_wmb() after updating all
1400 * dependent data and before modifying the groups count
1401 *
1402 * * Readers must perform an smp_rmb() after reading the groups
1403 * count and before reading any dependent data.
1404 *
1405 * NB. These rules can be relaxed when checking the group count
1406 * while freeing data, as we can only allocate from a block
1407 * group after serialising against the group count, and we can
1408 * only then free after serialising in turn against that
1409 * allocation.
1410 */
1411 smp_wmb();
1412
1413 /* Update the global fs size fields */
1414 sbi->s_groups_count += flex_gd->count;
c5c72d81
TT
1415 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1416 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
2e10e2f2
YY
1417
1418 /* Update the reserved block counts only once the new group is
1419 * active. */
1420 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1421 reserved_blocks);
1422
1423 /* Update the free space counts */
1424 percpu_counter_add(&sbi->s_freeclusters_counter,
810da240 1425 EXT4_NUM_B2C(sbi, free_blocks));
2e10e2f2
YY
1426 percpu_counter_add(&sbi->s_freeinodes_counter,
1427 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1428
01f795f9
YY
1429 ext4_debug("free blocks count %llu",
1430 percpu_counter_read(&sbi->s_freeclusters_counter));
e2b911c5 1431 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
2e10e2f2
YY
1432 ext4_group_t flex_group;
1433 flex_group = ext4_flex_group(sbi, group_data[0].group);
90ba983f
TT
1434 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1435 &sbi->s_flex_groups[flex_group].free_clusters);
2e10e2f2
YY
1436 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1437 &sbi->s_flex_groups[flex_group].free_inodes);
1438 }
1439
952fc18e
TT
1440 /*
1441 * Update the fs overhead information
1442 */
1443 ext4_calculate_overhead(sb);
1444
2e10e2f2
YY
1445 if (test_opt(sb, DEBUG))
1446 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1447 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1448 blocks_count, free_blocks, reserved_blocks);
1449}
1450
4bac1f8c
YY
1451/* Add a flex group to an fs. Ensure we handle all possible error conditions
1452 * _before_ we start modifying the filesystem, because we cannot abort the
1453 * transaction and not have it write the data to disk.
1454 */
1455static int ext4_flex_group_add(struct super_block *sb,
1456 struct inode *resize_inode,
1457 struct ext4_new_flex_group_data *flex_gd)
1458{
1459 struct ext4_sb_info *sbi = EXT4_SB(sb);
1460 struct ext4_super_block *es = sbi->s_es;
1461 ext4_fsblk_t o_blocks_count;
1462 ext4_grpblk_t last;
1463 ext4_group_t group;
1464 handle_t *handle;
1465 unsigned reserved_gdb;
1466 int err = 0, err2 = 0, credit;
1467
1468 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1469
1470 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1471 o_blocks_count = ext4_blocks_count(es);
1472 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1473 BUG_ON(last);
1474
1475 err = setup_new_flex_group_blocks(sb, flex_gd);
1476 if (err)
1477 goto exit;
1478 /*
1479 * We will always be modifying at least the superblock and GDT
2c869b26 1480 * blocks. If we are adding a group past the last current GDT block,
4bac1f8c
YY
1481 * we will also modify the inode and the dindirect block. If we
1482 * are adding a group with superblock/GDT backups we will also
1483 * modify each of the reserved GDT dindirect blocks.
