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 #define EXT4FS_DEBUG
14
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17
18 #include "ext4_jbd2.h"
19
20 struct ext4_rcu_ptr {
21 struct rcu_head rcu;
22 void *ptr;
23 };
24
ext4_rcu_ptr_callback(struct rcu_head * head)25 static 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
ext4_kvfree_array_rcu(void * to_free)34 void 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
ext4_resize_begin(struct super_block * sb)47 int 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 the reserved GDT blocks is non-zero, the resize_inode feature
57 * should always be set.
58 */
59 if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
60 !ext4_has_feature_resize_inode(sb)) {
61 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
62 return -EFSCORRUPTED;
63 }
64
65 /*
66 * If we are not using the primary superblock/GDT copy don't resize,
67 * because the user tools have no way of handling this. Probably a
68 * bad time to do it anyways.
69 */
70 if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
71 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
72 ext4_warning(sb, "won't resize using backup superblock at %llu",
73 (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
74 return -EPERM;
75 }
76
77 /*
78 * We are not allowed to do online-resizing on a filesystem mounted
79 * with error, because it can destroy the filesystem easily.
80 */
81 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
82 ext4_warning(sb, "There are errors in the filesystem, "
83 "so online resizing is not allowed");
84 return -EPERM;
85 }
86
87 if (ext4_has_feature_sparse_super2(sb)) {
88 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
89 return -EOPNOTSUPP;
90 }
91
92 if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
93 &EXT4_SB(sb)->s_ext4_flags))
94 ret = -EBUSY;
95
96 return ret;
97 }
98
ext4_resize_end(struct super_block * sb)99 void ext4_resize_end(struct super_block *sb)
100 {
101 clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
102 smp_mb__after_atomic();
103 }
104
ext4_meta_bg_first_group(struct super_block * sb,ext4_group_t group)105 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
106 ext4_group_t group) {
107 return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
108 EXT4_DESC_PER_BLOCK_BITS(sb);
109 }
110
ext4_meta_bg_first_block_no(struct super_block * sb,ext4_group_t group)111 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
112 ext4_group_t group) {
113 group = ext4_meta_bg_first_group(sb, group);
114 return ext4_group_first_block_no(sb, group);
115 }
116
ext4_group_overhead_blocks(struct super_block * sb,ext4_group_t group)117 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
118 ext4_group_t group) {
119 ext4_grpblk_t overhead;
120 overhead = ext4_bg_num_gdb(sb, group);
121 if (ext4_bg_has_super(sb, group))
122 overhead += 1 +
123 le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
124 return overhead;
125 }
126
127 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
128 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
129
verify_group_input(struct super_block * sb,struct ext4_new_group_data * input)130 static int verify_group_input(struct super_block *sb,
131 struct ext4_new_group_data *input)
132 {
133 struct ext4_sb_info *sbi = EXT4_SB(sb);
134 struct ext4_super_block *es = sbi->s_es;
135 ext4_fsblk_t start = ext4_blocks_count(es);
136 ext4_fsblk_t end = start + input->blocks_count;
137 ext4_group_t group = input->group;
138 ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
139 unsigned overhead;
140 ext4_fsblk_t metaend;
141 struct buffer_head *bh = NULL;
142 ext4_grpblk_t free_blocks_count, offset;
143 int err = -EINVAL;
144
145 if (group != sbi->s_groups_count) {
146 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
147 input->group, sbi->s_groups_count);
148 return -EINVAL;
149 }
150
151 overhead = ext4_group_overhead_blocks(sb, group);
152 metaend = start + overhead;
153 input->free_clusters_count = free_blocks_count =
154 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
155
156 if (test_opt(sb, DEBUG))
157 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
158 "(%d free, %u reserved)\n",
159 ext4_bg_has_super(sb, input->group) ? "normal" :
160 "no-super", input->group, input->blocks_count,
161 free_blocks_count, input->reserved_blocks);
162
163 ext4_get_group_no_and_offset(sb, start, NULL, &offset);
164 if (offset != 0)
165 ext4_warning(sb, "Last group not full");
166 else if (input->reserved_blocks > input->blocks_count / 5)
