1// 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#include <linux/errno.h>
14#include <linux/slab.h>
15#include <linux/jiffies.h>
16
17#include "ext4_jbd2.h"
18
19struct ext4_rcu_ptr {
20 struct rcu_head rcu;
21 void *ptr;
22};
23
24static void ext4_rcu_ptr_callback(struct rcu_head *head)
25{
26 struct ext4_rcu_ptr *ptr;
27
28 ptr = container_of(head, struct ext4_rcu_ptr, rcu);
29 kvfree(addr: ptr->ptr);
30 kfree(objp: ptr);
31}
32
33void ext4_kvfree_array_rcu(void *to_free)
34{
35 struct ext4_rcu_ptr *ptr = kzalloc(size: sizeof(*ptr), GFP_KERNEL);
36
37 if (ptr) {
38 ptr->ptr = to_free;
39 call_rcu(head: &ptr->rcu, func: ext4_rcu_ptr_callback);
40 return;
41 }
42 synchronize_rcu();
43 kvfree(addr: to_free);
44}
45
46int ext4_resize_begin(struct super_block *sb)
47{
48 struct ext4_sb_info *sbi = EXT4_SB(sb);
49 int ret = 0;
50
51 if (!capable(CAP_SYS_RESOURCE))
52 return -EPERM;
53
54 /*
55 * If the reserved GDT blocks is non-zero, the resize_inode feature
56 * should always be set.
57 */
58 if (sbi->s_es->s_reserved_gdt_blocks &&
59 !ext4_has_feature_resize_inode(sb)) {
60 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
61 return -EFSCORRUPTED;
62 }
63
64 /*
65 * If we are not using the primary superblock/GDT copy don't resize,
66 * because the user tools have no way of handling this. Probably a
67 * bad time to do it anyways.
68 */
69 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
70 le32_to_cpu(sbi->s_es->s_first_data_block)) {
71 ext4_warning(sb, "won't resize using backup superblock at %llu",
72 (unsigned long long)sbi->s_sbh->b_blocknr);
73 return -EPERM;
74 }
75
76 /*
77 * We are not allowed to do online-resizing on a filesystem mounted
78 * with error, because it can destroy the filesystem easily.
79 */
80 if (sbi->s_mount_state & EXT4_ERROR_FS) {
81 ext4_warning(sb, "There are errors in the filesystem, "
82 "so online resizing is not allowed");
83 return -EPERM;
84 }
85
86 if (ext4_has_feature_sparse_super2(sb)) {
87 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
88 return -EOPNOTSUPP;
89 }
90
91 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
92 addr: &sbi->s_ext4_flags))
93 ret = -EBUSY;
94
95 return ret;
96}
97
98int ext4_resize_end(struct super_block *sb, bool update_backups)
99{
100 clear_bit_unlock(EXT4_FLAGS_RESIZING, addr: &EXT4_SB(sb)->s_ext4_flags);
101 smp_mb__after_atomic();
102 if (update_backups)
103 return ext4_update_overhead(sb, force: true);
104 return 0;
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 free_blocks_count = input->blocks_count - 2 - overhead -
144 sbi->s_itb_per_group;
145 input->free_clusters_count = EXT4_B2C(sbi, free_blocks_count);
146
147 if (test_opt(sb, DEBUG))
148 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
149 "(%d free, %u reserved)\n",
150 ext4_bg_has_super(sb, input->group) ? "normal" :
151 "no-super", input->group, input->blocks_count,
152 free_blocks_count, input->reserved_blocks);
153
154 ext4_get_group_no_and_offset(sb, blocknr: start, NULL, offsetp: &offset);
155 if (offset != 0)
156 ext4_warning(sb, "Last group not full");
157 else if (input->reserved_blocks > input->blocks_count / 5)
158 ext4_warning(sb, "Reserved blocks too high (%u)",
159 input->reserved_blocks);
160 else if (free_blocks_count < 0)
161 ext4_warning(sb, "Bad blocks count %u",
162 input->blocks_count);
163 else if (IS_ERR(ptr: bh = ext4_sb_bread(sb, block: end - 1, op_flags: 0))) {
164 err = PTR_ERR(ptr: bh);
165 bh = NULL;
166 ext4_warning(sb, "Cannot read last block (%llu)",
167 end - 1);
168 } else if (outside(input->block_bitmap, start, end))
169 ext4_warning(sb, "Block bitmap not in group (block %llu)",
170 (unsigned long long)input->block_bitmap);
171 else if (outside(input->inode_bitmap, start, end))
172 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
173 (unsigned long long)input->inode_bitmap);
174 else if (outside(input->inode_table, start, end) ||
175 outside(itend - 1, start, end))
176 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
177 (unsigned long long)input->inode_table, itend - 1);
178 else if (input->inode_bitmap == input->block_bitmap)
179 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
180 (unsigned long long)input->block_bitmap);
181 else if (inside(input->block_bitmap, input->inode_table, itend))
182 ext4_warning(sb, "Block bitmap (%llu) in inode table "
183 "(%llu-%llu)",
184 (unsigned long long)input->block_bitmap,
185 (unsigned long long)input->inode_table, itend - 1);
186 else if (inside(input->inode_bitmap, input->inode_table, itend))
187 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
188 "(%llu-%llu)",
189 (unsigned long long)input->inode_bitmap,
190 (unsigned long long)input->inode_table, itend - 1);
191 else if (inside(input->block_bitmap, start, metaend))
192 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
193 (unsigned long long)input->block_bitmap,
194 start, metaend - 1);
195 else if (inside(input->inode_bitmap, start, metaend))
196 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
197 (unsigned long long)input->inode_bitmap,
198 start, metaend - 1);
199 else if (inside(input->inode_table, start, metaend) ||
200 inside(itend - 1, start, metaend))
201 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
202 "(%llu-%llu)",
203 (unsigned long long)input->inode_table,
204 itend - 1, start, metaend - 1);
205 else
206 err = 0;
207 brelse(bh);
208
209 return err;
210}
211
212/*
213 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
214 * group each time.
215 */
216struct ext4_new_flex_group_data {
217 struct ext4_new_group_data *groups; /* new_group_data for groups
218 in the flex group */
219 __u16 *bg_flags; /* block group flags of groups
220 in @groups */
221 ext4_group_t count; /* number of groups in @groups
222 */
223};
224
225/*
226 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
227 * @flexbg_size.
228 *
229 * Returns NULL on failure otherwise address of the allocated structure.
230 */
231static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
232{
233 struct ext4_new_flex_group_data *flex_gd;
234
235 flex_gd = kmalloc(size: sizeof(*flex_gd), GFP_NOFS);
236 if (flex_gd == NULL)
237 goto out3;
238
239 if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
240 goto out2;
241 flex_gd->count = flexbg_size;
242
243 flex_gd->groups = kmalloc_array(n: flexbg_size,
244 size: sizeof(struct ext4_new_group_data),
245 GFP_NOFS);
246 if (flex_gd->groups == NULL)
247 goto out2;
248
249 flex_gd->bg_flags = kmalloc_array(n: flexbg_size, size: sizeof(__u16),
250 GFP_NOFS);
251 if (flex_gd->bg_flags == NULL)
252 goto out1;
253
254 return flex_gd;
255
256out1:
257 kfree(objp: flex_gd->groups);
258out2:
259 kfree(objp: flex_gd);
260out3:
261 return NULL;
262}
263
264static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
265{
266 kfree(objp: flex_gd->bg_flags);
267 kfree(objp: flex_gd->groups);
268 kfree(objp: flex_gd);
269}
270
271/*
272 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
273 * and inode tables for a flex group.
274 *
275 * This function is used by 64bit-resize. Note that this function allocates
276 * group tables from the 1st group of groups contained by @flexgd, which may
277 * be a partial of a flex group.
278 *
279 * @sb: super block of fs to which the groups belongs
280 *
281 * Returns 0 on a successful allocation of the metadata blocks in the
282 * block group.
