blob: 6513079c728be35448812ca120127f390f62a2fc [file] [log] [blame]
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/resize.c
4 *
5 * Support for resizing an ext4 filesystem while it is mounted.
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
13#define EXT4FS_DEBUG
14
15#include <linux/errno.h>
16#include <linux/slab.h>
17
18#include "ext4_jbd2.h"
19
Olivier Deprez0e641232021-09-23 10:07:05 +020020struct 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
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000047int ext4_resize_begin(struct super_block *sb)
48{
49 struct ext4_sb_info *sbi = EXT4_SB(sb);
50 int ret = 0;
51
52 if (!capable(CAP_SYS_RESOURCE))
53 return -EPERM;
54
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 */
60 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
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
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, "
73 "so online resizing is not allowed");
74 return -EPERM;
75 }
76
Olivier Deprez157378f2022-04-04 15:47:50 +020077 if (ext4_has_feature_sparse_super2(sb)) {
78 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
79 return -EOPNOTSUPP;
80 }
81
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000082 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
83 &EXT4_SB(sb)->s_ext4_flags))
84 ret = -EBUSY;
85
86 return ret;
87}
88
89void ext4_resize_end(struct super_block *sb)
90{
91 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
92 smp_mb__after_atomic();
93}
94
95static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
96 ext4_group_t group) {
97 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
98 EXT4_DESC_PER_BLOCK_BITS(sb);
99}
100
101static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
102 ext4_group_t group) {
103 group = ext4_meta_bg_first_group(sb, group);
104 return ext4_group_first_block_no(sb, group);
105}
106
107static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
108 ext4_group_t group) {
109 ext4_grpblk_t overhead;
110 overhead = ext4_bg_num_gdb(sb, group);
111 if (ext4_bg_has_super(sb, group))
112 overhead += 1 +
113 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
114 return overhead;
115}
116
117#define outside(b, first, last) ((b) < (first) || (b) >= (last))
118#define inside(b, first, last) ((b) >= (first) && (b) < (last))
119
120static int verify_group_input(struct super_block *sb,
121 struct ext4_new_group_data *input)
122{
123 struct ext4_sb_info *sbi = EXT4_SB(sb);
124 struct ext4_super_block *es = sbi->s_es;
125 ext4_fsblk_t start = ext4_blocks_count(es);
126 ext4_fsblk_t end = start + input->blocks_count;
127 ext4_group_t group = input->group;
128 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
129 unsigned overhead;
130 ext4_fsblk_t metaend;
131 struct buffer_head *bh = NULL;
132 ext4_grpblk_t free_blocks_count, offset;
133 int err = -EINVAL;
134
135 if (group != sbi->s_groups_count) {
136 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
137 input->group, sbi->s_groups_count);
138 return -EINVAL;
139 }
140
141 overhead = ext4_group_overhead_blocks(sb, group);
142 metaend = start + overhead;
143 input->free_clusters_count = free_blocks_count =
144 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
145
146 if (test_opt(sb, DEBUG))
147 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
148 "(%d free, %u reserved)\n",
149 ext4_bg_has_super(sb, input->group) ? "normal" :
150 "no-super", input->group, input->blocks_count,
151 free_blocks_count, input->reserved_blocks);
152
153 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
154 if (offset != 0)
155 ext4_warning(sb, "Last group not full");
156 else if (input->reserved_blocks > input->blocks_count / 5)
157 ext4_warning(sb, "Reserved blocks too high (%u)",
158 input->reserved_blocks);
159 else if (free_blocks_count < 0)
160 ext4_warning(sb, "Bad blocks count %u",
161 input->blocks_count);
David Brazdil0f672f62019-12-10 10:32:29 +0000162 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
163 err = PTR_ERR(bh);
164 bh = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000165 ext4_warning(sb, "Cannot read last block (%llu)",
166 end - 1);
David Brazdil0f672f62019-12-10 10:32:29 +0000167 } else if (outside(input->block_bitmap, start, end))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000168 ext4_warning(sb, "Block bitmap not in group (block %llu)",
169 (unsigned long long)input->block_bitmap);
170 else if (outside(input->inode_bitmap, start, end))
171 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
172 (unsigned long long)input->inode_bitmap);
173 else if (outside(input->inode_table, start, end) ||
174 outside(itend - 1, start, end))
175 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
176 (unsigned long long)input->inode_table, itend - 1);
177 else if (input->inode_bitmap == input->block_bitmap)
178 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (inside(input->block_bitmap, input->inode_table, itend))
181 ext4_warning(sb, "Block bitmap (%llu) in inode table "
182 "(%llu-%llu)",
183 (unsigned long long)input->block_bitmap,
184 (unsigned long long)input->inode_table, itend - 1);
185 else if (inside(input->inode_bitmap, input->inode_table, itend))
186 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
187 "(%llu-%llu)",
188 (unsigned long long)input->inode_bitmap,
189 (unsigned long long)input->inode_table, itend - 1);
190 else if (inside(input->block_bitmap, start, metaend))
191 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
192 (unsigned long long)input->block_bitmap,
193 start, metaend - 1);
194 else if (inside(input->inode_bitmap, start, metaend))
195 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
196 (unsigned long long)input->inode_bitmap,
197 start, metaend - 1);
198 else if (inside(input->inode_table, start, metaend) ||
199 inside(itend - 1, start, metaend))
200 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
201 "(%llu-%llu)",
202 (unsigned long long)input->inode_table,
203 itend - 1, start, metaend - 1);
204 else
205 err = 0;
206 brelse(bh);
207
208 return err;
209}
210
211/*
212 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
213 * group each time.
214 */
215struct ext4_new_flex_group_data {
216 struct ext4_new_group_data *groups; /* new_group_data for groups
217 in the flex group */
218 __u16 *bg_flags; /* block group flags of groups
219 in @groups */
220 ext4_group_t count; /* number of groups in @groups
221 */
222};
223
224/*
225 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
226 * @flexbg_size.
227 *
228 * Returns NULL on failure otherwise address of the allocated structure.
229 */
230static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
231{
232 struct ext4_new_flex_group_data *flex_gd;
233
234 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
235 if (flex_gd == NULL)
236 goto out3;
237
238 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
239 goto out2;
240 flex_gd->count = flexbg_size;
241
242 flex_gd->groups = kmalloc_array(flexbg_size,
243 sizeof(struct ext4_new_group_data),
244 GFP_NOFS);
245 if (flex_gd->groups == NULL)
246 goto out2;
247
248 flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
249 GFP_NOFS);
250 if (flex_gd->bg_flags == NULL)
251 goto out1;
252
253 return flex_gd;
254
255out1:
256 kfree(flex_gd->groups);
257out2:
258 kfree(flex_gd);
259out3:
260 return NULL;
261}
262
263static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
264{
265 kfree(flex_gd->bg_flags);
266 kfree(flex_gd->groups);
267 kfree(flex_gd);
268}
269
270/*
271 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
272 * and inode tables for a flex group.
273 *
274 * This function is used by 64bit-resize. Note that this function allocates
275 * group tables from the 1st group of groups contained by @flexgd, which may
276 * be a partial of a flex group.
277 *
278 * @sb: super block of fs to which the groups belongs
279 *
280 * Returns 0 on a successful allocation of the metadata blocks in the
281 * block group.