167 ext4_warning(sb, "Reserved blocks too high (%u)",
168 input->reserved_blocks);
169 else if (free_blocks_count < 0)
170 ext4_warning(sb, "Bad blocks count %u",
171 input->blocks_count);
172 else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
173 err = PTR_ERR(bh);
174 bh = NULL;
175 ext4_warning(sb, "Cannot read last block (%llu)",
176 end - 1);
177 } else if (outside(input->block_bitmap, start, end))
178 ext4_warning(sb, "Block bitmap not in group (block %llu)",
179 (unsigned long long)input->block_bitmap);
180 else if (outside(input->inode_bitmap, start, end))
181 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
182 (unsigned long long)input->inode_bitmap);
183 else if (outside(input->inode_table, start, end) ||
184 outside(itend - 1, start, end))
185 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
186 (unsigned long long)input->inode_table, itend - 1);
187 else if (input->inode_bitmap == input->block_bitmap)
188 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
189 (unsigned long long)input->block_bitmap);
190 else if (inside(input->block_bitmap, input->inode_table, itend))
191 ext4_warning(sb, "Block bitmap (%llu) in inode table "
192 "(%llu-%llu)",
193 (unsigned long long)input->block_bitmap,
194 (unsigned long long)input->inode_table, itend - 1);
195 else if (inside(input->inode_bitmap, input->inode_table, itend))
196 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
197 "(%llu-%llu)",
198 (unsigned long long)input->inode_bitmap,
199 (unsigned long long)input->inode_table, itend - 1);
200 else if (inside(input->block_bitmap, start, metaend))
201 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
202 (unsigned long long)input->block_bitmap,
203 start, metaend - 1);
204 else if (inside(input->inode_bitmap, start, metaend))
205 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
206 (unsigned long long)input->inode_bitmap,
207 start, metaend - 1);
208 else if (inside(input->inode_table, start, metaend) ||
209 inside(itend - 1, start, metaend))
210 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
211 "(%llu-%llu)",
212 (unsigned long long)input->inode_table,
213 itend - 1, start, metaend - 1);
214 else
215 err = 0;
216 brelse(bh);
217
218 return err;
219 }
220
221 /*
222 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
223 * group each time.
224 */
225 struct ext4_new_flex_group_data {
226 struct ext4_new_group_data *groups; /* new_group_data for groups
227 in the flex group */
228 __u16 *bg_flags; /* block group flags of groups
229 in @groups */
230 ext4_group_t resize_bg; /* number of allocated
231 new_group_data */
232 ext4_group_t count; /* number of groups in @groups
233 */
234 };
235
236 /*
237 * Avoiding memory allocation failures due to too many groups added each time.
238 */
239 #define MAX_RESIZE_BG 16384
240
241 /*
242 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
243 * @flexbg_size.
244 *
245 * Returns NULL on failure otherwise address of the allocated structure.
246 */
alloc_flex_gd(unsigned int flexbg_size)247 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size)
248 {
249 struct ext4_new_flex_group_data *flex_gd;
250
251 flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
252 if (flex_gd == NULL)
253 goto out3;
254
255 if (unlikely(flexbg_size > MAX_RESIZE_BG))
256 flex_gd->resize_bg = MAX_RESIZE_BG;
257 else
258 flex_gd->resize_bg = flexbg_size;
259
260 flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
261 sizeof(struct ext4_new_group_data),
262 GFP_NOFS);
263 if (flex_gd->groups == NULL)
264 goto out2;
265
266 flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
267 GFP_NOFS);
268 if (flex_gd->bg_flags == NULL)
269 goto out1;
270
271 return flex_gd;
272
273 out1:
274 kfree(flex_gd->groups);
275 out2:
276 kfree(flex_gd);
277 out3:
278 return NULL;
279 }
280
free_flex_gd(struct ext4_new_flex_group_data * flex_gd)281 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
282 {
283 kfree(flex_gd->bg_flags);
284 kfree(flex_gd->groups);
285 kfree(flex_gd);
286 }
287
288 /*
289 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
290 * and inode tables for a flex group.
291 *
292 * This function is used by 64bit-resize. Note that this function allocates
293 * group tables from the 1st group of groups contained by @flexgd, which may
294 * be a partial of a flex group.
295 *
296 * @sb: super block of fs to which the groups belongs
297 *
298 * Returns 0 on a successful allocation of the metadata blocks in the
299 * block group.