283 */
284static int ext4_alloc_group_tables(struct super_block *sb,
285 struct ext4_new_flex_group_data *flex_gd,
286 int flexbg_size)
287{
288 struct ext4_new_group_data *group_data = flex_gd->groups;
289 ext4_fsblk_t start_blk;
290 ext4_fsblk_t last_blk;
291 ext4_group_t src_group;
292 ext4_group_t bb_index = 0;
293 ext4_group_t ib_index = 0;
294 ext4_group_t it_index = 0;
295 ext4_group_t group;
296 ext4_group_t last_group;
297 unsigned overhead;
298 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
299 int i;
300
301 BUG_ON(flex_gd->count == 0 || group_data == NULL);
302
303 src_group = group_data[0].group;
304 last_group = src_group + flex_gd->count - 1;
305
306 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
307 (last_group & ~(flexbg_size - 1))));
308next_group:
309 group = group_data[0].group;
310 if (src_group >= group_data[0].group + flex_gd->count)
311 return -ENOSPC;
312 start_blk = ext4_group_first_block_no(sb, group_no: src_group);
313 last_blk = start_blk + group_data[src_group - group].blocks_count;
314
315 overhead = ext4_group_overhead_blocks(sb, group: src_group);
316
317 start_blk += overhead;
318
319 /* We collect contiguous blocks as much as possible. */
320 src_group++;
321 for (; src_group <= last_group; src_group++) {
322 overhead = ext4_group_overhead_blocks(sb, group: src_group);
323 if (overhead == 0)
324 last_blk += group_data[src_group - group].blocks_count;
325 else
326 break;
327 }
328
329 /* Allocate block bitmaps */
330 for (; bb_index < flex_gd->count; bb_index++) {
331 if (start_blk >= last_blk)
332 goto next_group;
333 group_data[bb_index].block_bitmap = start_blk++;
334 group = ext4_get_group_number(sb, block: start_blk - 1);
335 group -= group_data[0].group;
336 group_data[group].mdata_blocks++;
337 flex_gd->bg_flags[group] &= uninit_mask;
338 }
339
340 /* Allocate inode bitmaps */
341 for (; ib_index < flex_gd->count; ib_index++) {
342 if (start_blk >= last_blk)
343 goto next_group;
344 group_data[ib_index].inode_bitmap = start_blk++;
345 group = ext4_get_group_number(sb, block: start_blk - 1);
346 group -= group_data[0].group;
347 group_data[group].mdata_blocks++;
348 flex_gd->bg_flags[group] &= uninit_mask;
349 }
350
351 /* Allocate inode tables */
352 for (; it_index < flex_gd->count; it_index++) {
353 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
354 ext4_fsblk_t next_group_start;
355
356 if (start_blk + itb > last_blk)
357 goto next_group;
358 group_data[it_index].inode_table = start_blk;
359 group = ext4_get_group_number(sb, block: start_blk);
360 next_group_start = ext4_group_first_block_no(sb, group_no: group + 1);
361 group -= group_data[0].group;
362
363 if (start_blk + itb > next_group_start) {
364 flex_gd->bg_flags[group + 1] &= uninit_mask;
365 overhead = start_blk + itb - next_group_start;
366 group_data[group + 1].mdata_blocks += overhead;
367 itb -= overhead;
368 }
369
370 group_data[group].mdata_blocks += itb;
371 flex_gd->bg_flags[group] &= uninit_mask;
372 start_blk += EXT4_SB(sb)->s_itb_per_group;
373 }
374
375 /* Update free clusters count to exclude metadata blocks */
376 for (i = 0; i < flex_gd->count; i++) {
377 group_data[i].free_clusters_count -=
378 EXT4_NUM_B2C(EXT4_SB(sb),
379 group_data[i].mdata_blocks);
380 }
381
382 if (test_opt(sb, DEBUG)) {
383 int i;
384 group = group_data[0].group;
385
386 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
387 "%d groups, flexbg size is %d:\n", flex_gd->count,
388 flexbg_size);
389
390 for (i = 0; i < flex_gd->count; i++) {
391 ext4_debug(
392 "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
393 ext4_bg_has_super(sb, group + i) ? "normal" :
394 "no-super", group + i,
395 group_data[i].blocks_count,
396 group_data[i].free_clusters_count,
397 group_data[i].mdata_blocks);
398 }
399 }
400 return 0;
401}
402
403static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
404 ext4_fsblk_t blk)
405{
406 struct buffer_head *bh;
407 int err;
408
409 bh = sb_getblk(sb, block: blk);
410 if (unlikely(!bh))
411 return ERR_PTR(error: -ENOMEM);
412 BUFFER_TRACE(bh, "get_write_access");
413 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
414 if (err) {
415 brelse(bh);
416 bh = ERR_PTR(error: err);
417 } else {
418 memset(bh->b_data, 0, sb->s_blocksize);
419 set_buffer_uptodate(bh);
420 }
421
422 return bh;
423}
424
425static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
426{
427 return ext4_journal_ensure_credits_fn(handle, credits,
428 EXT4_MAX_TRANS_DATA, 0, 0);
429}
430
431/*
432 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
433 *
434 * Helper function for ext4_setup_new_group_blocks() which set .
435 *
436 * @sb: super block
437 * @handle: journal handle
438 * @flex_gd: flex group data
439 */
440static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
441 struct ext4_new_flex_group_data *flex_gd,
442 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
443{
444 struct ext4_sb_info *sbi = EXT4_SB(sb);
445 ext4_group_t count = last_cluster - first_cluster + 1;
446 ext4_group_t count2;
447
448 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
449 last_cluster);
450 for (; count > 0; count -= count2, first_cluster += count2) {
451 ext4_fsblk_t start;
452 struct buffer_head *bh;
453 ext4_group_t group;
454 int err;
455
456 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
457 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
458 group -= flex_gd->groups[0].group;
459
460 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
461 if (count2 > count)
462 count2 = count;
463
464 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
465 BUG_ON(flex_gd->count > 1);
466 continue;
467 }
468
469 err = ext4_resize_ensure_credits_batch(handle, credits: 1);
470 if (err < 0)
471 return err;
472
473 bh = sb_getblk(sb, block: flex_gd->groups[group].block_bitmap);
474 if (unlikely(!bh))
475 return -ENOMEM;
476
477 BUFFER_TRACE(bh, "get_write_access");
478 err = ext4_journal_get_write_access(handle, sb, bh,
479 EXT4_JTR_NONE);
480 if (err) {
481 brelse(bh);
482 return err;
483 }
484 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
485 first_cluster, first_cluster - start, count2);
486 mb_set_bits(bm: bh->b_data, cur: first_cluster - start, len: count2);
487
488 err = ext4_handle_dirty_metadata(handle, NULL, bh);
489 brelse(bh);
490 if (unlikely(err))
491 return err;
492 }
493
494 return 0;
495}
496
497/*
498 * Set up the block and inode bitmaps, and the inode table for the new groups.
499 * This doesn't need to be part of the main transaction, since we are only
500 * changing blocks outside the actual filesystem. We still do journaling to
501 * ensure the recovery is correct in case of a failure just after resize.
502 * If any part of this fails, we simply abort the resize.
503 *
504 * setup_new_flex_group_blocks handles a flex group as follow:
505 * 1. copy super block and GDT, and initialize group tables if necessary.
506 * In this step, we only set bits in blocks bitmaps for blocks taken by
507 * super block and GDT.
508 * 2. allocate group tables in block bitmaps, that is, set bits in block
509 * bitmap for blocks taken by group tables.