282 */
283static int ext4_alloc_group_tables(struct super_block *sb,
284 struct ext4_new_flex_group_data *flex_gd,
285 int flexbg_size)
286{
287 struct ext4_new_group_data *group_data = flex_gd->groups;
288 ext4_fsblk_t start_blk;
289 ext4_fsblk_t last_blk;
290 ext4_group_t src_group;
291 ext4_group_t bb_index = 0;
292 ext4_group_t ib_index = 0;
293 ext4_group_t it_index = 0;
294 ext4_group_t group;
295 ext4_group_t last_group;
296 unsigned overhead;
297 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
298 int i;
299
300 BUG_ON(flex_gd->count == 0 || group_data == NULL);
301
302 src_group = group_data[0].group;
303 last_group = src_group + flex_gd->count - 1;
304
305 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
306 (last_group & ~(flexbg_size - 1))));
307next_group:
308 group = group_data[0].group;
309 if (src_group >= group_data[0].group + flex_gd->count)
310 return -ENOSPC;
311 start_blk = ext4_group_first_block_no(sb, src_group);
312 last_blk = start_blk + group_data[src_group - group].blocks_count;
313
314 overhead = ext4_group_overhead_blocks(sb, src_group);
315
316 start_blk += overhead;
317
318 /* We collect contiguous blocks as much as possible. */
319 src_group++;
320 for (; src_group <= last_group; src_group++) {
321 overhead = ext4_group_overhead_blocks(sb, src_group);
322 if (overhead == 0)
323 last_blk += group_data[src_group - group].blocks_count;
324 else
325 break;
326 }
327
328 /* Allocate block bitmaps */
329 for (; bb_index < flex_gd->count; bb_index++) {
330 if (start_blk >= last_blk)
331 goto next_group;
332 group_data[bb_index].block_bitmap = start_blk++;
333 group = ext4_get_group_number(sb, start_blk - 1);
334 group -= group_data[0].group;
335 group_data[group].mdata_blocks++;
336 flex_gd->bg_flags[group] &= uninit_mask;
337 }
338
339 /* Allocate inode bitmaps */
340 for (; ib_index < flex_gd->count; ib_index++) {
341 if (start_blk >= last_blk)
342 goto next_group;
343 group_data[ib_index].inode_bitmap = start_blk++;
344 group = ext4_get_group_number(sb, start_blk - 1);
345 group -= group_data[0].group;
346 group_data[group].mdata_blocks++;
347 flex_gd->bg_flags[group] &= uninit_mask;
348 }
349
350 /* Allocate inode tables */
351 for (; it_index < flex_gd->count; it_index++) {
352 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
353 ext4_fsblk_t next_group_start;
354
355 if (start_blk + itb > last_blk)
356 goto next_group;
357 group_data[it_index].inode_table = start_blk;
358 group = ext4_get_group_number(sb, start_blk);
359 next_group_start = ext4_group_first_block_no(sb, group + 1);
360 group -= group_data[0].group;
361
362 if (start_blk + itb > next_group_start) {
363 flex_gd->bg_flags[group + 1] &= uninit_mask;
364 overhead = start_blk + itb - next_group_start;
365 group_data[group + 1].mdata_blocks += overhead;
366 itb -= overhead;
367 }
368
369 group_data[group].mdata_blocks += itb;
370 flex_gd->bg_flags[group] &= uninit_mask;
371 start_blk += EXT4_SB(sb)->s_itb_per_group;
372 }
373
374 /* Update free clusters count to exclude metadata blocks */
375 for (i = 0; i < flex_gd->count; i++) {
376 group_data[i].free_clusters_count -=
377 EXT4_NUM_B2C(EXT4_SB(sb),
378 group_data[i].mdata_blocks);
379 }
380
381 if (test_opt(sb, DEBUG)) {
382 int i;
383 group = group_data[0].group;
384
385 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
386 "%d groups, flexbg size is %d:\n", flex_gd->count,
387 flexbg_size);
388
389 for (i = 0; i < flex_gd->count; i++) {
390 ext4_debug(
391 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
392 ext4_bg_has_super(sb, group + i) ? "normal" :
393 "no-super", group + i,
394 group_data[i].blocks_count,
395 group_data[i].free_clusters_count,
396 group_data[i].mdata_blocks);
397 }
398 }
399 return 0;
400}
401
402static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
403 ext4_fsblk_t blk)
404{
405 struct buffer_head *bh;
406 int err;
407
408 bh = sb_getblk(sb, blk);
409 if (unlikely(!bh))
410 return ERR_PTR(-ENOMEM);
411 BUFFER_TRACE(bh, "get_write_access");
412 if ((err = ext4_journal_get_write_access(handle, bh))) {
413 brelse(bh);
414 bh = ERR_PTR(err);
415 } else {
416 memset(bh->b_data, 0, sb->s_blocksize);
417 set_buffer_uptodate(bh);
418 }
419
420 return bh;
421}
422
Olivier Deprez157378f2022-04-04 15:47:50 +0200423static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000424{
Olivier Deprez157378f2022-04-04 15:47:50 +0200425 return ext4_journal_ensure_credits_fn(handle, credits,
426 EXT4_MAX_TRANS_DATA, 0, 0);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000427}
428
429/*
430 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
431 *
432 * Helper function for ext4_setup_new_group_blocks() which set .
433 *
434 * @sb: super block
435 * @handle: journal handle
436 * @flex_gd: flex group data
437 */
438static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
439 struct ext4_new_flex_group_data *flex_gd,
440 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
441{
442 struct ext4_sb_info *sbi = EXT4_SB(sb);
443 ext4_group_t count = last_cluster - first_cluster + 1;
444 ext4_group_t count2;
445
446 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
447 last_cluster);
448 for (count2 = count; count > 0;
449 count -= count2, first_cluster += count2) {
450 ext4_fsblk_t start;
451 struct buffer_head *bh;
452 ext4_group_t group;
453 int err;
454
455 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
456 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
457 group -= flex_gd->groups[0].group;
458
459 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
460 if (count2 > count)
461 count2 = count;
462
463 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
464 BUG_ON(flex_gd->count > 1);
465 continue;
466 }
467
Olivier Deprez157378f2022-04-04 15:47:50 +0200468 err = ext4_resize_ensure_credits_batch(handle, 1);
469 if (err < 0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000470 return err;
471
472 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
473 if (unlikely(!bh))
474 return -ENOMEM;
475
476 BUFFER_TRACE(bh, "get_write_access");
477 err = ext4_journal_get_write_access(handle, bh);
478 if (err) {
479 brelse(bh);
480 return err;
481 }
482 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
483 first_cluster, first_cluster - start, count2);
484 ext4_set_bits(bh->b_data, first_cluster - start, count2);
485
486 err = ext4_handle_dirty_metadata(handle, NULL, bh);
487 brelse(bh);
488 if (unlikely(err))
489 return err;
490 }
491
492 return 0;
493}
494
495/*
496 * Set up the block and inode bitmaps, and the inode table for the new groups.
497 * This doesn't need to be part of the main transaction, since we are only
498 * changing blocks outside the actual filesystem. We still do journaling to
499 * ensure the recovery is correct in case of a failure just after resize.
500 * If any part of this fails, we simply abort the resize.
501 *
502 * setup_new_flex_group_blocks handles a flex group as follow:
503 * 1. copy super block and GDT, and initialize group tables if necessary.
504 * In this step, we only set bits in blocks bitmaps for blocks taken by
505 * super block and GDT.
506 * 2. allocate group tables in block bitmaps, that is, set bits in block
507 * bitmap for blocks taken by group tables.