1484 */
2c869b26
JK
1485 credit = 3; /* sb, resize inode, resize inode dindirect */
1486 /* GDT blocks */
1487 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1488 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
9924a92a 1489 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
4bac1f8c
YY
1490 if (IS_ERR(handle)) {
1491 err = PTR_ERR(handle);
1492 goto exit;
1493 }
1494
5d601255 1495 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
4bac1f8c
YY
1496 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1497 if (err)
1498 goto exit_journal;
1499
1500 group = flex_gd->groups[0].group;
49598e04 1501 BUG_ON(group != sbi->s_groups_count);
4bac1f8c
YY
1502 err = ext4_add_new_descs(handle, sb, group,
1503 resize_inode, flex_gd->count);
1504 if (err)
1505 goto exit_journal;
1506
1507 err = ext4_setup_new_descs(handle, sb, flex_gd);
1508 if (err)
1509 goto exit_journal;
1510
1511 ext4_update_super(sb, flex_gd);
1512
1513 err = ext4_handle_dirty_super(handle, sb);
1514
1515exit_journal:
1516 err2 = ext4_journal_stop(handle);
1517 if (!err)
1518 err = err2;
1519
1520 if (!err) {
2ebd1704
YY
1521 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1522 int gdb_num_end = ((group + flex_gd->count - 1) /
1523 EXT4_DESC_PER_BLOCK(sb));
e2b911c5 1524 int meta_bg = ext4_has_feature_meta_bg(sb);
0acdb887 1525 sector_t old_gdb = 0;
2ebd1704 1526
4bac1f8c 1527 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
01f795f9 1528 sizeof(struct ext4_super_block), 0);
2ebd1704 1529 for (; gdb_num <= gdb_num_end; gdb_num++) {
4bac1f8c 1530 struct buffer_head *gdb_bh;
2ebd1704 1531
1d0c3924
TT
1532 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1533 gdb_num);
0acdb887
TM
1534 if (old_gdb == gdb_bh->b_blocknr)
1535 continue;
4bac1f8c 1536 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
01f795f9 1537 gdb_bh->b_size, meta_bg);
0acdb887 1538 old_gdb = gdb_bh->b_blocknr;
4bac1f8c
YY
1539 }
1540 }
1541exit:
1542 return err;
1543}
1544
19c5246d
YY
1545static int ext4_setup_next_flex_gd(struct super_block *sb,
1546 struct ext4_new_flex_group_data *flex_gd,
1547 ext4_fsblk_t n_blocks_count,
1548 unsigned long flexbg_size)
1549{
d77147ff 1550 struct ext4_sb_info *sbi = EXT4_SB(sb);
1551 struct ext4_super_block *es = sbi->s_es;
19c5246d
YY
1552 struct ext4_new_group_data *group_data = flex_gd->groups;
1553 ext4_fsblk_t o_blocks_count;
1554 ext4_group_t n_group;
1555 ext4_group_t group;
1556 ext4_group_t last_group;
1557 ext4_grpblk_t last;
d77147ff 1558 ext4_grpblk_t clusters_per_group;
19c5246d
YY
1559 unsigned long i;
1560
d77147ff 1561 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
19c5246d
YY
1562
1563 o_blocks_count = ext4_blocks_count(es);
1564
1565 if (o_blocks_count == n_blocks_count)
1566 return 0;
1567
1568 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1569 BUG_ON(last);
1570 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1571
1572 last_group = group | (flexbg_size - 1);
1573 if (last_group > n_group)
1574 last_group = n_group;
1575
1576 flex_gd->count = last_group - group + 1;
1577
1578 for (i = 0; i < flex_gd->count; i++) {
1579 int overhead;
1580
1581 group_data[i].group = group + i;
d77147ff 1582 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
01f795f9 1583 overhead = ext4_group_overhead_blocks(sb, group + i);
d77147ff 1584 group_data[i].mdata_blocks = overhead;
1585 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
7f511862 1586 if (ext4_has_group_desc_csum(sb)) {
19c5246d
YY
1587 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1588 EXT4_BG_INODE_UNINIT;
7f511862
TT
1589 if (!test_opt(sb, INIT_INODE_TABLE))
1590 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1591 } else
19c5246d
YY
1592 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1593 }
1594
feb0ab32 1595 if (last_group == n_group && ext4_has_group_desc_csum(sb))
19c5246d
YY
1596 /* We need to initialize block bitmap of last group. */
1597 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1598
d77147ff 1599 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1600 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1601 group_data[i - 1].free_clusters_count -= clusters_per_group -
1602 last - 1;
19c5246d
YY
1603 }
1604
1605 return 1;
1606}
1607
ac27a0ec
DK
1608/* Add group descriptor data to an existing or new group descriptor block.
1609 * Ensure we handle all possible error conditions _before_ we start modifying
1610 * the filesystem, because we cannot abort the transaction and not have it
1611 * write the data to disk.
1612 *
1613 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1614 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1615 *
1616 * We only need to hold the superblock lock while we are actually adding
1617 * in the new group's counts to the superblock. Prior to that we have
1618 * not really "added" the group at all. We re-check that we are still
1619 * adding in the last group in case things have changed since verifying.