510 */
511static int setup_new_flex_group_blocks(struct super_block *sb,
512 struct ext4_new_flex_group_data *flex_gd)
513{
514 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
515 ext4_fsblk_t start;
516 ext4_fsblk_t block;
517 struct ext4_sb_info *sbi = EXT4_SB(sb);
518 struct ext4_super_block *es = sbi->s_es;
519 struct ext4_new_group_data *group_data = flex_gd->groups;
520 __u16 *bg_flags = flex_gd->bg_flags;
521 handle_t *handle;
522 ext4_group_t group, count;
523 struct buffer_head *bh = NULL;
524 int reserved_gdb, i, j, err = 0, err2;
525 int meta_bg;
526
527 BUG_ON(!flex_gd->count || !group_data ||
528 group_data[0].group != sbi->s_groups_count);
529
530 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
531 meta_bg = ext4_has_feature_meta_bg(sb);
532
533 /* This transaction may be extended/restarted along the way */
534 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
535 if (IS_ERR(ptr: handle))
536 return PTR_ERR(ptr: handle);
537
538 group = group_data[0].group;
539 for (i = 0; i < flex_gd->count; i++, group++) {
540 unsigned long gdblocks;
541 ext4_grpblk_t overhead;
542
543 gdblocks = ext4_bg_num_gdb(sb, group);
544 start = ext4_group_first_block_no(sb, group_no: group);
545
546 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
547 goto handle_itb;
548
549 if (meta_bg == 1)
550 goto handle_itb;
551
552 block = start + ext4_bg_has_super(sb, group);
553 /* Copy all of the GDT blocks into the backup in this group */
554 for (j = 0; j < gdblocks; j++, block++) {
555 struct buffer_head *gdb;
556
557 ext4_debug("update backup group %#04llx\n", block);
558 err = ext4_resize_ensure_credits_batch(handle, credits: 1);
559 if (err < 0)
560 goto out;
561
562 gdb = sb_getblk(sb, block);
563 if (unlikely(!gdb)) {
564 err = -ENOMEM;
565 goto out;
566 }
567
568 BUFFER_TRACE(gdb, "get_write_access");
569 err = ext4_journal_get_write_access(handle, sb, gdb,
570 EXT4_JTR_NONE);
571 if (err) {
572 brelse(bh: gdb);
573 goto out;
574 }
575 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
576 s_group_desc, j)->b_data, gdb->b_size);
577 set_buffer_uptodate(gdb);
578
579 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
580 if (unlikely(err)) {
581 brelse(bh: gdb);
582 goto out;
583 }
584 brelse(bh: gdb);
585 }
586
587 /* Zero out all of the reserved backup group descriptor
588 * table blocks
589 */
590 if (ext4_bg_has_super(sb, group)) {
591 err = sb_issue_zeroout(sb, block: gdblocks + start + 1,
592 nr_blocks: reserved_gdb, GFP_NOFS);
593 if (err)
594 goto out;
595 }
596
597handle_itb:
598 /* Initialize group tables of the group @group */
599 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
600 goto handle_bb;
601
602 /* Zero out all of the inode table blocks */
603 block = group_data[i].inode_table;
604 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
605 block, sbi->s_itb_per_group);
606 err = sb_issue_zeroout(sb, block, nr_blocks: sbi->s_itb_per_group,
607 GFP_NOFS);
608 if (err)
609 goto out;
610
611handle_bb:
612 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
613 goto handle_ib;
614
615 /* Initialize block bitmap of the @group */
616 block = group_data[i].block_bitmap;
617 err = ext4_resize_ensure_credits_batch(handle, credits: 1);
618 if (err < 0)
619 goto out;
620
621 bh = bclean(handle, sb, blk: block);
622 if (IS_ERR(ptr: bh)) {
623 err = PTR_ERR(ptr: bh);
624 goto out;
625 }
626 overhead = ext4_group_overhead_blocks(sb, group);
627 if (overhead != 0) {
628 ext4_debug("mark backup superblock %#04llx (+0)\n",
629 start);
630 mb_set_bits(bm: bh->b_data, cur: 0,
631 EXT4_NUM_B2C(sbi, overhead));
632 }
633 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
634 end_bit: sb->s_blocksize * 8, bitmap: bh->b_data);
635 err = ext4_handle_dirty_metadata(handle, NULL, bh);
636 brelse(bh);
637 if (err)
638 goto out;
639
640handle_ib:
641 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
642 continue;
643
644 /* Initialize inode bitmap of the @group */
645 block = group_data[i].inode_bitmap;
646 err = ext4_resize_ensure_credits_batch(handle, credits: 1);
647 if (err < 0)
648 goto out;
649 /* Mark unused entries in inode bitmap used */
650 bh = bclean(handle, sb, blk: block);
651 if (IS_ERR(ptr: bh)) {
652 err = PTR_ERR(ptr: bh);
653 goto out;
654 }
655
656 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
657 end_bit: sb->s_blocksize * 8, bitmap: bh->b_data);
658 err = ext4_handle_dirty_metadata(handle, NULL, bh);
659 brelse(bh);
660 if (err)
661 goto out;
662 }
663
664 /* Mark group tables in block bitmap */
665 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
666 count = group_table_count[j];
667 start = (&group_data[0].block_bitmap)[j];
668 block = start;
669 for (i = 1; i < flex_gd->count; i++) {
670 block += group_table_count[j];
671 if (block == (&group_data[i].block_bitmap)[j]) {
672 count += group_table_count[j];
673 continue;
674 }
675 err = set_flexbg_block_bitmap(sb, handle,
676 flex_gd,
677 EXT4_B2C(sbi, start),
678 EXT4_B2C(sbi,
679 start + count
680 - 1));
681 if (err)
682 goto out;
683 count = group_table_count[j];
684 start = (&group_data[i].block_bitmap)[j];
685 block = start;
686 }
687
688 err = set_flexbg_block_bitmap(sb, handle,
689 flex_gd,
690 EXT4_B2C(sbi, start),
691 EXT4_B2C(sbi,
692 start + count
693 - 1));
694 if (err)
695 goto out;
696 }
697
698out:
699 err2 = ext4_journal_stop(handle);
700 if (err2 && !err)
701 err = err2;
702
703 return err;
704}
705
706/*
707 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
708 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
709 * calling this for the first time. In a sparse filesystem it will be the
710 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
711 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
712 */
713unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
714 unsigned int *five, unsigned int *seven)
715{
716 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
717 unsigned int *min = three;
718 int mult = 3;
719 unsigned int ret;
720
721 if (ext4_has_feature_sparse_super2(sb)) {
722 do {
723 if (*min > 2)
724 return UINT_MAX;
725 ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
726 *min += 1;
727 } while (!ret);
728 return ret;
729 }
730
731 if (!ext4_has_feature_sparse_super(sb)) {
732 ret = *min;
733 *min += 1;
734 return ret;
735 }
736
737 if (*five < *min) {
738 min = five;
739 mult = 5;
740 }
741 if (*seven < *min) {
742 min = seven;
743 mult = 7;
744 }
745
746 ret = *min;
747 *min *= mult;
748
749 return ret;
750}
751
752/*
753 * Check that all of the backup GDT blocks are held in the primary GDT block.
754 * It is assumed that they are stored in group order. Returns the number of
755 * groups in current filesystem that have BACKUPS, or -ve error code.
756 */
757static int verify_reserved_gdb(struct super_block *sb,
758 ext4_group_t end,
759 struct buffer_head *primary)
760{
761 const ext4_fsblk_t blk = primary->b_blocknr;
762 unsigned three = 1;
763 unsigned five = 5;
764 unsigned seven = 7;
765 unsigned grp;
766 __le32 *p = (__le32 *)primary->b_data;
767 int gdbackups = 0;
768
769 while ((grp = ext4_list_backups(sb, three: &three, five: &five, seven: &seven)) < end) {
770 if (le32_to_cpu(*p++) !=
771 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
772 ext4_warning(sb, "reserved GDT %llu"
773 " missing grp %d (%llu)",
774 blk, grp,
775 grp *
776 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
777 blk);
778 return -EINVAL;
779 }
780 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
781 return -EFBIG;
782 }
783
784 return gdbackups;
785}
786
787/*
788 * Called when we need to bring a reserved group descriptor table block into
789 * use from the resize inode. The primary copy of the new GDT block currently
790 * is an indirect block (under the double indirect block in the resize inode).
791 * The new backup GDT blocks will be stored as leaf blocks in this indirect
792 * block, in group order. Even though we know all the block numbers we need,
793 * we check to ensure that the resize inode has actually reserved these blocks.
794 *
795 * Don't need to update the block bitmaps because the blocks are still in use.
796 *
797 * We get all of the error cases out of the way, so that we are sure to not
798 * fail once we start modifying the data on disk, because JBD has no rollback.
799 */
800static int add_new_gdb(handle_t *handle, struct inode *inode,
801 ext4_group_t group)
802{
803 struct super_block *sb = inode->i_sb;
804 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
805 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
806 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
807 struct buffer_head **o_group_desc, **n_group_desc = NULL;
808 struct buffer_head *dind = NULL;
809 struct buffer_head *gdb_bh = NULL;
810 int gdbackups;
811 struct ext4_iloc iloc = { .bh = NULL };
812 __le32 *data;
813 int err;
814
815 if (test_opt(sb, DEBUG))
816 printk(KERN_DEBUG
817 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
818 gdb_num);
819
820 gdb_bh = ext4_sb_bread(sb, block: gdblock, op_flags: 0);
821 if (IS_ERR(ptr: gdb_bh))
822 return PTR_ERR(ptr: gdb_bh);
823
824 gdbackups = verify_reserved_gdb(sb, end: group, primary: gdb_bh);
825 if (gdbackups < 0) {
826 err = gdbackups;
827 goto errout;
828 }
829
830 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
831 dind = ext4_sb_bread(sb, le32_to_cpu(*data), op_flags: 0);
832 if (IS_ERR(ptr: dind)) {
833 err = PTR_ERR(ptr: dind);
834 dind = NULL;
835 goto errout;
836 }
837
838 data = (__le32 *)dind->b_data;
839 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
840 ext4_warning(sb, "new group %u GDT block %llu not reserved",
841 group, gdblock);
842 err = -EINVAL;
843 goto errout;
844 }
845
846 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
847 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
848 EXT4_JTR_NONE);
849 if (unlikely(err))
850 goto errout;
851
852 BUFFER_TRACE(gdb_bh, "get_write_access");
853 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
854 if (unlikely(err))
855 goto errout;
856
857 BUFFER_TRACE(dind, "get_write_access");
858 err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
859 if (unlikely(err)) {
860 ext4_std_error(sb, err);
861 goto errout;
862 }
863
864 /* ext4_reserve_inode_write() gets a reference on the iloc */
865 err = ext4_reserve_inode_write(handle, inode, iloc: &iloc);
866 if (unlikely(err))
867 goto errout;
868
869 n_group_desc = kvmalloc(size: (gdb_num + 1) * sizeof(struct buffer_head *),
870 GFP_KERNEL);
871 if (!n_group_desc) {
872 err = -ENOMEM;
873 ext4_warning(sb, "not enough memory for %lu groups",
874 gdb_num + 1);
875 goto errout;
876 }
877
878 /*
879 * Finally, we have all of the possible failures behind us...
880 *
881 * Remove new GDT block from inode double-indirect block and clear out
882 * the new GDT block for use (which also "frees" the backup GDT blocks
883 * from the reserved inode). We don't need to change the bitmaps for
884 * these blocks, because they are marked as in-use from being in the
885 * reserved inode, and will become GDT blocks (primary and backup).