508 */
509static int setup_new_flex_group_blocks(struct super_block *sb,
510 struct ext4_new_flex_group_data *flex_gd)
511{
512 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
513 ext4_fsblk_t start;
514 ext4_fsblk_t block;
515 struct ext4_sb_info *sbi = EXT4_SB(sb);
516 struct ext4_super_block *es = sbi->s_es;
517 struct ext4_new_group_data *group_data = flex_gd->groups;
518 __u16 *bg_flags = flex_gd->bg_flags;
519 handle_t *handle;
520 ext4_group_t group, count;
521 struct buffer_head *bh = NULL;
522 int reserved_gdb, i, j, err = 0, err2;
523 int meta_bg;
524
525 BUG_ON(!flex_gd->count || !group_data ||
526 group_data[0].group != sbi->s_groups_count);
527
528 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
529 meta_bg = ext4_has_feature_meta_bg(sb);
530
531 /* This transaction may be extended/restarted along the way */
532 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
533 if (IS_ERR(handle))
534 return PTR_ERR(handle);
535
536 group = group_data[0].group;
537 for (i = 0; i < flex_gd->count; i++, group++) {
538 unsigned long gdblocks;
539 ext4_grpblk_t overhead;
540
541 gdblocks = ext4_bg_num_gdb(sb, group);
542 start = ext4_group_first_block_no(sb, group);
543
544 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
545 goto handle_itb;
546
547 if (meta_bg == 1) {
548 ext4_group_t first_group;
549 first_group = ext4_meta_bg_first_group(sb, group);
550 if (first_group != group + 1 &&
551 first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
552 goto handle_itb;
553 }
554
555 block = start + ext4_bg_has_super(sb, group);
556 /* Copy all of the GDT blocks into the backup in this group */
557 for (j = 0; j < gdblocks; j++, block++) {
558 struct buffer_head *gdb;
559
560 ext4_debug("update backup group %#04llx\n", block);
Olivier Deprez157378f2022-04-04 15:47:50 +0200561 err = ext4_resize_ensure_credits_batch(handle, 1);
562 if (err < 0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000563 goto out;
564
565 gdb = sb_getblk(sb, block);
566 if (unlikely(!gdb)) {
567 err = -ENOMEM;
568 goto out;
569 }
570
571 BUFFER_TRACE(gdb, "get_write_access");
572 err = ext4_journal_get_write_access(handle, gdb);
573 if (err) {
574 brelse(gdb);
575 goto out;
576 }
Olivier Deprez0e641232021-09-23 10:07:05 +0200577 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
578 s_group_desc, j)->b_data, gdb->b_size);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000579 set_buffer_uptodate(gdb);
580
581 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
582 if (unlikely(err)) {
583 brelse(gdb);
584 goto out;
585 }
586 brelse(gdb);
587 }
588
589 /* Zero out all of the reserved backup group descriptor
590 * table blocks
591 */
592 if (ext4_bg_has_super(sb, group)) {
593 err = sb_issue_zeroout(sb, gdblocks + start + 1,
594 reserved_gdb, GFP_NOFS);
595 if (err)
596 goto out;
597 }
598
599handle_itb:
600 /* Initialize group tables of the grop @group */
601 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
602 goto handle_bb;
603
604 /* Zero out all of the inode table blocks */
605 block = group_data[i].inode_table;
606 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
607 block, sbi->s_itb_per_group);
608 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
609 GFP_NOFS);
610 if (err)
611 goto out;
612
613handle_bb:
614 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
615 goto handle_ib;
616
617 /* Initialize block bitmap of the @group */
618 block = group_data[i].block_bitmap;
Olivier Deprez157378f2022-04-04 15:47:50 +0200619 err = ext4_resize_ensure_credits_batch(handle, 1);
620 if (err < 0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000621 goto out;
622
623 bh = bclean(handle, sb, block);
624 if (IS_ERR(bh)) {
625 err = PTR_ERR(bh);
626 goto out;
627 }
628 overhead = ext4_group_overhead_blocks(sb, group);
629 if (overhead != 0) {
630 ext4_debug("mark backup superblock %#04llx (+0)\n",
631 start);
632 ext4_set_bits(bh->b_data, 0,
633 EXT4_NUM_B2C(sbi, overhead));
634 }
635 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
636 sb->s_blocksize * 8, bh->b_data);
637 err = ext4_handle_dirty_metadata(handle, NULL, bh);
638 brelse(bh);
639 if (err)
640 goto out;
641
642handle_ib:
643 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
644 continue;
645
646 /* Initialize inode bitmap of the @group */
647 block = group_data[i].inode_bitmap;
Olivier Deprez157378f2022-04-04 15:47:50 +0200648 err = ext4_resize_ensure_credits_batch(handle, 1);
649 if (err < 0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000650 goto out;
651 /* Mark unused entries in inode bitmap used */
652 bh = bclean(handle, sb, block);
653 if (IS_ERR(bh)) {
654 err = PTR_ERR(bh);
655 goto out;
656 }
657
658 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
659 sb->s_blocksize * 8, bh->b_data);
660 err = ext4_handle_dirty_metadata(handle, NULL, bh);
661 brelse(bh);
662 if (err)
663 goto out;
664 }
665
666 /* Mark group tables in block bitmap */
667 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
668 count = group_table_count[j];
669 start = (&group_data[0].block_bitmap)[j];
670 block = start;
671 for (i = 1; i < flex_gd->count; i++) {
672 block += group_table_count[j];
673 if (block == (&group_data[i].block_bitmap)[j]) {
674 count += group_table_count[j];
675 continue;
676 }
677 err = set_flexbg_block_bitmap(sb, handle,
678 flex_gd,
679 EXT4_B2C(sbi, start),
680 EXT4_B2C(sbi,
681 start + count
682 - 1));
683 if (err)
684 goto out;
685 count = group_table_count[j];
686 start = (&group_data[i].block_bitmap)[j];
687 block = start;
688 }
689
690 if (count) {
691 err = set_flexbg_block_bitmap(sb, handle,
692 flex_gd,
693 EXT4_B2C(sbi, start),
694 EXT4_B2C(sbi,
695 start + count
696 - 1));
697 if (err)
698 goto out;
699 }
700 }
701
702out:
703 err2 = ext4_journal_stop(handle);
704 if (err2 && !err)
705 err = err2;
706
707 return err;
708}
709
710/*
711 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
712 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
713 * calling this for the first time. In a sparse filesystem it will be the
714 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
715 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
716 */
717static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
718 unsigned *five, unsigned *seven)
719{
720 unsigned *min = three;
721 int mult = 3;
722 unsigned ret;
723
724 if (!ext4_has_feature_sparse_super(sb)) {
725 ret = *min;
726 *min += 1;
727 return ret;
728 }
729
730 if (*five < *min) {
731 min = five;
732 mult = 5;
733 }
734 if (*seven < *min) {
735 min = seven;
736 mult = 7;
737 }
738
739 ret = *min;
740 *min *= mult;
741
742 return ret;
743}
744
745/*
746 * Check that all of the backup GDT blocks are held in the primary GDT block.
747 * It is assumed that they are stored in group order. Returns the number of
748 * groups in current filesystem that have BACKUPS, or -ve error code.
749 */
750static int verify_reserved_gdb(struct super_block *sb,
751 ext4_group_t end,
752 struct buffer_head *primary)
753{
754 const ext4_fsblk_t blk = primary->b_blocknr;
755 unsigned three = 1;
756 unsigned five = 5;
757 unsigned seven = 7;
758 unsigned grp;
759 __le32 *p = (__le32 *)primary->b_data;
760 int gdbackups = 0;
761
762 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
763 if (le32_to_cpu(*p++) !=
764 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
765 ext4_warning(sb, "reserved GDT %llu"
766 " missing grp %d (%llu)",
767 blk, grp,
768 grp *
769 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
770 blk);
771 return -EINVAL;
772 }
773 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
774 return -EFBIG;
775 }
776
777 return gdbackups;
778}
779
780/*
781 * Called when we need to bring a reserved group descriptor table block into
782 * use from the resize inode. The primary copy of the new GDT block currently
783 * is an indirect block (under the double indirect block in the resize inode).
784 * The new backup GDT blocks will be stored as leaf blocks in this indirect
785 * block, in group order. Even though we know all the block numbers we need,
786 * we check to ensure that the resize inode has actually reserved these blocks.
787 *
788 * Don't need to update the block bitmaps because the blocks are still in use.
789 *
790 * We get all of the error cases out of the way, so that we are sure to not
791 * fail once we start modifying the data on disk, because JBD has no rollback.
792 */
793static int add_new_gdb(handle_t *handle, struct inode *inode,
794 ext4_group_t group)
795{
796 struct super_block *sb = inode->i_sb;
797 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
798 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
799 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
David Brazdil0f672f62019-12-10 10:32:29 +0000800 struct buffer_head **o_group_desc, **n_group_desc = NULL;
801 struct buffer_head *dind = NULL;
802 struct buffer_head *gdb_bh = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000803 int gdbackups;
David Brazdil0f672f62019-12-10 10:32:29 +0000804 struct ext4_iloc iloc = { .bh = NULL };
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000805 __le32 *data;
806 int err;
807
808 if (test_opt(sb, DEBUG))
809 printk(KERN_DEBUG
810 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
811 gdb_num);
812
David Brazdil0f672f62019-12-10 10:32:29 +0000813 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
814 if (IS_ERR(gdb_bh))
815 return PTR_ERR(gdb_bh);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000816
817 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
818 if (gdbackups < 0) {
819 err = gdbackups;
David Brazdil0f672f62019-12-10 10:32:29 +0000820 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000821 }
822
823 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
David Brazdil0f672f62019-12-10 10:32:29 +0000824 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
825 if (IS_ERR(dind)) {
826 err = PTR_ERR(dind);
827 dind = NULL;
828 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000829 }
830
831 data = (__le32 *)dind->b_data;
832 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
833 ext4_warning(sb, "new group %u GDT block %llu not reserved",
834 group, gdblock);
835 err = -EINVAL;
David Brazdil0f672f62019-12-10 10:32:29 +0000836 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000837 }
838
839 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
840 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
841 if (unlikely(err))
David Brazdil0f672f62019-12-10 10:32:29 +0000842 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000843
844 BUFFER_TRACE(gdb_bh, "get_write_access");
845 err = ext4_journal_get_write_access(handle, gdb_bh);
846 if (unlikely(err))
David Brazdil0f672f62019-12-10 10:32:29 +0000847 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000848
849 BUFFER_TRACE(dind, "get_write_access");
850 err = ext4_journal_get_write_access(handle, dind);
Olivier Deprez0e641232021-09-23 10:07:05 +0200851 if (unlikely(err)) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000852 ext4_std_error(sb, err);
Olivier Deprez0e641232021-09-23 10:07:05 +0200853 goto errout;
854 }
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000855
856 /* ext4_reserve_inode_write() gets a reference on the iloc */
857 err = ext4_reserve_inode_write(handle, inode, &iloc);
858 if (unlikely(err))
David Brazdil0f672f62019-12-10 10:32:29 +0000859 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000860
Olivier Deprez157378f2022-04-04 15:47:50 +0200861 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
862 GFP_KERNEL);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000863 if (!n_group_desc) {
864 err = -ENOMEM;
865 ext4_warning(sb, "not enough memory for %lu groups",
866 gdb_num + 1);
David Brazdil0f672f62019-12-10 10:32:29 +0000867 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000868 }
869
870 /*
871 * Finally, we have all of the possible failures behind us...