300 */
ext4_alloc_group_tables(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,unsigned int flexbg_size)301 static int ext4_alloc_group_tables(struct super_block *sb,
302 struct ext4_new_flex_group_data *flex_gd,
303 unsigned int flexbg_size)
304 {
305 struct ext4_new_group_data *group_data = flex_gd->groups;
306 ext4_fsblk_t start_blk;
307 ext4_fsblk_t last_blk;
308 ext4_group_t src_group;
309 ext4_group_t bb_index = 0;
310 ext4_group_t ib_index = 0;
311 ext4_group_t it_index = 0;
312 ext4_group_t group;
313 ext4_group_t last_group;
314 unsigned overhead;
315 __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
316 int i;
317
318 BUG_ON(flex_gd->count == 0 || group_data == NULL);
319
320 src_group = group_data[0].group;
321 last_group = src_group + flex_gd->count - 1;
322
323 BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
324 (last_group & ~(flexbg_size - 1))));
325 next_group:
326 group = group_data[0].group;
327 if (src_group >= group_data[0].group + flex_gd->count)
328 return -ENOSPC;
329 start_blk = ext4_group_first_block_no(sb, src_group);
330 last_blk = start_blk + group_data[src_group - group].blocks_count;
331
332 overhead = ext4_group_overhead_blocks(sb, src_group);
333
334 start_blk += overhead;
335
336 /* We collect contiguous blocks as much as possible. */
337 src_group++;
338 for (; src_group <= last_group; src_group++) {
339 overhead = ext4_group_overhead_blocks(sb, src_group);
340 if (overhead == 0)
341 last_blk += group_data[src_group - group].blocks_count;
342 else
343 break;
344 }
345
346 /* Allocate block bitmaps */
347 for (; bb_index < flex_gd->count; bb_index++) {
348 if (start_blk >= last_blk)
349 goto next_group;
350 group_data[bb_index].block_bitmap = start_blk++;
351 group = ext4_get_group_number(sb, start_blk - 1);
352 group -= group_data[0].group;
353 group_data[group].mdata_blocks++;
354 flex_gd->bg_flags[group] &= uninit_mask;
355 }
356
357 /* Allocate inode bitmaps */
358 for (; ib_index < flex_gd->count; ib_index++) {
359 if (start_blk >= last_blk)
360 goto next_group;
361 group_data[ib_index].inode_bitmap = start_blk++;
362 group = ext4_get_group_number(sb, start_blk - 1);
363 group -= group_data[0].group;
364 group_data[group].mdata_blocks++;
365 flex_gd->bg_flags[group] &= uninit_mask;
366 }
367
368 /* Allocate inode tables */
369 for (; it_index < flex_gd->count; it_index++) {
370 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
371 ext4_fsblk_t next_group_start;
372
373 if (start_blk + itb > last_blk)
374 goto next_group;
375 group_data[it_index].inode_table = start_blk;
376 group = ext4_get_group_number(sb, start_blk);
377 next_group_start = ext4_group_first_block_no(sb, group + 1);
378 group -= group_data[0].group;
379
380 if (start_blk + itb > next_group_start) {
381 flex_gd->bg_flags[group + 1] &= uninit_mask;
382 overhead = start_blk + itb - next_group_start;
383 group_data[group + 1].mdata_blocks += overhead;
384 itb -= overhead;
385 }
386
387 group_data[group].mdata_blocks += itb;
388 flex_gd->bg_flags[group] &= uninit_mask;
389 start_blk += EXT4_SB(sb)->s_itb_per_group;
390 }
391
392 /* Update free clusters count to exclude metadata blocks */
393 for (i = 0; i < flex_gd->count; i++) {
394 group_data[i].free_clusters_count -=
395 EXT4_NUM_B2C(EXT4_SB(sb),
396 group_data[i].mdata_blocks);
397 }
398
399 if (test_opt(sb, DEBUG)) {
400 int i;
401 group = group_data[0].group;
402
403 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
404 "%u groups, flexbg size is %u:\n", flex_gd->count,
405 flexbg_size);
406
407 for (i = 0; i < flex_gd->count; i++) {
408 ext4_debug(
409 "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
410 ext4_bg_has_super(sb, group + i) ? "normal" :
411 "no-super", group + i,
412 group_data[i].blocks_count,
413 group_data[i].free_clusters_count,
414 group_data[i].mdata_blocks);
415 }
416 }
417 return 0;
418 }
419
bclean(handle_t * handle,struct super_block * sb,ext4_fsblk_t blk)420 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
421 ext4_fsblk_t blk)
422 {
423 struct buffer_head *bh;
424 int err;
425
426 bh = sb_getblk(sb, blk);
427 if (unlikely(!bh))
428 return ERR_PTR(-ENOMEM);
429 BUFFER_TRACE(bh, "get_write_access");
430 if ((err = ext4_journal_get_write_access(handle, bh))) {
431 brelse(bh);
432 bh = ERR_PTR(err);
433 } else {
434 memset(bh->b_data, 0, sb->s_blocksize);
435 set_buffer_uptodate(bh);
436 }
437
438 return bh;
439 }
440
ext4_resize_ensure_credits_batch(handle_t * handle,int credits)441 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
442 {
443 return ext4_journal_ensure_credits_fn(handle, credits,
444 EXT4_MAX_TRANS_DATA, 0, 0);
445 }
446
447 /*
448 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
449 *
450 * Helper function for ext4_setup_new_group_blocks() which set .