886 */
887 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
888 err = ext4_handle_dirty_metadata(handle, NULL, dind);
889 if (unlikely(err)) {
890 ext4_std_error(sb, err);
891 goto errout;
892 }
893 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
894 (9 - EXT4_SB(sb)->s_cluster_bits);
895 ext4_mark_iloc_dirty(handle, inode, iloc: &iloc);
896 memset(gdb_bh->b_data, 0, sb->s_blocksize);
897 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
898 if (unlikely(err)) {
899 ext4_std_error(sb, err);
900 iloc.bh = NULL;
901 goto errout;
902 }
903 brelse(bh: dind);
904
905 rcu_read_lock();
906 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
907 memcpy(n_group_desc, o_group_desc,
908 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
909 rcu_read_unlock();
910 n_group_desc[gdb_num] = gdb_bh;
911 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
912 EXT4_SB(sb)->s_gdb_count++;
913 ext4_kvfree_array_rcu(to_free: o_group_desc);
914
915 lock_buffer(bh: EXT4_SB(sb)->s_sbh);
916 le16_add_cpu(var: &es->s_reserved_gdt_blocks, val: -1);
917 ext4_superblock_csum_set(sb);
918 unlock_buffer(bh: EXT4_SB(sb)->s_sbh);
919 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
920 if (err)
921 ext4_std_error(sb, err);
922 return err;
923errout:
924 kvfree(addr: n_group_desc);
925 brelse(bh: iloc.bh);
926 brelse(bh: dind);
927 brelse(bh: gdb_bh);
928
929 ext4_debug("leaving with error %d\n", err);
930 return err;
931}
932
933/*
934 * If there is no available space in the existing block group descriptors for
935 * the new block group and there are no reserved block group descriptors, then
936 * the meta_bg feature will get enabled, and es->s_first_meta_bg will get set
937 * to the first block group that is managed using meta_bg and s_first_meta_bg
938 * must be a multiple of EXT4_DESC_PER_BLOCK(sb).
939 * This function will be called when first group of meta_bg is added to bring
940 * new group descriptors block of new added meta_bg.
941 */
942static int add_new_gdb_meta_bg(struct super_block *sb,
943 handle_t *handle, ext4_group_t group) {
944 ext4_fsblk_t gdblock;
945 struct buffer_head *gdb_bh;
946 struct buffer_head **o_group_desc, **n_group_desc;
947 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
948 int err;
949
950 gdblock = ext4_group_first_block_no(sb, group_no: group) +
951 ext4_bg_has_super(sb, group);
952 gdb_bh = ext4_sb_bread(sb, block: gdblock, op_flags: 0);
953 if (IS_ERR(ptr: gdb_bh))
954 return PTR_ERR(ptr: gdb_bh);
955 n_group_desc = kvmalloc(size: (gdb_num + 1) * sizeof(struct buffer_head *),
956 GFP_KERNEL);
957 if (!n_group_desc) {
958 brelse(bh: gdb_bh);
959 err = -ENOMEM;
960 ext4_warning(sb, "not enough memory for %lu groups",
961 gdb_num + 1);
962 return err;
963 }
964
965 rcu_read_lock();
966 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
967 memcpy(n_group_desc, o_group_desc,
968 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
969 rcu_read_unlock();
970 n_group_desc[gdb_num] = gdb_bh;
971
972 BUFFER_TRACE(gdb_bh, "get_write_access");
973 err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
974 if (err) {
975 kvfree(addr: n_group_desc);
976 brelse(bh: gdb_bh);
977 return err;
978 }
979
980 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
981 EXT4_SB(sb)->s_gdb_count++;
982 ext4_kvfree_array_rcu(to_free: o_group_desc);
983 return err;
984}
985
986/*
987 * Called when we are adding a new group which has a backup copy of each of
988 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
989 * We need to add these reserved backup GDT blocks to the resize inode, so
990 * that they are kept for future resizing and not allocated to files.
991 *
992 * Each reserved backup GDT block will go into a different indirect block.
993 * The indirect blocks are actually the primary reserved GDT blocks,
994 * so we know in advance what their block numbers are. We only get the
995 * double-indirect block to verify it is pointing to the primary reserved
996 * GDT blocks so we don't overwrite a data block by accident. The reserved
997 * backup GDT blocks are stored in their reserved primary GDT block.
998 */
999static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1000 ext4_group_t group)
1001{
1002 struct super_block *sb = inode->i_sb;
1003 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1004 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1005 struct buffer_head **primary;
1006 struct buffer_head *dind;
1007 struct ext4_iloc iloc;
1008 ext4_fsblk_t blk;
1009 __le32 *data, *end;
1010 int gdbackups = 0;
1011 int res, i;
1012 int err;
1013
1014 primary = kmalloc_array(n: reserved_gdb, size: sizeof(*primary), GFP_NOFS);
1015 if (!primary)
1016 return -ENOMEM;
1017
1018 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1019 dind = ext4_sb_bread(sb, le32_to_cpu(*data), op_flags: 0);
1020 if (IS_ERR(ptr: dind)) {
1021 err = PTR_ERR(ptr: dind);
1022 dind = NULL;
1023 goto exit_free;
1024 }
1025
1026 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1027 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1028 EXT4_ADDR_PER_BLOCK(sb));
1029 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1030
1031 /* Get each reserved primary GDT block and verify it holds backups */
1032 for (res = 0; res < reserved_gdb; res++, blk++) {
1033 if (le32_to_cpu(*data) != blk) {
1034 ext4_warning(sb, "reserved block %llu"
1035 " not at offset %ld",
1036 blk,
1037 (long)(data - (__le32 *)dind->b_data));
1038 err = -EINVAL;
1039 goto exit_bh;
1040 }
1041 primary[res] = ext4_sb_bread(sb, block: blk, op_flags: 0);
1042 if (IS_ERR(ptr: primary[res])) {
1043 err = PTR_ERR(ptr: primary[res]);
1044 primary[res] = NULL;
1045 goto exit_bh;
1046 }
1047 gdbackups = verify_reserved_gdb(sb, end: group, primary: primary[res]);
1048 if (gdbackups < 0) {
1049 brelse(bh: primary[res]);
1050 err = gdbackups;
1051 goto exit_bh;
1052 }
1053 if (++data >= end)
1054 data = (__le32 *)dind->b_data;
1055 }
1056
1057 for (i = 0; i < reserved_gdb; i++) {
1058 BUFFER_TRACE(primary[i], "get_write_access");
1059 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1060 EXT4_JTR_NONE)))
1061 goto exit_bh;
1062 }
1063
1064 if ((err = ext4_reserve_inode_write(handle, inode, iloc: &iloc)))
1065 goto exit_bh;
1066
1067 /*
1068 * Finally we can add each of the reserved backup GDT blocks from
1069 * the new group to its reserved primary GDT block.
1070 */
1071 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1072 for (i = 0; i < reserved_gdb; i++) {
1073 int err2;
1074 data = (__le32 *)primary[i]->b_data;
1075 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1076 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1077 if (!err)
1078 err = err2;
1079 }
1080
1081 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1082 ext4_mark_iloc_dirty(handle, inode, iloc: &iloc);
1083
1084exit_bh:
1085 while (--res >= 0)
1086 brelse(bh: primary[res]);
1087 brelse(bh: dind);
1088
1089exit_free:
1090 kfree(objp: primary);
1091
1092 return err;
1093}
1094
1095static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1096 ext4_group_t group)
1097{
1098 struct ext4_super_block *es = (struct ext4_super_block *) data;
1099
1100 es->s_block_group_nr = cpu_to_le16(group);
1101 if (ext4_has_metadata_csum(sb))
1102 es->s_checksum = ext4_superblock_csum(sb, es);
1103}
1104
1105/*
1106 * Update the backup copies of the ext4 metadata. These don't need to be part
1107 * of the main resize transaction, because e2fsck will re-write them if there
1108 * is a problem (basically only OOM will cause a problem). However, we
1109 * _should_ update the backups if possible, in case the primary gets trashed
1110 * for some reason and we need to run e2fsck from a backup superblock. The
1111 * important part is that the new block and inode counts are in the backup
1112 * superblocks, and the location of the new group metadata in the GDT backups.
1113 *
1114 * We do not need take the s_resize_lock for this, because these
1115 * blocks are not otherwise touched by the filesystem code when it is
1116 * mounted. We don't need to worry about last changing from
1117 * sbi->s_groups_count, because the worst that can happen is that we
1118 * do not copy the full number of backups at this time. The resize
1119 * which changed s_groups_count will backup again.