872 *
873 * Remove new GDT block from inode double-indirect block and clear out
874 * the new GDT block for use (which also "frees" the backup GDT blocks
875 * from the reserved inode). We don't need to change the bitmaps for
876 * these blocks, because they are marked as in-use from being in the
877 * reserved inode, and will become GDT blocks (primary and backup).
878 */
879 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
880 err = ext4_handle_dirty_metadata(handle, NULL, dind);
881 if (unlikely(err)) {
882 ext4_std_error(sb, err);
David Brazdil0f672f62019-12-10 10:32:29 +0000883 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000884 }
885 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
886 (9 - EXT4_SB(sb)->s_cluster_bits);
887 ext4_mark_iloc_dirty(handle, inode, &iloc);
888 memset(gdb_bh->b_data, 0, sb->s_blocksize);
889 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
890 if (unlikely(err)) {
891 ext4_std_error(sb, err);
892 iloc.bh = NULL;
David Brazdil0f672f62019-12-10 10:32:29 +0000893 goto errout;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000894 }
895 brelse(dind);
896
Olivier Deprez0e641232021-09-23 10:07:05 +0200897 rcu_read_lock();
898 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000899 memcpy(n_group_desc, o_group_desc,
900 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Olivier Deprez0e641232021-09-23 10:07:05 +0200901 rcu_read_unlock();
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000902 n_group_desc[gdb_num] = gdb_bh;
Olivier Deprez0e641232021-09-23 10:07:05 +0200903 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000904 EXT4_SB(sb)->s_gdb_count++;
Olivier Deprez0e641232021-09-23 10:07:05 +0200905 ext4_kvfree_array_rcu(o_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000906
907 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
908 err = ext4_handle_dirty_super(handle, sb);
909 if (err)
910 ext4_std_error(sb, err);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000911 return err;
David Brazdil0f672f62019-12-10 10:32:29 +0000912errout:
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000913 kvfree(n_group_desc);
914 brelse(iloc.bh);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000915 brelse(dind);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000916 brelse(gdb_bh);
917
918 ext4_debug("leaving with error %d\n", err);
919 return err;
920}
921
922/*
923 * add_new_gdb_meta_bg is the sister of add_new_gdb.
924 */
925static int add_new_gdb_meta_bg(struct super_block *sb,
926 handle_t *handle, ext4_group_t group) {
927 ext4_fsblk_t gdblock;
928 struct buffer_head *gdb_bh;
929 struct buffer_head **o_group_desc, **n_group_desc;
930 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
931 int err;
932
933 gdblock = ext4_meta_bg_first_block_no(sb, group) +
934 ext4_bg_has_super(sb, group);
David Brazdil0f672f62019-12-10 10:32:29 +0000935 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
936 if (IS_ERR(gdb_bh))
937 return PTR_ERR(gdb_bh);
Olivier Deprez157378f2022-04-04 15:47:50 +0200938 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
939 GFP_KERNEL);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000940 if (!n_group_desc) {
941 brelse(gdb_bh);
942 err = -ENOMEM;
943 ext4_warning(sb, "not enough memory for %lu groups",
944 gdb_num + 1);
945 return err;
946 }
947
Olivier Deprez0e641232021-09-23 10:07:05 +0200948 rcu_read_lock();
949 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000950 memcpy(n_group_desc, o_group_desc,
951 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
Olivier Deprez0e641232021-09-23 10:07:05 +0200952 rcu_read_unlock();
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000953 n_group_desc[gdb_num] = gdb_bh;
David Brazdil0f672f62019-12-10 10:32:29 +0000954
955 BUFFER_TRACE(gdb_bh, "get_write_access");
956 err = ext4_journal_get_write_access(handle, gdb_bh);
957 if (err) {
958 kvfree(n_group_desc);
959 brelse(gdb_bh);
960 return err;
961 }
962
Olivier Deprez0e641232021-09-23 10:07:05 +0200963 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000964 EXT4_SB(sb)->s_gdb_count++;
Olivier Deprez0e641232021-09-23 10:07:05 +0200965 ext4_kvfree_array_rcu(o_group_desc);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000966 return err;
967}
968
969/*
970 * Called when we are adding a new group which has a backup copy of each of
971 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
972 * We need to add these reserved backup GDT blocks to the resize inode, so
973 * that they are kept for future resizing and not allocated to files.
974 *
975 * Each reserved backup GDT block will go into a different indirect block.
976 * The indirect blocks are actually the primary reserved GDT blocks,
977 * so we know in advance what their block numbers are. We only get the
978 * double-indirect block to verify it is pointing to the primary reserved
979 * GDT blocks so we don't overwrite a data block by accident. The reserved
980 * backup GDT blocks are stored in their reserved primary GDT block.
981 */
982static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
983 ext4_group_t group)
984{
985 struct super_block *sb = inode->i_sb;
986 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
987 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
988 struct buffer_head **primary;
989 struct buffer_head *dind;
990 struct ext4_iloc iloc;
991 ext4_fsblk_t blk;
992 __le32 *data, *end;
993 int gdbackups = 0;
994 int res, i;
995 int err;
996
997 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
998 if (!primary)
999 return -ENOMEM;
1000
1001 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
David Brazdil0f672f62019-12-10 10:32:29 +00001002 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1003 if (IS_ERR(dind)) {
1004 err = PTR_ERR(dind);
1005 dind = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001006 goto exit_free;
1007 }
1008
1009 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1010 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1011 EXT4_ADDR_PER_BLOCK(sb));
1012 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1013
1014 /* Get each reserved primary GDT block and verify it holds backups */
1015 for (res = 0; res < reserved_gdb; res++, blk++) {
1016 if (le32_to_cpu(*data) != blk) {
1017 ext4_warning(sb, "reserved block %llu"
1018 " not at offset %ld",
1019 blk,
1020 (long)(data - (__le32 *)dind->b_data));
1021 err = -EINVAL;
1022 goto exit_bh;
1023 }
David Brazdil0f672f62019-12-10 10:32:29 +00001024 primary[res] = ext4_sb_bread(sb, blk, 0);
1025 if (IS_ERR(primary[res])) {
1026 err = PTR_ERR(primary[res]);
1027 primary[res] = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001028 goto exit_bh;
1029 }
1030 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1031 if (gdbackups < 0) {
1032 brelse(primary[res]);
1033 err = gdbackups;
1034 goto exit_bh;
1035 }
1036 if (++data >= end)
1037 data = (__le32 *)dind->b_data;
1038 }
1039
1040 for (i = 0; i < reserved_gdb; i++) {
1041 BUFFER_TRACE(primary[i], "get_write_access");
1042 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1043 goto exit_bh;
1044 }
1045
1046 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1047 goto exit_bh;
1048
1049 /*
1050 * Finally we can add each of the reserved backup GDT blocks from
1051 * the new group to its reserved primary GDT block.