451 *
452 * @sb: super block
453 * @handle: journal handle
454 * @flex_gd: flex group data
455 */
set_flexbg_block_bitmap(struct super_block * sb,handle_t * handle,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t first_cluster,ext4_fsblk_t last_cluster)456 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
457 struct ext4_new_flex_group_data *flex_gd,
458 ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
459 {
460 struct ext4_sb_info *sbi = EXT4_SB(sb);
461 ext4_group_t count = last_cluster - first_cluster + 1;
462 ext4_group_t count2;
463
464 ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
465 last_cluster);
466 for (count2 = count; count > 0;
467 count -= count2, first_cluster += count2) {
468 ext4_fsblk_t start;
469 struct buffer_head *bh;
470 ext4_group_t group;
471 int err;
472
473 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
474 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
475 group -= flex_gd->groups[0].group;
476
477 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
478 if (count2 > count)
479 count2 = count;
480
481 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
482 BUG_ON(flex_gd->count > 1);
483 continue;
484 }
485
486 err = ext4_resize_ensure_credits_batch(handle, 1);
487 if (err < 0)
488 return err;
489
490 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
491 if (unlikely(!bh))
492 return -ENOMEM;
493
494 BUFFER_TRACE(bh, "get_write_access");
495 err = ext4_journal_get_write_access(handle, bh);
496 if (err) {
497 brelse(bh);
498 return err;
499 }
500 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
501 first_cluster, first_cluster - start, count2);
502 ext4_set_bits(bh->b_data, first_cluster - start, count2);
503
504 err = ext4_handle_dirty_metadata(handle, NULL, bh);
505 brelse(bh);
506 if (unlikely(err))
507 return err;
508 }
509
510 return 0;
511 }
512
513 /*
514 * Set up the block and inode bitmaps, and the inode table for the new groups.
515 * This doesn't need to be part of the main transaction, since we are only
516 * changing blocks outside the actual filesystem. We still do journaling to
517 * ensure the recovery is correct in case of a failure just after resize.
518 * If any part of this fails, we simply abort the resize.
519 *
520 * setup_new_flex_group_blocks handles a flex group as follow:
521 * 1. copy super block and GDT, and initialize group tables if necessary.
522 * In this step, we only set bits in blocks bitmaps for blocks taken by
523 * super block and GDT.
524 * 2. allocate group tables in block bitmaps, that is, set bits in block
525 * bitmap for blocks taken by group tables.