1120 */
1121static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1122 int size, int meta_bg)
1123{
1124 struct ext4_sb_info *sbi = EXT4_SB(sb);
1125 ext4_group_t last;
1126 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1127 unsigned three = 1;
1128 unsigned five = 5;
1129 unsigned seven = 7;
1130 ext4_group_t group = 0;
1131 int rest = sb->s_blocksize - size;
1132 handle_t *handle;
1133 int err = 0, err2;
1134
1135 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1136 if (IS_ERR(ptr: handle)) {
1137 group = 1;
1138 err = PTR_ERR(ptr: handle);
1139 goto exit_err;
1140 }
1141
1142 if (meta_bg == 0) {
1143 group = ext4_list_backups(sb, three: &three, five: &five, seven: &seven);
1144 last = sbi->s_groups_count;
1145 } else {
1146 group = ext4_get_group_number(sb, block: blk_off) + 1;
1147 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1148 }
1149
1150 while (group < sbi->s_groups_count) {
1151 struct buffer_head *bh;
1152 ext4_fsblk_t backup_block;
1153 int has_super = ext4_bg_has_super(sb, group);
1154 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group_no: group);
1155
1156 /* Out of journal space, and can't get more - abort - so sad */
1157 err = ext4_resize_ensure_credits_batch(handle, credits: 1);
1158 if (err < 0)
1159 break;
1160
1161 if (meta_bg == 0)
1162 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1163 else
1164 backup_block = first_block + has_super;
1165
1166 bh = sb_getblk(sb, block: backup_block);
1167 if (unlikely(!bh)) {
1168 err = -ENOMEM;
1169 break;
1170 }
1171 ext4_debug("update metadata backup %llu(+%llu)\n",
1172 backup_block, backup_block -
1173 ext4_group_first_block_no(sb, group));
1174 BUFFER_TRACE(bh, "get_write_access");
1175 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1176 EXT4_JTR_NONE))) {
1177 brelse(bh);
1178 break;
1179 }
1180 lock_buffer(bh);
1181 memcpy(bh->b_data, data, size);
1182 if (rest)
1183 memset(bh->b_data + size, 0, rest);
1184 if (has_super && (backup_block == first_block))
1185 ext4_set_block_group_nr(sb, data: bh->b_data, group);
1186 set_buffer_uptodate(bh);
1187 unlock_buffer(bh);
1188 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1189 if (unlikely(err))
1190 ext4_std_error(sb, err);
1191 brelse(bh);
1192
1193 if (meta_bg == 0)
1194 group = ext4_list_backups(sb, three: &three, five: &five, seven: &seven);
1195 else if (group == last)
1196 break;
1197 else
1198 group = last;
1199 }
1200 if ((err2 = ext4_journal_stop(handle)) && !err)
1201 err = err2;
1202
1203 /*
1204 * Ugh! Need to have e2fsck write the backup copies. It is too
1205 * late to revert the resize, we shouldn't fail just because of
1206 * the backup copies (they are only needed in case of corruption).
1207 *
1208 * However, if we got here we have a journal problem too, so we
1209 * can't really start a transaction to mark the superblock.
1210 * Chicken out and just set the flag on the hope it will be written
1211 * to disk, and if not - we will simply wait until next fsck.
1212 */
1213exit_err:
1214 if (err) {
1215 ext4_warning(sb, "can't update backup for group %u (err %d), "
1216 "forcing fsck on next reboot", group, err);
1217 sbi->s_mount_state &= ~EXT4_VALID_FS;
1218 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1219 mark_buffer_dirty(bh: sbi->s_sbh);
1220 }
1221}
1222
1223/*
1224 * ext4_add_new_descs() adds @count group descriptor of groups
1225 * starting at @group
1226 *
1227 * @handle: journal handle
1228 * @sb: super block
1229 * @group: the group no. of the first group desc to be added
1230 * @resize_inode: the resize inode
1231 * @count: number of group descriptors to be added
1232 */
1233static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1234 ext4_group_t group, struct inode *resize_inode,
1235 ext4_group_t count)
1236{
1237 struct ext4_sb_info *sbi = EXT4_SB(sb);
1238 struct ext4_super_block *es = sbi->s_es;
1239 struct buffer_head *gdb_bh;
1240 int i, gdb_off, gdb_num, err = 0;
1241 int meta_bg;
1242
1243 meta_bg = ext4_has_feature_meta_bg(sb);
1244 for (i = 0; i < count; i++, group++) {
1245 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1246 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1247
1248 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1249 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1250
1251 /*
1252 * We will only either add reserved group blocks to a backup group
1253 * or remove reserved blocks for the first group in a new group block.
1254 * Doing both would be mean more complex code, and sane people don't
1255 * use non-sparse filesystems anymore. This is already checked above.
1256 */
1257 if (gdb_off) {
1258 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1259 gdb_num);
1260 BUFFER_TRACE(gdb_bh, "get_write_access");
1261 err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1262 EXT4_JTR_NONE);
1263
1264 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1265 err = reserve_backup_gdb(handle, inode: resize_inode, group);
1266 } else if (meta_bg != 0) {
1267 err = add_new_gdb_meta_bg(sb, handle, group);
1268 } else {
1269 err = add_new_gdb(handle, inode: resize_inode, group);
1270 }
1271 if (err)
1272 break;
1273 }
1274 return err;
1275}
1276
1277static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1278{
1279 struct buffer_head *bh = sb_getblk(sb, block);
1280 if (unlikely(!bh))
1281 return NULL;
1282 if (!bh_uptodate_or_lock(bh)) {
1283 if (ext4_read_bh(bh, op_flags: 0, NULL) < 0) {
1284 brelse(bh);
1285 return NULL;
1286 }
1287 }
1288
1289 return bh;
1290}
1291
1292static int ext4_set_bitmap_checksums(struct super_block *sb,
1293 struct ext4_group_desc *gdp,
1294 struct ext4_new_group_data *group_data)
1295{
1296 struct buffer_head *bh;
1297
1298 if (!ext4_has_metadata_csum(sb))
1299 return 0;
1300
1301 bh = ext4_get_bitmap(sb, block: group_data->inode_bitmap);
1302 if (!bh)
1303 return -EIO;
1304 ext4_inode_bitmap_csum_set(sb, gdp, bh,
1305 EXT4_INODES_PER_GROUP(sb) / 8);
1306 brelse(bh);
1307
1308 bh = ext4_get_bitmap(sb, block: group_data->block_bitmap);
1309 if (!bh)
1310 return -EIO;
1311 ext4_block_bitmap_csum_set(sb, gdp, bh);
1312 brelse(bh);
1313
1314 return 0;
1315}
1316
1317/*
1318 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1319 */
1320static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1321 struct ext4_new_flex_group_data *flex_gd)
1322{
1323 struct ext4_new_group_data *group_data = flex_gd->groups;
1324 struct ext4_group_desc *gdp;
1325 struct ext4_sb_info *sbi = EXT4_SB(sb);
1326 struct buffer_head *gdb_bh;
1327 ext4_group_t group;
1328 __u16 *bg_flags = flex_gd->bg_flags;
1329 int i, gdb_off, gdb_num, err = 0;
1330
1331
1332 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1333 group = group_data->group;
1334
1335 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1336 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1337
1338 /*
1339 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1340 */
1341 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1342 /* Update group descriptor block for new group */
1343 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1344 gdb_off * EXT4_DESC_SIZE(sb));
1345
1346 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1347 ext4_block_bitmap_set(sb, bg: gdp, blk: group_data->block_bitmap);
1348 ext4_inode_bitmap_set(sb, bg: gdp, blk: group_data->inode_bitmap);
1349 err = ext4_set_bitmap_checksums(sb, gdp, group_data);
1350 if (err) {
1351 ext4_std_error(sb, err);
1352 break;
1353 }
1354
1355 ext4_inode_table_set(sb, bg: gdp, blk: group_data->inode_table);
1356 ext4_free_group_clusters_set(sb, bg: gdp,
1357 count: group_data->free_clusters_count);
1358 ext4_free_inodes_set(sb, bg: gdp, EXT4_INODES_PER_GROUP(sb));
1359 if (ext4_has_group_desc_csum(sb))
1360 ext4_itable_unused_set(sb, bg: gdp,
1361 EXT4_INODES_PER_GROUP(sb));
1362 gdp->bg_flags = cpu_to_le16(*bg_flags);
1363 ext4_group_desc_csum_set(sb, group, gdp);
1364
1365 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1366 if (unlikely(err)) {
1367 ext4_std_error(sb, err);
1368 break;
1369 }
1370
1371 /*
1372 * We can allocate memory for mb_alloc based on the new group
1373 * descriptor
1374 */
1375 err = ext4_mb_add_groupinfo(sb, i: group, desc: gdp);
1376 if (err)
1377 break;
1378 }
1379 return err;
1380}
1381
1382static void ext4_add_overhead(struct super_block *sb,
1383 const ext4_fsblk_t overhead)
1384{
1385 struct ext4_sb_info *sbi = EXT4_SB(sb);
1386 struct ext4_super_block *es = sbi->s_es;
1387
1388 sbi->s_overhead += overhead;
1389 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1390 smp_wmb();
1391}
1392
1393/*
1394 * ext4_update_super() updates the super block so that the newly added
1395 * groups can be seen by the filesystem.