1052 */
1053 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1054 for (i = 0; i < reserved_gdb; i++) {
1055 int err2;
1056 data = (__le32 *)primary[i]->b_data;
1057 /* printk("reserving backup %lu[%u] = %lu\n",
1058 primary[i]->b_blocknr, gdbackups,
1059 blk + primary[i]->b_blocknr); */
1060 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1061 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1062 if (!err)
1063 err = err2;
1064 }
1065
1066 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1067 ext4_mark_iloc_dirty(handle, inode, &iloc);
1068
1069exit_bh:
1070 while (--res >= 0)
1071 brelse(primary[res]);
1072 brelse(dind);
1073
1074exit_free:
1075 kfree(primary);
1076
1077 return err;
1078}
1079
1080/*
1081 * Update the backup copies of the ext4 metadata. These don't need to be part
1082 * of the main resize transaction, because e2fsck will re-write them if there
1083 * is a problem (basically only OOM will cause a problem). However, we
1084 * _should_ update the backups if possible, in case the primary gets trashed
1085 * for some reason and we need to run e2fsck from a backup superblock. The
1086 * important part is that the new block and inode counts are in the backup
1087 * superblocks, and the location of the new group metadata in the GDT backups.
1088 *
1089 * We do not need take the s_resize_lock for this, because these
1090 * blocks are not otherwise touched by the filesystem code when it is
1091 * mounted. We don't need to worry about last changing from
1092 * sbi->s_groups_count, because the worst that can happen is that we
1093 * do not copy the full number of backups at this time. The resize
1094 * which changed s_groups_count will backup again.
1095 */
1096static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1097 int size, int meta_bg)
1098{
1099 struct ext4_sb_info *sbi = EXT4_SB(sb);
1100 ext4_group_t last;
1101 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1102 unsigned three = 1;
1103 unsigned five = 5;
1104 unsigned seven = 7;
1105 ext4_group_t group = 0;
1106 int rest = sb->s_blocksize - size;
1107 handle_t *handle;
1108 int err = 0, err2;
1109
1110 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1111 if (IS_ERR(handle)) {
1112 group = 1;
1113 err = PTR_ERR(handle);
1114 goto exit_err;
1115 }
1116
1117 if (meta_bg == 0) {
1118 group = ext4_list_backups(sb, &three, &five, &seven);
1119 last = sbi->s_groups_count;
1120 } else {
1121 group = ext4_get_group_number(sb, blk_off) + 1;
1122 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1123 }
1124
1125 while (group < sbi->s_groups_count) {
1126 struct buffer_head *bh;
1127 ext4_fsblk_t backup_block;
1128
1129 /* Out of journal space, and can't get more - abort - so sad */
Olivier Deprez157378f2022-04-04 15:47:50 +02001130 err = ext4_resize_ensure_credits_batch(handle, 1);
1131 if (err < 0)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001132 break;
1133
1134 if (meta_bg == 0)
1135 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1136 else
1137 backup_block = (ext4_group_first_block_no(sb, group) +
1138 ext4_bg_has_super(sb, group));
1139
1140 bh = sb_getblk(sb, backup_block);
1141 if (unlikely(!bh)) {
1142 err = -ENOMEM;
1143 break;
1144 }
1145 ext4_debug("update metadata backup %llu(+%llu)\n",
1146 backup_block, backup_block -
1147 ext4_group_first_block_no(sb, group));
1148 BUFFER_TRACE(bh, "get_write_access");
1149 if ((err = ext4_journal_get_write_access(handle, bh))) {
1150 brelse(bh);
1151 break;
1152 }
1153 lock_buffer(bh);
1154 memcpy(bh->b_data, data, size);
1155 if (rest)
1156 memset(bh->b_data + size, 0, rest);
1157 set_buffer_uptodate(bh);
1158 unlock_buffer(bh);
1159 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1160 if (unlikely(err))
1161 ext4_std_error(sb, err);
1162 brelse(bh);
1163
1164 if (meta_bg == 0)
1165 group = ext4_list_backups(sb, &three, &five, &seven);
1166 else if (group == last)
1167 break;
1168 else
1169 group = last;
1170 }
1171 if ((err2 = ext4_journal_stop(handle)) && !err)
1172 err = err2;
1173
1174 /*
1175 * Ugh! Need to have e2fsck write the backup copies. It is too
1176 * late to revert the resize, we shouldn't fail just because of
1177 * the backup copies (they are only needed in case of corruption).
1178 *
1179 * However, if we got here we have a journal problem too, so we
1180 * can't really start a transaction to mark the superblock.
1181 * Chicken out and just set the flag on the hope it will be written
1182 * to disk, and if not - we will simply wait until next fsck.
1183 */
1184exit_err:
1185 if (err) {
1186 ext4_warning(sb, "can't update backup for group %u (err %d), "
1187 "forcing fsck on next reboot", group, err);
1188 sbi->s_mount_state &= ~EXT4_VALID_FS;
1189 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1190 mark_buffer_dirty(sbi->s_sbh);
1191 }
1192}
1193
1194/*
1195 * ext4_add_new_descs() adds @count group descriptor of groups
1196 * starting at @group
1197 *
1198 * @handle: journal handle
1199 * @sb: super block
1200 * @group: the group no. of the first group desc to be added
1201 * @resize_inode: the resize inode
1202 * @count: number of group descriptors to be added
1203 */
1204static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1205 ext4_group_t group, struct inode *resize_inode,
1206 ext4_group_t count)
1207{
1208 struct ext4_sb_info *sbi = EXT4_SB(sb);
1209 struct ext4_super_block *es = sbi->s_es;
1210 struct buffer_head *gdb_bh;
1211 int i, gdb_off, gdb_num, err = 0;
1212 int meta_bg;
1213
1214 meta_bg = ext4_has_feature_meta_bg(sb);
1215 for (i = 0; i < count; i++, group++) {
1216 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1217 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1218
1219 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1220 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1221
1222 /*
1223 * We will only either add reserved group blocks to a backup group
1224 * or remove reserved blocks for the first group in a new group block.
1225 * Doing both would be mean more complex code, and sane people don't
1226 * use non-sparse filesystems anymore. This is already checked above.
1227 */
1228 if (gdb_off) {
Olivier Deprez0e641232021-09-23 10:07:05 +02001229 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1230 gdb_num);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001231 BUFFER_TRACE(gdb_bh, "get_write_access");
1232 err = ext4_journal_get_write_access(handle, gdb_bh);
1233
1234 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1235 err = reserve_backup_gdb(handle, resize_inode, group);
1236 } else if (meta_bg != 0) {
1237 err = add_new_gdb_meta_bg(sb, handle, group);
1238 } else {
1239 err = add_new_gdb(handle, resize_inode, group);
1240 }
1241 if (err)
1242 break;
1243 }
1244 return err;
1245}
1246
1247static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1248{
1249 struct buffer_head *bh = sb_getblk(sb, block);
1250 if (unlikely(!bh))
1251 return NULL;
1252 if (!bh_uptodate_or_lock(bh)) {
Olivier Deprez157378f2022-04-04 15:47:50 +02001253 if (ext4_read_bh(bh, 0, NULL) < 0) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001254 brelse(bh);
1255 return NULL;
1256 }
1257 }
1258
1259 return bh;
1260}
1261
1262static int ext4_set_bitmap_checksums(struct super_block *sb,
1263 ext4_group_t group,
1264 struct ext4_group_desc *gdp,
1265 struct ext4_new_group_data *group_data)
1266{
1267 struct buffer_head *bh;
1268
1269 if (!ext4_has_metadata_csum(sb))
1270 return 0;
1271
1272 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1273 if (!bh)
1274 return -EIO;
1275 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1276 EXT4_INODES_PER_GROUP(sb) / 8);
1277 brelse(bh);
1278
1279 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1280 if (!bh)
1281 return -EIO;
1282 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1283 brelse(bh);
1284
1285 return 0;
1286}
1287
1288/*
1289 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1290 */
1291static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1292 struct ext4_new_flex_group_data *flex_gd)
1293{
1294 struct ext4_new_group_data *group_data = flex_gd->groups;
1295 struct ext4_group_desc *gdp;
1296 struct ext4_sb_info *sbi = EXT4_SB(sb);
1297 struct buffer_head *gdb_bh;
1298 ext4_group_t group;
1299 __u16 *bg_flags = flex_gd->bg_flags;
1300 int i, gdb_off, gdb_num, err = 0;
1301
1302
1303 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1304 group = group_data->group;
1305
1306 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1307 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1308
1309 /*
1310 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1311 */
Olivier Deprez0e641232021-09-23 10:07:05 +02001312 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001313 /* Update group descriptor block for new group */
1314 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1315 gdb_off * EXT4_DESC_SIZE(sb));
1316
1317 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1318 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1319 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1320 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1321 if (err) {
1322 ext4_std_error(sb, err);
1323 break;
1324 }
1325
1326 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1327 ext4_free_group_clusters_set(sb, gdp,
1328 group_data->free_clusters_count);
1329 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1330 if (ext4_has_group_desc_csum(sb))
1331 ext4_itable_unused_set(sb, gdp,
1332 EXT4_INODES_PER_GROUP(sb));
1333 gdp->bg_flags = cpu_to_le16(*bg_flags);
1334 ext4_group_desc_csum_set(sb, group, gdp);
1335
1336 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1337 if (unlikely(err)) {
1338 ext4_std_error(sb, err);
1339 break;
1340 }
1341
1342 /*
1343 * We can allocate memory for mb_alloc based on the new group
1344 * descriptor
1345 */
1346 err = ext4_mb_add_groupinfo(sb, group, gdp);
1347 if (err)
1348 break;
1349 }
1350 return err;
1351}
1352
1353/*
1354 * ext4_update_super() updates the super block so that the newly added
1355 * groups can be seen by the filesystem.