1620 */
617ba13b 1621int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
ac27a0ec 1622{
61f296cc 1623 struct ext4_new_flex_group_data flex_gd;
617ba13b
MC
1624 struct ext4_sb_info *sbi = EXT4_SB(sb);
1625 struct ext4_super_block *es = sbi->s_es;
1626 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
ac27a0ec 1627 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
ac27a0ec 1628 struct inode *inode = NULL;
03b40e34 1629 int gdb_off;
61f296cc
YY
1630 int err;
1631 __u16 bg_flags = 0;
ac27a0ec 1632
617ba13b 1633 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 1634
e2b911c5 1635 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
12062ddd 1636 ext4_warning(sb, "Can't resize non-sparse filesystem further");
ac27a0ec
DK
1637 return -EPERM;
1638 }
1639
bd81d8ee
LV
1640 if (ext4_blocks_count(es) + input->blocks_count <
1641 ext4_blocks_count(es)) {
12062ddd 1642 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1643 return -EINVAL;
1644 }
1645
617ba13b 1646 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
ac27a0ec 1647 le32_to_cpu(es->s_inodes_count)) {
12062ddd 1648 ext4_warning(sb, "inodes_count overflow");
ac27a0ec
DK
1649 return -EINVAL;
1650 }
1651
1652 if (reserved_gdb || gdb_off == 0) {
e647e291 1653 if (!ext4_has_feature_resize_inode(sb) ||
e2b911c5 1654 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
12062ddd 1655 ext4_warning(sb,
ac27a0ec
DK
1656 "No reserved GDT blocks, can't resize");
1657 return -EPERM;
1658 }
8a363970 1659 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1d1fe1ee 1660 if (IS_ERR(inode)) {
12062ddd 1661 ext4_warning(sb, "Error opening resize inode");
1d1fe1ee 1662 return PTR_ERR(inode);
ac27a0ec
DK
1663 }
1664 }
1665
920313a7 1666
61f296cc 1667 err = verify_group_input(sb, input);
08c3a813 1668 if (err)
61f296cc 1669 goto out;
ac27a0ec 1670
117fff10
TT
1671 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1672 if (err)
7f511862 1673 goto out;
117fff10 1674
28623c2f
TT
1675 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1676 if (err)
1677 goto out;
1678
61f296cc
YY
1679 flex_gd.count = 1;
1680 flex_gd.groups = input;
1681 flex_gd.bg_flags = &bg_flags;
1682 err = ext4_flex_group_add(sb, inode, &flex_gd);
1683out:
ac27a0ec
DK
1684 iput(inode);
1685 return err;
617ba13b 1686} /* ext4_group_add */
ac27a0ec 1687
18e31438
YY
1688/*
1689 * extend a group without checking assuming that checking has been done.
1690 */
1691static int ext4_group_extend_no_check(struct super_block *sb,
1692 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1693{
1694 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1695 handle_t *handle;
1696 int err = 0, err2;
1697
1698 /* We will update the superblock, one block bitmap, and
1699 * one group descriptor via ext4_group_add_blocks().
1700 */
9924a92a 1701 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
18e31438
YY
1702 if (IS_ERR(handle)) {
1703 err = PTR_ERR(handle);
1704 ext4_warning(sb, "error %d on journal start", err);
1705 return err;
1706 }
1707
5d601255 1708 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
18e31438
YY
1709 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1710 if (err) {
1711 ext4_warning(sb, "error %d on journal write access", err);
1712 goto errout;
1713 }
1714
1715 ext4_blocks_count_set(es, o_blocks_count + add);
636d7e2e 1716 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
18e31438
YY
1717 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1718 o_blocks_count + add);
1719 /* We add the blocks to the bitmap and set the group need init bit */
1720 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1721 if (err)
1722 goto errout;
1723 ext4_handle_dirty_super(handle, sb);
1724 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1725 o_blocks_count + add);
1726errout:
1727 err2 = ext4_journal_stop(handle);
1728 if (err2 && !err)
1729 err = err2;
1730
1731 if (!err) {
1732 if (test_opt(sb, DEBUG))
1733 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1734 "blocks\n", ext4_blocks_count(es));
6ca792ed 1735 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
01f795f9 1736 (char *)es, sizeof(struct ext4_super_block), 0);
18e31438
YY
1737 }
1738 return err;
1739}
1740
2b2d6d01
TT
1741/*
1742 * Extend the filesystem to the new number of blocks specified. This entry
ac27a0ec
DK
1743 * point is only used to extend the current filesystem to the end of the last
1744 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1745 * for emergencies (because it has no dependencies on reserved blocks).
1746 *
617ba13b 1747 * If we _really_ wanted, we could use default values to call ext4_group_add()
ac27a0ec
DK
1748 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1749 * GDT blocks are reserved to grow to the desired size.