1396 *
1397 * @sb: super block
1398 * @flex_gd: new added groups
1399 */
1400static void ext4_update_super(struct super_block *sb,
1401 struct ext4_new_flex_group_data *flex_gd)
1402{
1403 ext4_fsblk_t blocks_count = 0;
1404 ext4_fsblk_t free_blocks = 0;
1405 ext4_fsblk_t reserved_blocks = 0;
1406 struct ext4_new_group_data *group_data = flex_gd->groups;
1407 struct ext4_sb_info *sbi = EXT4_SB(sb);
1408 struct ext4_super_block *es = sbi->s_es;
1409 int i;
1410
1411 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1412 /*
1413 * Make the new blocks and inodes valid next. We do this before
1414 * increasing the group count so that once the group is enabled,
1415 * all of its blocks and inodes are already valid.
1416 *
1417 * We always allocate group-by-group, then block-by-block or
1418 * inode-by-inode within a group, so enabling these
1419 * blocks/inodes before the group is live won't actually let us
1420 * allocate the new space yet.
1421 */
1422 for (i = 0; i < flex_gd->count; i++) {
1423 blocks_count += group_data[i].blocks_count;
1424 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1425 }
1426
1427 reserved_blocks = ext4_r_blocks_count(es) * 100;
1428 reserved_blocks = div64_u64(dividend: reserved_blocks, divisor: ext4_blocks_count(es));
1429 reserved_blocks *= blocks_count;
1430 do_div(reserved_blocks, 100);
1431
1432 lock_buffer(bh: sbi->s_sbh);
1433 ext4_blocks_count_set(es, blk: ext4_blocks_count(es) + blocks_count);
1434 ext4_free_blocks_count_set(es, blk: ext4_free_blocks_count(es) + free_blocks);
1435 le32_add_cpu(var: &es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1436 flex_gd->count);
1437 le32_add_cpu(var: &es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1438 flex_gd->count);
1439
1440 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1441 /*
1442 * We need to protect s_groups_count against other CPUs seeing
1443 * inconsistent state in the superblock.
1444 *
1445 * The precise rules we use are:
1446 *
1447 * * Writers must perform a smp_wmb() after updating all
1448 * dependent data and before modifying the groups count
1449 *
1450 * * Readers must perform an smp_rmb() after reading the groups
1451 * count and before reading any dependent data.
1452 *
1453 * NB. These rules can be relaxed when checking the group count
1454 * while freeing data, as we can only allocate from a block
1455 * group after serialising against the group count, and we can
1456 * only then free after serialising in turn against that
1457 * allocation.
1458 */
1459 smp_wmb();
1460
1461 /* Update the global fs size fields */
1462 sbi->s_groups_count += flex_gd->count;
1463 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1464 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1465
1466 /* Update the reserved block counts only once the new group is
1467 * active. */
1468 ext4_r_blocks_count_set(es, blk: ext4_r_blocks_count(es) +
1469 reserved_blocks);
1470
1471 /* Update the free space counts */
1472 percpu_counter_add(fbc: &sbi->s_freeclusters_counter,
1473 EXT4_NUM_B2C(sbi, free_blocks));
1474 percpu_counter_add(fbc: &sbi->s_freeinodes_counter,
1475 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1476
1477 ext4_debug("free blocks count %llu",
1478 percpu_counter_read(&sbi->s_freeclusters_counter));
1479 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1480 ext4_group_t flex_group;
1481 struct flex_groups *fg;
1482
1483 flex_group = ext4_flex_group(sbi, block_group: group_data[0].group);
1484 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1485 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1486 v: &fg->free_clusters);
1487 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1488 v: &fg->free_inodes);
1489 }
1490
1491 /*
1492 * Update the fs overhead information.
1493 *
1494 * For bigalloc, if the superblock already has a properly calculated
1495 * overhead, update it with a value based on numbers already computed
1496 * above for the newly allocated capacity.
1497 */
1498 if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1499 ext4_add_overhead(sb,
1500 EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1501 else
1502 ext4_calculate_overhead(sb);
1503 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1504
1505 ext4_superblock_csum_set(sb);
1506 unlock_buffer(bh: sbi->s_sbh);
1507 if (test_opt(sb, DEBUG))
1508 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1509 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1510 blocks_count, free_blocks, reserved_blocks);
1511}
1512
1513/* Add a flex group to an fs. Ensure we handle all possible error conditions
1514 * _before_ we start modifying the filesystem, because we cannot abort the
1515 * transaction and not have it write the data to disk.
1516 */
1517static int ext4_flex_group_add(struct super_block *sb,
1518 struct inode *resize_inode,
1519 struct ext4_new_flex_group_data *flex_gd)
1520{
1521 struct ext4_sb_info *sbi = EXT4_SB(sb);
1522 struct ext4_super_block *es = sbi->s_es;
1523 ext4_fsblk_t o_blocks_count;
1524 ext4_grpblk_t last;
1525 ext4_group_t group;
1526 handle_t *handle;
1527 unsigned reserved_gdb;
1528 int err = 0, err2 = 0, credit;
1529
1530 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1531
1532 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1533 o_blocks_count = ext4_blocks_count(es);
1534 ext4_get_group_no_and_offset(sb, blocknr: o_blocks_count, blockgrpp: &group, offsetp: &last);
1535 BUG_ON(last);
1536
1537 err = setup_new_flex_group_blocks(sb, flex_gd);
1538 if (err)
1539 goto exit;
1540 /*
1541 * We will always be modifying at least the superblock and GDT
1542 * blocks. If we are adding a group past the last current GDT block,
1543 * we will also modify the inode and the dindirect block. If we
1544 * are adding a group with superblock/GDT backups we will also
1545 * modify each of the reserved GDT dindirect blocks.
1546 */
1547 credit = 3; /* sb, resize inode, resize inode dindirect */
1548 /* GDT blocks */
1549 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1550 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1551 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1552 if (IS_ERR(ptr: handle)) {
1553 err = PTR_ERR(ptr: handle);
1554 goto exit;
1555 }
1556
1557 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1558 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1559 EXT4_JTR_NONE);
1560 if (err)
1561 goto exit_journal;
1562
1563 group = flex_gd->groups[0].group;
1564 BUG_ON(group != sbi->s_groups_count);
1565 err = ext4_add_new_descs(handle, sb, group,
1566 resize_inode, count: flex_gd->count);
1567 if (err)
1568 goto exit_journal;
1569
1570 err = ext4_setup_new_descs(handle, sb, flex_gd);
1571 if (err)
1572 goto exit_journal;
1573
1574 ext4_update_super(sb, flex_gd);
1575
1576 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1577
1578exit_journal:
1579 err2 = ext4_journal_stop(handle);
1580 if (!err)
1581 err = err2;
1582
1583 if (!err) {
1584 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1585 int gdb_num_end = ((group + flex_gd->count - 1) /
1586 EXT4_DESC_PER_BLOCK(sb));
1587 int meta_bg = ext4_has_feature_meta_bg(sb);
1588 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1589 ext4_group_first_block_no(sb, group_no: 0);
1590
1591 update_backups(sb, blk_off: ext4_group_first_block_no(sb, group_no: 0),
1592 data: (char *)es, size: sizeof(struct ext4_super_block), meta_bg: 0);
1593 for (; gdb_num <= gdb_num_end; gdb_num++) {
1594 struct buffer_head *gdb_bh;
1595
1596 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1597 gdb_num);
1598 update_backups(sb, blk_off: gdb_bh->b_blocknr - padding_blocks,
1599 data: gdb_bh->b_data, size: gdb_bh->b_size, meta_bg);
1600 }
1601 }
1602exit:
1603 return err;
1604}
1605
1606static int ext4_setup_next_flex_gd(struct super_block *sb,
1607 struct ext4_new_flex_group_data *flex_gd,
1608 ext4_fsblk_t n_blocks_count,
1609 unsigned long flexbg_size)
1610{
1611 struct ext4_sb_info *sbi = EXT4_SB(sb);
1612 struct ext4_super_block *es = sbi->s_es;
1613 struct ext4_new_group_data *group_data = flex_gd->groups;
1614 ext4_fsblk_t o_blocks_count;
1615 ext4_group_t n_group;
1616 ext4_group_t group;
1617 ext4_group_t last_group;
1618 ext4_grpblk_t last;
1619 ext4_grpblk_t clusters_per_group;
1620 unsigned long i;
1621
1622 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1623
1624 o_blocks_count = ext4_blocks_count(es);
1625
1626 if (o_blocks_count == n_blocks_count)
1627 return 0;
1628
1629 ext4_get_group_no_and_offset(sb, blocknr: o_blocks_count, blockgrpp: &group, offsetp: &last);
1630 BUG_ON(last);
1631 ext4_get_group_no_and_offset(sb, blocknr: n_blocks_count - 1, blockgrpp: &n_group, offsetp: &last);
1632
1633 last_group = group | (flexbg_size - 1);
1634 if (last_group > n_group)
1635 last_group = n_group;
1636
1637 flex_gd->count = last_group - group + 1;
1638
1639 for (i = 0; i < flex_gd->count; i++) {
1640 int overhead;
1641
1642 group_data[i].group = group + i;
1643 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1644 overhead = ext4_group_overhead_blocks(sb, group: group + i);
1645 group_data[i].mdata_blocks = overhead;
1646 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1647 if (ext4_has_group_desc_csum(sb)) {
1648 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1649 EXT4_BG_INODE_UNINIT;
1650 if (!test_opt(sb, INIT_INODE_TABLE))
1651 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1652 } else
1653 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1654 }
1655
1656 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1657 /* We need to initialize block bitmap of last group. */
1658 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1659
1660 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1661 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1662 group_data[i - 1].free_clusters_count -= clusters_per_group -
1663 last - 1;
1664 }
1665
1666 return 1;
1667}
1668
1669/* Add group descriptor data to an existing or new group descriptor block.