1356 *
1357 * @sb: super block
1358 * @flex_gd: new added groups
1359 */
1360static void ext4_update_super(struct super_block *sb,
1361 struct ext4_new_flex_group_data *flex_gd)
1362{
1363 ext4_fsblk_t blocks_count = 0;
1364 ext4_fsblk_t free_blocks = 0;
1365 ext4_fsblk_t reserved_blocks = 0;
1366 struct ext4_new_group_data *group_data = flex_gd->groups;
1367 struct ext4_sb_info *sbi = EXT4_SB(sb);
1368 struct ext4_super_block *es = sbi->s_es;
1369 int i;
1370
1371 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1372 /*
1373 * Make the new blocks and inodes valid next. We do this before
1374 * increasing the group count so that once the group is enabled,
1375 * all of its blocks and inodes are already valid.
1376 *
1377 * We always allocate group-by-group, then block-by-block or
1378 * inode-by-inode within a group, so enabling these
1379 * blocks/inodes before the group is live won't actually let us
1380 * allocate the new space yet.
1381 */
1382 for (i = 0; i < flex_gd->count; i++) {
1383 blocks_count += group_data[i].blocks_count;
1384 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1385 }
1386
1387 reserved_blocks = ext4_r_blocks_count(es) * 100;
1388 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1389 reserved_blocks *= blocks_count;
1390 do_div(reserved_blocks, 100);
1391
1392 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1393 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1394 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1395 flex_gd->count);
1396 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1397 flex_gd->count);
1398
1399 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1400 /*
1401 * We need to protect s_groups_count against other CPUs seeing
1402 * inconsistent state in the superblock.
1403 *
1404 * The precise rules we use are:
1405 *
1406 * * Writers must perform a smp_wmb() after updating all
1407 * dependent data and before modifying the groups count
1408 *
1409 * * Readers must perform an smp_rmb() after reading the groups
1410 * count and before reading any dependent data.
1411 *
1412 * NB. These rules can be relaxed when checking the group count
1413 * while freeing data, as we can only allocate from a block
1414 * group after serialising against the group count, and we can
1415 * only then free after serialising in turn against that
1416 * allocation.
1417 */
1418 smp_wmb();
1419
1420 /* Update the global fs size fields */
1421 sbi->s_groups_count += flex_gd->count;
1422 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1423 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1424
1425 /* Update the reserved block counts only once the new group is
1426 * active. */
1427 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1428 reserved_blocks);
1429
1430 /* Update the free space counts */
1431 percpu_counter_add(&sbi->s_freeclusters_counter,
1432 EXT4_NUM_B2C(sbi, free_blocks));
1433 percpu_counter_add(&sbi->s_freeinodes_counter,
1434 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1435
1436 ext4_debug("free blocks count %llu",
1437 percpu_counter_read(&sbi->s_freeclusters_counter));
1438 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1439 ext4_group_t flex_group;
Olivier Deprez0e641232021-09-23 10:07:05 +02001440 struct flex_groups *fg;
1441
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001442 flex_group = ext4_flex_group(sbi, group_data[0].group);
Olivier Deprez0e641232021-09-23 10:07:05 +02001443 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001444 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
Olivier Deprez0e641232021-09-23 10:07:05 +02001445 &fg->free_clusters);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001446 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
Olivier Deprez0e641232021-09-23 10:07:05 +02001447 &fg->free_inodes);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001448 }
1449
1450 /*
1451 * Update the fs overhead information
1452 */
1453 ext4_calculate_overhead(sb);
1454
1455 if (test_opt(sb, DEBUG))
1456 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1457 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1458 blocks_count, free_blocks, reserved_blocks);
1459}
1460
1461/* Add a flex group to an fs. Ensure we handle all possible error conditions
1462 * _before_ we start modifying the filesystem, because we cannot abort the
1463 * transaction and not have it write the data to disk.
1464 */
1465static int ext4_flex_group_add(struct super_block *sb,
1466 struct inode *resize_inode,
1467 struct ext4_new_flex_group_data *flex_gd)
1468{
1469 struct ext4_sb_info *sbi = EXT4_SB(sb);
1470 struct ext4_super_block *es = sbi->s_es;
1471 ext4_fsblk_t o_blocks_count;
1472 ext4_grpblk_t last;
1473 ext4_group_t group;
1474 handle_t *handle;
1475 unsigned reserved_gdb;
1476 int err = 0, err2 = 0, credit;
1477
1478 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1479
1480 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1481 o_blocks_count = ext4_blocks_count(es);
1482 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1483 BUG_ON(last);
1484
1485 err = setup_new_flex_group_blocks(sb, flex_gd);
1486 if (err)
1487 goto exit;
1488 /*
1489 * We will always be modifying at least the superblock and GDT
1490 * blocks. If we are adding a group past the last current GDT block,
1491 * we will also modify the inode and the dindirect block. If we
1492 * are adding a group with superblock/GDT backups we will also
1493 * modify each of the reserved GDT dindirect blocks.
1494 */
1495 credit = 3; /* sb, resize inode, resize inode dindirect */
1496 /* GDT blocks */
1497 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1498 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1499 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1500 if (IS_ERR(handle)) {
1501 err = PTR_ERR(handle);
1502 goto exit;
1503 }
1504
1505 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1506 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1507 if (err)
1508 goto exit_journal;
1509
1510 group = flex_gd->groups[0].group;
1511 BUG_ON(group != sbi->s_groups_count);
1512 err = ext4_add_new_descs(handle, sb, group,
1513 resize_inode, flex_gd->count);
1514 if (err)
1515 goto exit_journal;
1516
1517 err = ext4_setup_new_descs(handle, sb, flex_gd);
1518 if (err)
1519 goto exit_journal;
1520
1521 ext4_update_super(sb, flex_gd);
1522
1523 err = ext4_handle_dirty_super(handle, sb);
1524
1525exit_journal:
1526 err2 = ext4_journal_stop(handle);
1527 if (!err)
1528 err = err2;
1529
1530 if (!err) {
1531 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1532 int gdb_num_end = ((group + flex_gd->count - 1) /
1533 EXT4_DESC_PER_BLOCK(sb));
1534 int meta_bg = ext4_has_feature_meta_bg(sb);
1535 sector_t old_gdb = 0;
1536
1537 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1538 sizeof(struct ext4_super_block), 0);
1539 for (; gdb_num <= gdb_num_end; gdb_num++) {
1540 struct buffer_head *gdb_bh;
1541
Olivier Deprez0e641232021-09-23 10:07:05 +02001542 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1543 gdb_num);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001544 if (old_gdb == gdb_bh->b_blocknr)
1545 continue;
1546 update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1547 gdb_bh->b_size, meta_bg);
1548 old_gdb = gdb_bh->b_blocknr;
1549 }
1550 }
1551exit:
1552 return err;
1553}
1554
1555static int ext4_setup_next_flex_gd(struct super_block *sb,
1556 struct ext4_new_flex_group_data *flex_gd,
1557 ext4_fsblk_t n_blocks_count,
1558 unsigned long flexbg_size)
1559{
1560 struct ext4_sb_info *sbi = EXT4_SB(sb);
1561 struct ext4_super_block *es = sbi->s_es;
1562 struct ext4_new_group_data *group_data = flex_gd->groups;
1563 ext4_fsblk_t o_blocks_count;
1564 ext4_group_t n_group;
1565 ext4_group_t group;
1566 ext4_group_t last_group;
1567 ext4_grpblk_t last;
1568 ext4_grpblk_t clusters_per_group;
1569 unsigned long i;
1570
1571 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1572
1573 o_blocks_count = ext4_blocks_count(es);
1574
1575 if (o_blocks_count == n_blocks_count)
1576 return 0;
1577
1578 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1579 BUG_ON(last);
1580 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1581
1582 last_group = group | (flexbg_size - 1);
1583 if (last_group > n_group)
1584 last_group = n_group;
1585
1586 flex_gd->count = last_group - group + 1;
1587
1588 for (i = 0; i < flex_gd->count; i++) {
1589 int overhead;
1590
1591 group_data[i].group = group + i;
1592 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1593 overhead = ext4_group_overhead_blocks(sb, group + i);
1594 group_data[i].mdata_blocks = overhead;
1595 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1596 if (ext4_has_group_desc_csum(sb)) {
1597 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1598 EXT4_BG_INODE_UNINIT;
1599 if (!test_opt(sb, INIT_INODE_TABLE))
1600 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1601 } else
1602 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1603 }
1604
1605 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1606 /* We need to initialize block bitmap of last group. */
1607 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1608
1609 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1610 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1611 group_data[i - 1].free_clusters_count -= clusters_per_group -
1612 last - 1;
1613 }
1614
1615 return 1;
1616}
1617
1618/* Add group descriptor data to an existing or new group descriptor block.