1750 */
617ba13b
MC
1751int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1752 ext4_fsblk_t n_blocks_count)
ac27a0ec 1753{
617ba13b 1754 ext4_fsblk_t o_blocks_count;
617ba13b
MC
1755 ext4_grpblk_t last;
1756 ext4_grpblk_t add;
af5bc92d 1757 struct buffer_head *bh;
d89651c8 1758 int err;
5f21b0e6 1759 ext4_group_t group;
ac27a0ec 1760
bd81d8ee 1761 o_blocks_count = ext4_blocks_count(es);
ac27a0ec
DK
1762
1763 if (test_opt(sb, DEBUG))
92b97816
TT
1764 ext4_msg(sb, KERN_DEBUG,
1765 "extending last group from %llu to %llu blocks",
1766 o_blocks_count, n_blocks_count);
ac27a0ec
DK
1767
1768 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1769 return 0;
1770
1771 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
92b97816
TT
1772 ext4_msg(sb, KERN_ERR,
1773 "filesystem too large to resize to %llu blocks safely",
1774 n_blocks_count);
ac27a0ec
DK
1775 return -EINVAL;
1776 }
1777
1778 if (n_blocks_count < o_blocks_count) {
12062ddd 1779 ext4_warning(sb, "can't shrink FS - resize aborted");
8f82f840 1780 return -EINVAL;
ac27a0ec
DK
1781 }
1782
1783 /* Handle the remaining blocks in the last group only. */
5f21b0e6 1784 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
ac27a0ec
DK
1785
1786 if (last == 0) {
12062ddd 1787 ext4_warning(sb, "need to use ext2online to resize further");
ac27a0ec
DK
1788 return -EPERM;
1789 }
1790
617ba13b 1791 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
ac27a0ec
DK
1792
1793 if (o_blocks_count + add < o_blocks_count) {
12062ddd 1794 ext4_warning(sb, "blocks_count overflow");
ac27a0ec
DK
1795 return -EINVAL;
1796 }
1797
1798 if (o_blocks_count + add > n_blocks_count)
1799 add = n_blocks_count - o_blocks_count;
1800
1801 if (o_blocks_count + add < n_blocks_count)
12062ddd 1802 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
ac27a0ec
DK
1803 o_blocks_count + add, add);
1804
1805 /* See if the device is actually as big as what was requested */
2b2d6d01 1806 bh = sb_bread(sb, o_blocks_count + add - 1);
ac27a0ec 1807 if (!bh) {
12062ddd 1808 ext4_warning(sb, "can't read last block, resize aborted");
ac27a0ec
DK
1809 return -ENOSPC;
1810 }
1811 brelse(bh);
1812
d89651c8 1813 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
ac27a0ec 1814 return err;
617ba13b 1815} /* ext4_group_extend */
19c5246d 1816
1c6bd717
TT
1817
1818static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1819{
1820 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1821}
1822
1823/*
1824 * Release the resize inode and drop the resize_inode feature if there
1825 * are no more reserved gdt blocks, and then convert the file system
1826 * to enable meta_bg
1827 */
1828static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1829{
1830 handle_t *handle;
1831 struct ext4_sb_info *sbi = EXT4_SB(sb);
1832 struct ext4_super_block *es = sbi->s_es;
59e31c15 1833 struct ext4_inode_info *ei = EXT4_I(inode);
1c6bd717
TT
1834 ext4_fsblk_t nr;
1835 int i, ret, err = 0;
1836 int credits = 1;
1837
1838 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
59e31c15 1839 if (inode) {
1c6bd717
TT
1840 if (es->s_reserved_gdt_blocks) {
1841 ext4_error(sb, "Unexpected non-zero "
1842 "s_reserved_gdt_blocks");
1843 return -EPERM;
1844 }
1c6bd717
TT
1845
1846 /* Do a quick sanity check of the resize inode */
d77147ff 1847 if (inode->i_blocks != 1 << (inode->i_blkbits -
1848 (9 - sbi->s_cluster_bits)))
1c6bd717
TT
1849 goto invalid_resize_inode;
1850 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1851 if (i == EXT4_DIND_BLOCK) {
1852 if (ei->i_data[i])
1853 continue;
1854 else
1855 goto invalid_resize_inode;
1856 }
1857 if (ei->i_data[i])
1858 goto invalid_resize_inode;
1859 }
1860 credits += 3; /* block bitmap, bg descriptor, resize inode */
1861 }
1862
9924a92a 1863 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1c6bd717