1670 * Ensure we handle all possible error conditions _before_ we start modifying
1671 * the filesystem, because we cannot abort the transaction and not have it
1672 * write the data to disk.
1673 *
1674 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1675 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1676 *
1677 * We only need to hold the superblock lock while we are actually adding
1678 * in the new group's counts to the superblock. Prior to that we have
1679 * not really "added" the group at all. We re-check that we are still
1680 * adding in the last group in case things have changed since verifying.
1681 */
1682int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1683{
1684 struct ext4_new_flex_group_data flex_gd;
1685 struct ext4_sb_info *sbi = EXT4_SB(sb);
1686 struct ext4_super_block *es = sbi->s_es;
1687 int reserved_gdb = ext4_bg_has_super(sb, group: input->group) ?
1688 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1689 struct inode *inode = NULL;
1690 int gdb_off;
1691 int err;
1692 __u16 bg_flags = 0;
1693
1694 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1695
1696 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1697 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1698 return -EPERM;
1699 }
1700
1701 if (ext4_blocks_count(es) + input->blocks_count <
1702 ext4_blocks_count(es)) {
1703 ext4_warning(sb, "blocks_count overflow");
1704 return -EINVAL;
1705 }
1706
1707 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1708 le32_to_cpu(es->s_inodes_count)) {
1709 ext4_warning(sb, "inodes_count overflow");
1710 return -EINVAL;
1711 }
1712
1713 if (reserved_gdb || gdb_off == 0) {
1714 if (!ext4_has_feature_resize_inode(sb) ||
1715 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1716 ext4_warning(sb,
1717 "No reserved GDT blocks, can't resize");
1718 return -EPERM;
1719 }
1720 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1721 if (IS_ERR(ptr: inode)) {
1722 ext4_warning(sb, "Error opening resize inode");
1723 return PTR_ERR(ptr: inode);
1724 }
1725 }
1726
1727
1728 err = verify_group_input(sb, input);
1729 if (err)
1730 goto out;
1731
1732 err = ext4_alloc_flex_bg_array(sb, ngroup: input->group + 1);
1733 if (err)
1734 goto out;
1735
1736 err = ext4_mb_alloc_groupinfo(sb, ngroups: input->group + 1);
1737 if (err)
1738 goto out;
1739
1740 flex_gd.count = 1;
1741 flex_gd.groups = input;
1742 flex_gd.bg_flags = &bg_flags;
1743 err = ext4_flex_group_add(sb, resize_inode: inode, flex_gd: &flex_gd);
1744out:
1745 iput(inode);
1746 return err;
1747} /* ext4_group_add */
1748
1749/*
1750 * extend a group without checking assuming that checking has been done.
1751 */
1752static int ext4_group_extend_no_check(struct super_block *sb,
1753 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1754{
1755 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1756 handle_t *handle;
1757 int err = 0, err2;
1758
1759 /* We will update the superblock, one block bitmap, and
1760 * one group descriptor via ext4_group_add_blocks().
1761 */
1762 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1763 if (IS_ERR(ptr: handle)) {
1764 err = PTR_ERR(ptr: handle);
1765 ext4_warning(sb, "error %d on journal start", err);
1766 return err;
1767 }
1768
1769 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1770 err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1771 EXT4_JTR_NONE);
1772 if (err) {
1773 ext4_warning(sb, "error %d on journal write access", err);
1774 goto errout;
1775 }
1776
1777 lock_buffer(bh: EXT4_SB(sb)->s_sbh);
1778 ext4_blocks_count_set(es, blk: o_blocks_count + add);
1779 ext4_free_blocks_count_set(es, blk: ext4_free_blocks_count(es) + add);
1780 ext4_superblock_csum_set(sb);
1781 unlock_buffer(bh: EXT4_SB(sb)->s_sbh);
1782 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1783 o_blocks_count + add);
1784 /* We add the blocks to the bitmap and set the group need init bit */
1785 err = ext4_group_add_blocks(handle, sb, block: o_blocks_count, count: add);
1786 if (err)
1787 goto errout;
1788 ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1789 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1790 o_blocks_count + add);
1791errout:
1792 err2 = ext4_journal_stop(handle);
1793 if (err2 && !err)
1794 err = err2;
1795
1796 if (!err) {
1797 if (test_opt(sb, DEBUG))
1798 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1799 "blocks\n", ext4_blocks_count(es));
1800 update_backups(sb, blk_off: ext4_group_first_block_no(sb, group_no: 0),
1801 data: (char *)es, size: sizeof(struct ext4_super_block), meta_bg: 0);
1802 }
1803 return err;
1804}
1805
1806/*
1807 * Extend the filesystem to the new number of blocks specified. This entry
1808 * point is only used to extend the current filesystem to the end of the last
1809 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1810 * for emergencies (because it has no dependencies on reserved blocks).
1811 *
1812 * If we _really_ wanted, we could use default values to call ext4_group_add()
1813 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1814 * GDT blocks are reserved to grow to the desired size.
1815 */
1816int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1817 ext4_fsblk_t n_blocks_count)
1818{
1819 ext4_fsblk_t o_blocks_count;
1820 ext4_grpblk_t last;
1821 ext4_grpblk_t add;
1822 struct buffer_head *bh;
1823 ext4_group_t group;
1824
1825 o_blocks_count = ext4_blocks_count(es);
1826
1827 if (test_opt(sb, DEBUG))
1828 ext4_msg(sb, KERN_DEBUG,
1829 "extending last group from %llu to %llu blocks",
1830 o_blocks_count, n_blocks_count);
1831
1832 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1833 return 0;
1834
1835 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1836 ext4_msg(sb, KERN_ERR,
1837 "filesystem too large to resize to %llu blocks safely",
1838 n_blocks_count);
1839 return -EINVAL;
1840 }
1841
1842 if (n_blocks_count < o_blocks_count) {
1843 ext4_warning(sb, "can't shrink FS - resize aborted");
1844 return -EINVAL;
1845 }
1846
1847 /* Handle the remaining blocks in the last group only. */
1848 ext4_get_group_no_and_offset(sb, blocknr: o_blocks_count, blockgrpp: &group, offsetp: &last);
1849
1850 if (last == 0) {
1851 ext4_warning(sb, "need to use ext2online to resize further");
1852 return -EPERM;
1853 }
1854
1855 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1856
1857 if (o_blocks_count + add < o_blocks_count) {
1858 ext4_warning(sb, "blocks_count overflow");
1859 return -EINVAL;
1860 }
1861
1862 if (o_blocks_count + add > n_blocks_count)
1863 add = n_blocks_count - o_blocks_count;
1864
1865 if (o_blocks_count + add < n_blocks_count)
1866 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1867 o_blocks_count + add, add);
1868
1869 /* See if the device is actually as big as what was requested */
1870 bh = ext4_sb_bread(sb, block: o_blocks_count + add - 1, op_flags: 0);
1871 if (IS_ERR(ptr: bh)) {
1872 ext4_warning(sb, "can't read last block, resize aborted");
1873 return -ENOSPC;
1874 }
1875 brelse(bh);
1876
1877 return ext4_group_extend_no_check(sb, o_blocks_count, add);
1878} /* ext4_group_extend */
1879
1880
1881static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1882{
1883 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1884}
1885
1886/*
1887 * Release the resize inode and drop the resize_inode feature if there
1888 * are no more reserved gdt blocks, and then convert the file system
1889 * to enable meta_bg
1890 */
1891static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1892{
1893 handle_t *handle;
1894 struct ext4_sb_info *sbi = EXT4_SB(sb);
1895 struct ext4_super_block *es = sbi->s_es;
1896 struct ext4_inode_info *ei = EXT4_I(inode);
1897 ext4_fsblk_t nr;
1898 int i, ret, err = 0;
1899 int credits = 1;
1900
1901 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1902 if (inode) {
1903 if (es->s_reserved_gdt_blocks) {
1904 ext4_error(sb, "Unexpected non-zero "
1905 "s_reserved_gdt_blocks");
1906 return -EPERM;
1907 }
1908
1909 /* Do a quick sanity check of the resize inode */
1910 if (inode->i_blocks != 1 << (inode->i_blkbits -