526 */
setup_new_flex_group_blocks(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)527 static int setup_new_flex_group_blocks(struct super_block *sb,
528 struct ext4_new_flex_group_data *flex_gd)
529 {
530 int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
531 ext4_fsblk_t start;
532 ext4_fsblk_t block;
533 struct ext4_sb_info *sbi = EXT4_SB(sb);
534 struct ext4_super_block *es = sbi->s_es;
535 struct ext4_new_group_data *group_data = flex_gd->groups;
536 __u16 *bg_flags = flex_gd->bg_flags;
537 handle_t *handle;
538 ext4_group_t group, count;
539 struct buffer_head *bh = NULL;
540 int reserved_gdb, i, j, err = 0, err2;
541 int meta_bg;
542
543 BUG_ON(!flex_gd->count || !group_data ||
544 group_data[0].group != sbi->s_groups_count);
545
546 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
547 meta_bg = ext4_has_feature_meta_bg(sb);
548
549 /* This transaction may be extended/restarted along the way */
550 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
551 if (IS_ERR(handle))
552 return PTR_ERR(handle);
553
554 group = group_data[0].group;
555 for (i = 0; i < flex_gd->count; i++, group++) {
556 unsigned long gdblocks;
557 ext4_grpblk_t overhead;
558
559 gdblocks = ext4_bg_num_gdb(sb, group);
560 start = ext4_group_first_block_no(sb, group);
561
562 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
563 goto handle_itb;
564
565 if (meta_bg == 1)
566 goto handle_itb;
567
568 block = start + ext4_bg_has_super(sb, group);
569 /* Copy all of the GDT blocks into the backup in this group */
570 for (j = 0; j < gdblocks; j++, block++) {
571 struct buffer_head *gdb;
572
573 ext4_debug("update backup group %#04llx\n", block);
574 err = ext4_resize_ensure_credits_batch(handle, 1);
575 if (err < 0)
576 goto out;
577
578 gdb = sb_getblk(sb, block);
579 if (unlikely(!gdb)) {
580 err = -ENOMEM;
581 goto out;
582 }
583
584 BUFFER_TRACE(gdb, "get_write_access");
585 err = ext4_journal_get_write_access(handle, gdb);
586 if (err) {
587 brelse(gdb);
588 goto out;
589 }
590 memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
591 s_group_desc, j)->b_data, gdb->b_size);
592 set_buffer_uptodate(gdb);
593
594 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
595 if (unlikely(err)) {
596 brelse(gdb);
597 goto out;
598 }
599 brelse(gdb);
600 }
601
602 /* Zero out all of the reserved backup group descriptor
603 * table blocks
604 */
605 if (ext4_bg_has_super(sb, group)) {
606 err = sb_issue_zeroout(sb, gdblocks + start + 1,
607 reserved_gdb, GFP_NOFS);
608 if (err)
609 goto out;
610 }
611
612 handle_itb:
613 /* Initialize group tables of the grop @group */
614 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
615 goto handle_bb;
616
617 /* Zero out all of the inode table blocks */
618 block = group_data[i].inode_table;
619 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
620 block, sbi->s_itb_per_group);
621 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
622 GFP_NOFS);
623 if (err)
624 goto out;
625
626 handle_bb:
627 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
628 goto handle_ib;
629
630 /* Initialize block bitmap of the @group */
631 block = group_data[i].block_bitmap;
632 err = ext4_resize_ensure_credits_batch(handle, 1);
633 if (err < 0)
634 goto out;
635
636 bh = bclean(handle, sb, block);
637 if (IS_ERR(bh)) {
638 err = PTR_ERR(bh);
639 goto out;
640 }
641 overhead = ext4_group_overhead_blocks(sb, group);
642 if (overhead != 0) {
643 ext4_debug("mark backup superblock %#04llx (+0)\n",
644 start);
645 ext4_set_bits(bh->b_data, 0,
646 EXT4_NUM_B2C(sbi, overhead));
647 }
648 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
649 sb->s_blocksize * 8, bh->b_data);
650 err = ext4_handle_dirty_metadata(handle, NULL, bh);
651 brelse(bh);
652 if (err)
653 goto out;
654
655 handle_ib:
656 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
657 continue;
658
659 /* Initialize inode bitmap of the @group */
660 block = group_data[i].inode_bitmap;
661 err = ext4_resize_ensure_credits_batch(handle, 1);
662 if (err < 0)
663 goto out;
664 /* Mark unused entries in inode bitmap used */
665 bh = bclean(handle, sb, block);
666 if (IS_ERR(bh)) {
667 err = PTR_ERR(bh);
668 goto out;
669 }
670
671 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
672 sb->s_blocksize * 8, bh->b_data);
673 err = ext4_handle_dirty_metadata(handle, NULL, bh);
674 brelse(bh);
675 if (err)
676 goto out;
677 }
678
679 /* Mark group tables in block bitmap */
680 for (j = 0; j < GROUP_TABLE_COUNT; j++) {
681 count = group_table_count[j];
682 start = (&group_data[0].block_bitmap)[j];
683 block = start;
684 for (i = 1; i < flex_gd->count; i++) {
685 block += group_table_count[j];
686 if (block == (&group_data[i].block_bitmap)[j]) {
687 count += group_table_count[j];
688 continue;
689 }
690 err = set_flexbg_block_bitmap(sb, handle,
691 flex_gd,
692 EXT4_B2C(sbi, start),
693 EXT4_B2C(sbi,
694 start + count
695 - 1));
696 if (err)
697 goto out;
698 count = group_table_count[j];
699 start = (&group_data[i].block_bitmap)[j];
700 block = start;
701 }
702
703 if (count) {
704 err = set_flexbg_block_bitmap(sb, handle,
705 flex_gd,
706 EXT4_B2C(sbi, start),
707 EXT4_B2C(sbi,
708 start + count
709 - 1));
710 if (err)
711 goto out;
712 }
713 }
714
715 out:
716 err2 = ext4_journal_stop(handle);
717 if (err2 && !err)
718 err = err2;
719
720 return err;
721 }
722
723 /*
724 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
725 * ext4 filesystem. The counters should be initialized to 1, 5, and 7 before
726 * calling this for the first time. In a sparse filesystem it will be the
727 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
728 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
729 */
ext4_list_backups(struct super_block * sb,unsigned * three,unsigned * five,unsigned * seven)730 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
731 unsigned *five, unsigned *seven)
732 {
733 unsigned *min = three;
734 int mult = 3;
735 unsigned ret;
736
737 if (!ext4_has_feature_sparse_super(sb)) {
738 ret = *min;
739 *min += 1;
740 return ret;
741 }
742
743 if (*five < *min) {
744 min = five;
745 mult = 5;
746 }
747 if (*seven < *min) {
748 min = seven;
749 mult = 7;
750 }
751
752 ret = *min;
753 *min *= mult;
754
755 return ret;
756 }
757
758 /*
759 * Check that all of the backup GDT blocks are held in the primary GDT block.