1619 * Ensure we handle all possible error conditions _before_ we start modifying
1620 * the filesystem, because we cannot abort the transaction and not have it
1621 * write the data to disk.
1622 *
1623 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1624 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1625 *
1626 * We only need to hold the superblock lock while we are actually adding
1627 * in the new group's counts to the superblock. Prior to that we have
1628 * not really "added" the group at all. We re-check that we are still
1629 * adding in the last group in case things have changed since verifying.
1630 */
1631int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1632{
1633 struct ext4_new_flex_group_data flex_gd;
1634 struct ext4_sb_info *sbi = EXT4_SB(sb);
1635 struct ext4_super_block *es = sbi->s_es;
1636 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1637 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1638 struct inode *inode = NULL;
1639 int gdb_off;
1640 int err;
1641 __u16 bg_flags = 0;
1642
1643 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1644
1645 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1646 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1647 return -EPERM;
1648 }
1649
1650 if (ext4_blocks_count(es) + input->blocks_count <
1651 ext4_blocks_count(es)) {
1652 ext4_warning(sb, "blocks_count overflow");
1653 return -EINVAL;
1654 }
1655
1656 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1657 le32_to_cpu(es->s_inodes_count)) {
1658 ext4_warning(sb, "inodes_count overflow");
1659 return -EINVAL;
1660 }
1661
1662 if (reserved_gdb || gdb_off == 0) {
David Brazdil0f672f62019-12-10 10:32:29 +00001663 if (!ext4_has_feature_resize_inode(sb) ||
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001664 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1665 ext4_warning(sb,
1666 "No reserved GDT blocks, can't resize");
1667 return -EPERM;
1668 }
David Brazdil0f672f62019-12-10 10:32:29 +00001669 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001670 if (IS_ERR(inode)) {
1671 ext4_warning(sb, "Error opening resize inode");
1672 return PTR_ERR(inode);
1673 }
1674 }
1675
1676
1677 err = verify_group_input(sb, input);
1678 if (err)
1679 goto out;
1680
1681 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1682 if (err)
1683 goto out;
1684
1685 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1686 if (err)
1687 goto out;
1688
1689 flex_gd.count = 1;
1690 flex_gd.groups = input;
1691 flex_gd.bg_flags = &bg_flags;
1692 err = ext4_flex_group_add(sb, inode, &flex_gd);
1693out:
1694 iput(inode);
1695 return err;
1696} /* ext4_group_add */
1697
1698/*
1699 * extend a group without checking assuming that checking has been done.
1700 */
1701static int ext4_group_extend_no_check(struct super_block *sb,
1702 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1703{
1704 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1705 handle_t *handle;
1706 int err = 0, err2;
1707
1708 /* We will update the superblock, one block bitmap, and
1709 * one group descriptor via ext4_group_add_blocks().
1710 */
1711 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1712 if (IS_ERR(handle)) {
1713 err = PTR_ERR(handle);
1714 ext4_warning(sb, "error %d on journal start", err);
1715 return err;
1716 }
1717
1718 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1719 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1720 if (err) {
1721 ext4_warning(sb, "error %d on journal write access", err);
1722 goto errout;
1723 }
1724
1725 ext4_blocks_count_set(es, o_blocks_count + add);
1726 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1727 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1728 o_blocks_count + add);
1729 /* We add the blocks to the bitmap and set the group need init bit */
1730 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1731 if (err)
1732 goto errout;
1733 ext4_handle_dirty_super(handle, sb);
1734 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1735 o_blocks_count + add);
1736errout:
1737 err2 = ext4_journal_stop(handle);
1738 if (err2 && !err)
1739 err = err2;
1740
1741 if (!err) {
1742 if (test_opt(sb, DEBUG))
1743 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1744 "blocks\n", ext4_blocks_count(es));
1745 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1746 (char *)es, sizeof(struct ext4_super_block), 0);
1747 }
1748 return err;
1749}
1750
1751/*
1752 * Extend the filesystem to the new number of blocks specified. This entry
1753 * point is only used to extend the current filesystem to the end of the last
1754 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1755 * for emergencies (because it has no dependencies on reserved blocks).
1756 *
1757 * If we _really_ wanted, we could use default values to call ext4_group_add()
1758 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1759 * GDT blocks are reserved to grow to the desired size.
1760 */
1761int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1762 ext4_fsblk_t n_blocks_count)
1763{
1764 ext4_fsblk_t o_blocks_count;
1765 ext4_grpblk_t last;
1766 ext4_grpblk_t add;
1767 struct buffer_head *bh;
1768 int err;
1769 ext4_group_t group;
1770
1771 o_blocks_count = ext4_blocks_count(es);
1772
1773 if (test_opt(sb, DEBUG))
1774 ext4_msg(sb, KERN_DEBUG,
1775 "extending last group from %llu to %llu blocks",
1776 o_blocks_count, n_blocks_count);
1777
1778 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1779 return 0;
1780
1781 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1782 ext4_msg(sb, KERN_ERR,
1783 "filesystem too large to resize to %llu blocks safely",
1784 n_blocks_count);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001785 return -EINVAL;
1786 }
1787
1788 if (n_blocks_count < o_blocks_count) {
1789 ext4_warning(sb, "can't shrink FS - resize aborted");
1790 return -EINVAL;
1791 }
1792
1793 /* Handle the remaining blocks in the last group only. */
1794 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1795
1796 if (last == 0) {
1797 ext4_warning(sb, "need to use ext2online to resize further");
1798 return -EPERM;
1799 }
1800
1801 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1802
1803 if (o_blocks_count + add < o_blocks_count) {
1804 ext4_warning(sb, "blocks_count overflow");
1805 return -EINVAL;
1806 }
1807
1808 if (o_blocks_count + add > n_blocks_count)
1809 add = n_blocks_count - o_blocks_count;
1810
1811 if (o_blocks_count + add < n_blocks_count)
1812 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1813 o_blocks_count + add, add);
1814
1815 /* See if the device is actually as big as what was requested */
Olivier Deprez157378f2022-04-04 15:47:50 +02001816 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1817 if (IS_ERR(bh)) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001818 ext4_warning(sb, "can't read last block, resize aborted");
1819 return -ENOSPC;
1820 }
1821 brelse(bh);
1822
1823 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1824 return err;
1825} /* ext4_group_extend */
1826
1827
1828static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1829{
1830 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1831}
1832
1833/*
1834 * Release the resize inode and drop the resize_inode feature if there
1835 * are no more reserved gdt blocks, and then convert the file system
1836 * to enable meta_bg
1837 */
1838static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1839{
1840 handle_t *handle;
1841 struct ext4_sb_info *sbi = EXT4_SB(sb);
1842 struct ext4_super_block *es = sbi->s_es;
1843 struct ext4_inode_info *ei = EXT4_I(inode);
1844 ext4_fsblk_t nr;
1845 int i, ret, err = 0;
1846 int credits = 1;
1847
1848 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1849 if (inode) {
1850 if (es->s_reserved_gdt_blocks) {
1851 ext4_error(sb, "Unexpected non-zero "
1852 "s_reserved_gdt_blocks");