TT
1864 if (IS_ERR(handle))
1865 return PTR_ERR(handle);
1866
5d601255 1867 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1c6bd717
TT
1868 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1869 if (err)
1870 goto errout;
1871
e2b911c5
DW
1872 ext4_clear_feature_resize_inode(sb);
1873 ext4_set_feature_meta_bg(sb);
1c6bd717
TT
1874 sbi->s_es->s_first_meta_bg =
1875 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1876
1877 err = ext4_handle_dirty_super(handle, sb);
1878 if (err) {
1879 ext4_std_error(sb, err);
1880 goto errout;
1881 }
1882
1883 if (inode) {
1884 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1885 ext4_free_blocks(handle, inode, NULL, nr, 1,
1886 EXT4_FREE_BLOCKS_METADATA |
1887 EXT4_FREE_BLOCKS_FORGET);
1888 ei->i_data[EXT4_DIND_BLOCK] = 0;
1889 inode->i_blocks = 0;
1890
1891 err = ext4_mark_inode_dirty(handle, inode);
1892 if (err)
1893 ext4_std_error(sb, err);
1894 }
1895
1896errout:
1897 ret = ext4_journal_stop(handle);
1898 if (!err)
1899 err = ret;
1900 return ret;
1901
1902invalid_resize_inode:
1903 ext4_error(sb, "corrupted/inconsistent resize inode");
1904 return -EINVAL;
1905}
1906
19c5246d
YY
1907/*
1908 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1909 *
1910 * @sb: super block of the fs to be resized
1911 * @n_blocks_count: the number of blocks resides in the resized fs
1912 */
1913int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1914{
1915 struct ext4_new_flex_group_data *flex_gd = NULL;
1916 struct ext4_sb_info *sbi = EXT4_SB(sb);
1917 struct ext4_super_block *es = sbi->s_es;
1918 struct buffer_head *bh;
01f795f9
YY
1919 struct inode *resize_inode = NULL;
1920 ext4_grpblk_t add, offset;
19c5246d
YY
1921 unsigned long n_desc_blocks;
1922 unsigned long o_desc_blocks;
01f795f9
YY
1923 ext4_group_t o_group;
1924 ext4_group_t n_group;
1925 ext4_fsblk_t o_blocks_count;
1c6bd717 1926 ext4_fsblk_t n_blocks_count_retry = 0;
4da4a56e 1927 unsigned long last_update_time = 0;
117fff10 1928 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
01f795f9 1929 int meta_bg;
19c5246d 1930
59e31c15
TT
1931 /* See if the device is actually as big as what was requested */
1932 bh = sb_bread(sb, n_blocks_count - 1);
1933 if (!bh) {
1934 ext4_warning(sb, "can't read last block, resize aborted");
1935 return -ENOSPC;
1936 }
1937 brelse(bh);
1938
1c6bd717 1939retry:
19c5246d
YY
1940 o_blocks_count = ext4_blocks_count(es);
1941
59e31c15
TT
1942 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1943 "to %llu blocks", o_blocks_count, n_blocks_count);
19c5246d
YY
1944
1945 if (n_blocks_count < o_blocks_count) {
1946 /* On-line shrinking not supported */
1947 ext4_warning(sb, "can't shrink FS - resize aborted");
1948 return -EINVAL;
1949 }
1950
1951 if (n_blocks_count == o_blocks_count)
1952 /* Nothing need to do */
1953 return 0;
1954
bd86298e 1955 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
4f2f76f7 1956 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
3f8a6411
TT
1957 ext4_warning(sb, "resize would cause inodes_count overflow");
1958 return -EINVAL;
1959 }
a0ade1de 1960 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
19c5246d 1961
1c6bd717
TT
1962 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1963 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
19c5246d 1964
e2b911c5 1965 meta_bg = ext4_has_feature_meta_bg(sb);
19c5246d 1966
e2b911c5 1967 if (ext4_has_feature_resize_inode(sb)) {
01f795f9
YY
1968 if (meta_bg) {
1969 ext4_error(sb, "resize_inode and meta_bg enabled "
1970 "simultaneously");
1971 return -EINVAL;
1972 }
1c6bd717
TT
1973 if (n_desc_blocks > o_desc_blocks +
1974 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1975 n_blocks_count_retry = n_blocks_count;
1976 n_desc_blocks = o_desc_blocks +