1911 (9 - sbi->s_cluster_bits)))
1912 goto invalid_resize_inode;
1913 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1914 if (i == EXT4_DIND_BLOCK) {
1915 if (ei->i_data[i])
1916 continue;
1917 else
1918 goto invalid_resize_inode;
1919 }
1920 if (ei->i_data[i])
1921 goto invalid_resize_inode;
1922 }
1923 credits += 3; /* block bitmap, bg descriptor, resize inode */
1924 }
1925
1926 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1927 if (IS_ERR(ptr: handle))
1928 return PTR_ERR(ptr: handle);
1929
1930 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1931 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1932 EXT4_JTR_NONE);
1933 if (err)
1934 goto errout;
1935
1936 lock_buffer(bh: sbi->s_sbh);
1937 ext4_clear_feature_resize_inode(sb);
1938 ext4_set_feature_meta_bg(sb);
1939 sbi->s_es->s_first_meta_bg =
1940 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1941 ext4_superblock_csum_set(sb);
1942 unlock_buffer(bh: sbi->s_sbh);
1943
1944 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1945 if (err) {
1946 ext4_std_error(sb, err);
1947 goto errout;
1948 }
1949
1950 if (inode) {
1951 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1952 ext4_free_blocks(handle, inode, NULL, block: nr, count: 1,
1953 EXT4_FREE_BLOCKS_METADATA |
1954 EXT4_FREE_BLOCKS_FORGET);
1955 ei->i_data[EXT4_DIND_BLOCK] = 0;
1956 inode->i_blocks = 0;
1957
1958 err = ext4_mark_inode_dirty(handle, inode);
1959 if (err)
1960 ext4_std_error(sb, err);
1961 }
1962
1963errout:
1964 ret = ext4_journal_stop(handle);
1965 return err ? err : ret;
1966
1967invalid_resize_inode:
1968 ext4_error(sb, "corrupted/inconsistent resize inode");
1969 return -EINVAL;
1970}
1971
1972/*
1973 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1974 *
1975 * @sb: super block of the fs to be resized
1976 * @n_blocks_count: the number of blocks resides in the resized fs
1977 */
1978int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1979{
1980 struct ext4_new_flex_group_data *flex_gd = NULL;
1981 struct ext4_sb_info *sbi = EXT4_SB(sb);
1982 struct ext4_super_block *es = sbi->s_es;
1983 struct buffer_head *bh;
1984 struct inode *resize_inode = NULL;
1985 ext4_grpblk_t add, offset;
1986 unsigned long n_desc_blocks;
1987 unsigned long o_desc_blocks;
1988 ext4_group_t o_group;
1989 ext4_group_t n_group;
1990 ext4_fsblk_t o_blocks_count;
1991 ext4_fsblk_t n_blocks_count_retry = 0;
1992 unsigned long last_update_time = 0;
1993 int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1994 int meta_bg;
1995
1996 /* See if the device is actually as big as what was requested */
1997 bh = ext4_sb_bread(sb, block: n_blocks_count - 1, op_flags: 0);
1998 if (IS_ERR(ptr: bh)) {
1999 ext4_warning(sb, "can't read last block, resize aborted");
2000 return -ENOSPC;
2001 }
2002 brelse(bh);
2003
2004 /*
2005 * For bigalloc, trim the requested size to the nearest cluster
2006 * boundary to avoid creating an unusable filesystem. We do this
2007 * silently, instead of returning an error, to avoid breaking
2008 * callers that blindly resize the filesystem to the full size of
2009 * the underlying block device.
2010 */
2011 if (ext4_has_feature_bigalloc(sb))
2012 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2013
2014retry:
2015 o_blocks_count = ext4_blocks_count(es);
2016
2017 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2018 "to %llu blocks", o_blocks_count, n_blocks_count);
2019
2020 if (n_blocks_count < o_blocks_count) {
2021 /* On-line shrinking not supported */
2022 ext4_warning(sb, "can't shrink FS - resize aborted");
2023 return -EINVAL;
2024 }
2025
2026 if (n_blocks_count == o_blocks_count)
2027 /* Nothing need to do */
2028 return 0;
2029
2030 n_group = ext4_get_group_number(sb, block: n_blocks_count - 1);
2031 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2032 ext4_warning(sb, "resize would cause inodes_count overflow");
2033 return -EINVAL;
2034 }
2035 ext4_get_group_no_and_offset(sb, blocknr: o_blocks_count - 1, blockgrpp: &o_group, offsetp: &offset);
2036
2037 n_desc_blocks = num_desc_blocks(sb, groups: n_group + 1);
2038 o_desc_blocks = num_desc_blocks(sb, groups: sbi->s_groups_count);
2039
2040 meta_bg = ext4_has_feature_meta_bg(sb);
2041
2042 if (ext4_has_feature_resize_inode(sb)) {
2043 if (meta_bg) {
2044 ext4_error(sb, "resize_inode and meta_bg enabled "
2045 "simultaneously");
2046 return -EINVAL;
2047 }
2048 if (n_desc_blocks > o_desc_blocks +
2049 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2050 n_blocks_count_retry = n_blocks_count;
2051 n_desc_blocks = o_desc_blocks +
2052 le16_to_cpu(es->s_reserved_gdt_blocks);
2053 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2054 n_blocks_count = (ext4_fsblk_t)n_group *
2055 EXT4_BLOCKS_PER_GROUP(sb) +
2056 le32_to_cpu(es->s_first_data_block);
2057 n_group--; /* set to last group number */
2058 }
2059
2060 if (!resize_inode)
2061 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2062 EXT4_IGET_SPECIAL);
2063 if (IS_ERR(ptr: resize_inode)) {
2064 ext4_warning(sb, "Error opening resize inode");
2065 return PTR_ERR(ptr: resize_inode);
2066 }
2067 }
2068
2069 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2070 err = ext4_convert_meta_bg(sb, inode: resize_inode);
2071 if (err)
2072 goto out;
2073 if (resize_inode) {
2074 iput(resize_inode);
2075 resize_inode = NULL;
2076 }
2077 if (n_blocks_count_retry) {
2078 n_blocks_count = n_blocks_count_retry;
2079 n_blocks_count_retry = 0;
2080 goto retry;
2081 }
2082 }
2083
2084 /*
2085 * Make sure the last group has enough space so that it's
2086 * guaranteed to have enough space for all metadata blocks
2087 * that it might need to hold. (We might not need to store
2088 * the inode table blocks in the last block group, but there
2089 * will be cases where this might be needed.)
2090 */
2091 if ((ext4_group_first_block_no(sb, group_no: n_group) +
2092 ext4_group_overhead_blocks(sb, group: n_group) + 2 +
2093 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2094 n_blocks_count = ext4_group_first_block_no(sb, group_no: n_group);
2095 n_group--;
2096 n_blocks_count_retry = 0;
2097 if (resize_inode) {
2098 iput(resize_inode);
2099 resize_inode = NULL;
2100 }
2101 goto retry;
2102 }
2103
2104 /* extend the last group */
2105 if (n_group == o_group)
2106 add = n_blocks_count - o_blocks_count;
2107 else
2108 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2109 if (add > 0) {
2110 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2111 if (err)
2112 goto out;
2113 }
2114
2115 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2116 goto out;
2117
2118 err = ext4_alloc_flex_bg_array(sb, ngroup: n_group + 1);
2119 if (err)
2120 goto out;
2121
2122 err = ext4_mb_alloc_groupinfo(sb, ngroups: n_group + 1);
2123 if (err)
2124 goto out;
2125
2126 flex_gd = alloc_flex_gd(flexbg_size);
2127 if (flex_gd == NULL) {
2128 err = -ENOMEM;
2129 goto out;
2130 }
2131
2132 /* Add flex groups. Note that a regular group is a
2133 * flex group with 1 group.
2134 */
2135 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2136 flexbg_size)) {
2137 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2138 if (last_update_time)
2139 ext4_msg(sb, KERN_INFO,
2140 "resized to %llu blocks",
2141 ext4_blocks_count(es));
2142 last_update_time = jiffies;
2143 }
2144 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2145 break;
2146 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2147 if (unlikely(err))
2148 break;
2149 }
2150
2151 if (!err && n_blocks_count_retry) {
2152 n_blocks_count = n_blocks_count_retry;
2153 n_blocks_count_retry = 0;
2154 free_flex_gd(flex_gd);
2155 flex_gd = NULL;
2156 if (resize_inode) {
2157 iput(resize_inode);
2158 resize_inode = NULL;
2159 }
2160 goto retry;
2161 }
2162
2163out:
2164 if (flex_gd)
2165 free_flex_gd(flex_gd);
2166 if (resize_inode != NULL)
2167 iput(resize_inode);
2168 if (err)
2169 ext4_warning(sb, "error (%d) occurred during "
2170 "file system resize", err);
2171 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2172 ext4_blocks_count(es));
2173 return err;
2174}
2175

source code of linux/fs/ext4/resize.c