1853 return -EPERM;
1854 }
1855
1856 /* Do a quick sanity check of the resize inode */
1857 if (inode->i_blocks != 1 << (inode->i_blkbits -
1858 (9 - sbi->s_cluster_bits)))
1859 goto invalid_resize_inode;
1860 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1861 if (i == EXT4_DIND_BLOCK) {
1862 if (ei->i_data[i])
1863 continue;
1864 else
1865 goto invalid_resize_inode;
1866 }
1867 if (ei->i_data[i])
1868 goto invalid_resize_inode;
1869 }
1870 credits += 3; /* block bitmap, bg descriptor, resize inode */
1871 }
1872
1873 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1874 if (IS_ERR(handle))
1875 return PTR_ERR(handle);
1876
1877 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1878 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1879 if (err)
1880 goto errout;
1881
1882 ext4_clear_feature_resize_inode(sb);
1883 ext4_set_feature_meta_bg(sb);
1884 sbi->s_es->s_first_meta_bg =
1885 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1886
1887 err = ext4_handle_dirty_super(handle, sb);
1888 if (err) {
1889 ext4_std_error(sb, err);
1890 goto errout;
1891 }
1892
1893 if (inode) {
1894 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1895 ext4_free_blocks(handle, inode, NULL, nr, 1,
1896 EXT4_FREE_BLOCKS_METADATA |
1897 EXT4_FREE_BLOCKS_FORGET);
1898 ei->i_data[EXT4_DIND_BLOCK] = 0;
1899 inode->i_blocks = 0;
1900
1901 err = ext4_mark_inode_dirty(handle, inode);
1902 if (err)
1903 ext4_std_error(sb, err);
1904 }
1905
1906errout:
1907 ret = ext4_journal_stop(handle);
1908 if (!err)
1909 err = ret;
1910 return ret;
1911
1912invalid_resize_inode:
1913 ext4_error(sb, "corrupted/inconsistent resize inode");
1914 return -EINVAL;
1915}
1916
1917/*
1918 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1919 *
1920 * @sb: super block of the fs to be resized
1921 * @n_blocks_count: the number of blocks resides in the resized fs
1922 */
1923int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1924{
1925 struct ext4_new_flex_group_data *flex_gd = NULL;
1926 struct ext4_sb_info *sbi = EXT4_SB(sb);
1927 struct ext4_super_block *es = sbi->s_es;
1928 struct buffer_head *bh;
1929 struct inode *resize_inode = NULL;
1930 ext4_grpblk_t add, offset;
1931 unsigned long n_desc_blocks;
1932 unsigned long o_desc_blocks;
1933 ext4_group_t o_group;
1934 ext4_group_t n_group;
1935 ext4_fsblk_t o_blocks_count;
1936 ext4_fsblk_t n_blocks_count_retry = 0;
1937 unsigned long last_update_time = 0;
1938 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1939 int meta_bg;
1940
1941 /* See if the device is actually as big as what was requested */
Olivier Deprez157378f2022-04-04 15:47:50 +02001942 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1943 if (IS_ERR(bh)) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001944 ext4_warning(sb, "can't read last block, resize aborted");
1945 return -ENOSPC;
1946 }
1947 brelse(bh);
1948
1949retry:
1950 o_blocks_count = ext4_blocks_count(es);
1951
1952 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1953 "to %llu blocks", o_blocks_count, n_blocks_count);
1954
1955 if (n_blocks_count < o_blocks_count) {
1956 /* On-line shrinking not supported */
1957 ext4_warning(sb, "can't shrink FS - resize aborted");
1958 return -EINVAL;
1959 }
1960
1961 if (n_blocks_count == o_blocks_count)
1962 /* Nothing need to do */
1963 return 0;
1964
1965 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1966 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1967 ext4_warning(sb, "resize would cause inodes_count overflow");
1968 return -EINVAL;
1969 }
1970 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1971
1972 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1973 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1974
1975 meta_bg = ext4_has_feature_meta_bg(sb);
1976
1977 if (ext4_has_feature_resize_inode(sb)) {
1978 if (meta_bg) {
1979 ext4_error(sb, "resize_inode and meta_bg enabled "
1980 "simultaneously");
1981 return -EINVAL;
1982 }
1983 if (n_desc_blocks > o_desc_blocks +
1984 le16_to_cpu(es->s_reserved_gdt_blocks)) {
1985 n_blocks_count_retry = n_blocks_count;
1986 n_desc_blocks = o_desc_blocks +
1987 le16_to_cpu(es->s_reserved_gdt_blocks);
1988 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1989 n_blocks_count = (ext4_fsblk_t)n_group *
David Brazdil0f672f62019-12-10 10:32:29 +00001990 EXT4_BLOCKS_PER_GROUP(sb) +
1991 le32_to_cpu(es->s_first_data_block);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001992 n_group--; /* set to last group number */
1993 }
1994
1995 if (!resize_inode)
David Brazdil0f672f62019-12-10 10:32:29 +00001996 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1997 EXT4_IGET_SPECIAL);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001998 if (IS_ERR(resize_inode)) {
1999 ext4_warning(sb, "Error opening resize inode");
2000 return PTR_ERR(resize_inode);
2001 }
2002 }
2003
2004 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2005 err = ext4_convert_meta_bg(sb, resize_inode);
2006 if (err)
2007 goto out;
2008 if (resize_inode) {
2009 iput(resize_inode);
2010 resize_inode = NULL;
2011 }
2012 if (n_blocks_count_retry) {
2013 n_blocks_count = n_blocks_count_retry;
2014 n_blocks_count_retry = 0;
2015 goto retry;
2016 }
2017 }
2018
2019 /*
2020 * Make sure the last group has enough space so that it's
2021 * guaranteed to have enough space for all metadata blocks
2022 * that it might need to hold. (We might not need to store
2023 * the inode table blocks in the last block group, but there
2024 * will be cases where this might be needed.)
2025 */
2026 if ((ext4_group_first_block_no(sb, n_group) +
2027 ext4_group_overhead_blocks(sb, n_group) + 2 +
2028 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2029 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2030 n_group--;
2031 n_blocks_count_retry = 0;
2032 if (resize_inode) {
2033 iput(resize_inode);
2034 resize_inode = NULL;
2035 }
2036 goto retry;
2037 }
2038
2039 /* extend the last group */
2040 if (n_group == o_group)
2041 add = n_blocks_count - o_blocks_count;
2042 else
2043 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2044 if (add > 0) {
2045 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2046 if (err)
2047 goto out;
2048 }
2049
2050 if (ext4_blocks_count(es) == n_blocks_count)
2051 goto out;
2052
2053 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2054 if (err)
2055 goto out;
2056
2057 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2058 if (err)
2059 goto out;
2060
2061 flex_gd = alloc_flex_gd(flexbg_size);
2062 if (flex_gd == NULL) {
2063 err = -ENOMEM;
2064 goto out;
2065 }
2066
2067 /* Add flex groups. Note that a regular group is a
2068 * flex group with 1 group.
2069 */
2070 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2071 flexbg_size)) {
2072 if (jiffies - last_update_time > HZ * 10) {
2073 if (last_update_time)
2074 ext4_msg(sb, KERN_INFO,
2075 "resized to %llu blocks",
2076 ext4_blocks_count(es));
2077 last_update_time = jiffies;
2078 }
2079 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2080 break;
2081 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2082 if (unlikely(err))
2083 break;
2084 }
2085
2086 if (!err && n_blocks_count_retry) {
2087 n_blocks_count = n_blocks_count_retry;
2088 n_blocks_count_retry = 0;
2089 free_flex_gd(flex_gd);
2090 flex_gd = NULL;
2091 if (resize_inode) {
2092 iput(resize_inode);
2093 resize_inode = NULL;
2094 }
2095 goto retry;
2096 }
2097
2098out:
2099 if (flex_gd)
2100 free_flex_gd(flex_gd);
2101 if (resize_inode != NULL)
2102 iput(resize_inode);
David Brazdil0f672f62019-12-10 10:32:29 +00002103 if (err)
2104 ext4_warning(sb, "error (%d) occurred during "
2105 "file system resize", err);
2106 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2107 ext4_blocks_count(es));
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002108 return err;
2109}