1977 le16_to_cpu(es->s_reserved_gdt_blocks);
1978 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
aec51758 1979 n_blocks_count = (ext4_fsblk_t)n_group *
f96c3ac8
JK
1980 EXT4_BLOCKS_PER_GROUP(sb) +
1981 le32_to_cpu(es->s_first_data_block);
1c6bd717 1982 n_group--; /* set to last group number */
01f795f9 1983 }
1c6bd717
TT
1984
1985 if (!resize_inode)
8a363970
TT
1986 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1987 EXT4_IGET_SPECIAL);
01f795f9
YY
1988 if (IS_ERR(resize_inode)) {
1989 ext4_warning(sb, "Error opening resize inode");
1990 return PTR_ERR(resize_inode);
1991 }
1c6bd717
TT
1992 }
1993
59e31c15 1994 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1c6bd717
TT
1995 err = ext4_convert_meta_bg(sb, resize_inode);
1996 if (err)
1997 goto out;
1998 if (resize_inode) {
1999 iput(resize_inode);
2000 resize_inode = NULL;
2001 }
2002 if (n_blocks_count_retry) {
2003 n_blocks_count = n_blocks_count_retry;
2004 n_blocks_count_retry = 0;
2005 goto retry;
2006 }
19c5246d
YY
2007 }
2008
f0a459de
TT
2009 /*
2010 * Make sure the last group has enough space so that it's
2011 * guaranteed to have enough space for all metadata blocks
2012 * that it might need to hold. (We might not need to store
2013 * the inode table blocks in the last block group, but there
2014 * will be cases where this might be needed.)
2015 */
2016 if ((ext4_group_first_block_no(sb, n_group) +
2017 ext4_group_overhead_blocks(sb, n_group) + 2 +
2018 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2019 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2020 n_group--;
2021 n_blocks_count_retry = 0;
2022 if (resize_inode) {
2023 iput(resize_inode);
2024 resize_inode = NULL;
2025 }
2026 goto retry;
2027 }
2028
a0ade1de
LC
2029 /* extend the last group */
2030 if (n_group == o_group)
2031 add = n_blocks_count - o_blocks_count;
2032 else
d77147ff 2033 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
a0ade1de 2034 if (add > 0) {
19c5246d
YY
2035 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2036 if (err)
2037 goto out;
2038 }
2039
d7574ad0 2040 if (ext4_blocks_count(es) == n_blocks_count)
19c5246d
YY
2041 goto out;
2042
117fff10
TT
2043 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2044 if (err)
f348e224 2045 goto out;
117fff10 2046
28623c2f
TT
2047 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2048 if (err)
2049 goto out;
2050
19c5246d
YY
2051 flex_gd = alloc_flex_gd(flexbg_size);
2052 if (flex_gd == NULL) {
2053 err = -ENOMEM;
2054 goto out;
2055 }
2056
2057 /* Add flex groups. Note that a regular group is a
2058 * flex group with 1 group.
2059 */
2060 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2061 flexbg_size)) {
4da4a56e
TT
2062 if (jiffies - last_update_time > HZ * 10) {
2063 if (last_update_time)
2064 ext4_msg(sb, KERN_INFO,
2065 "resized to %llu blocks",
2066 ext4_blocks_count(es));
2067 last_update_time = jiffies;
2068 }
03c1c290
YY
2069 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2070 break;
19c5246d
YY
2071 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2072 if (unlikely(err))
2073 break;
2074 }
2075
1c6bd717
TT
2076 if (!err && n_blocks_count_retry) {
2077 n_blocks_count = n_blocks_count_retry;
2078 n_blocks_count_retry = 0;
2079 free_flex_gd(flex_gd);
2080 flex_gd = NULL;
db6aee62
VA
2081 if (resize_inode) {
2082 iput(resize_inode);
2083 resize_inode = NULL;
2084 }
1c6bd717
TT
2085 goto retry;
2086 }
2087
19c5246d
YY
2088out:
2089 if (flex_gd)
2090 free_flex_gd(flex_gd);
01f795f9
YY
2091 if (resize_inode != NULL)
2092 iput(resize_inode);
6c732840
LC
2093 if (err)
2094 ext4_warning(sb, "error (%d) occurred during "
2095 "file system resize", err);
2096 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2097 ext4_blocks_count(es));
19c5246d
YY
2098 return err;
2099}