760 * It is assumed that they are stored in group order. Returns the number of
761 * groups in current filesystem that have BACKUPS, or -ve error code.
762 */
verify_reserved_gdb(struct super_block * sb,ext4_group_t end,struct buffer_head * primary)763 static int verify_reserved_gdb(struct super_block *sb,
764 ext4_group_t end,
765 struct buffer_head *primary)
766 {
767 const ext4_fsblk_t blk = primary->b_blocknr;
768 unsigned three = 1;
769 unsigned five = 5;
770 unsigned seven = 7;
771 unsigned grp;
772 __le32 *p = (__le32 *)primary->b_data;
773 int gdbackups = 0;
774
775 while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
776 if (le32_to_cpu(*p++) !=
777 grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
778 ext4_warning(sb, "reserved GDT %llu"
779 " missing grp %d (%llu)",
780 blk, grp,
781 grp *
782 (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
783 blk);
784 return -EINVAL;
785 }
786 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
787 return -EFBIG;
788 }
789
790 return gdbackups;
791 }
792
793 /*
794 * Called when we need to bring a reserved group descriptor table block into
795 * use from the resize inode. The primary copy of the new GDT block currently
796 * is an indirect block (under the double indirect block in the resize inode).
797 * The new backup GDT blocks will be stored as leaf blocks in this indirect
798 * block, in group order. Even though we know all the block numbers we need,
799 * we check to ensure that the resize inode has actually reserved these blocks.
800 *
801 * Don't need to update the block bitmaps because the blocks are still in use.
802 *
803 * We get all of the error cases out of the way, so that we are sure to not
804 * fail once we start modifying the data on disk, because JBD has no rollback.
805 */
add_new_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)806 static int add_new_gdb(handle_t *handle, struct inode *inode,
807 ext4_group_t group)
808 {
809 struct super_block *sb = inode->i_sb;
810 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
811 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
812 ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
813 struct buffer_head **o_group_desc, **n_group_desc = NULL;
814 struct buffer_head *dind = NULL;
815 struct buffer_head *gdb_bh = NULL;
816 int gdbackups;
817 struct ext4_iloc iloc = { .bh = NULL };
818 __le32 *data;
819 int err;
820
821 if (test_opt(sb, DEBUG))
822 printk(KERN_DEBUG
823 "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
824 gdb_num);
825
826 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
827 if (IS_ERR(gdb_bh))
828 return PTR_ERR(gdb_bh);
829
830 gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
831 if (gdbackups < 0) {
832 err = gdbackups;
833 goto errout;
834 }
835
836 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
837 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
838 if (IS_ERR(dind)) {
839 err = PTR_ERR(dind);
840 dind = NULL;
841 goto errout;
842 }
843
844 data = (__le32 *)dind->b_data;
845 if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
846 ext4_warning(sb, "new group %u GDT block %llu not reserved",
847 group, gdblock);
848 err = -EINVAL;
849 goto errout;
850 }
851
852 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
853 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
854 if (unlikely(err))
855 goto errout;
856
857 BUFFER_TRACE(gdb_bh, "get_write_access");
858 err = ext4_journal_get_write_access(handle, gdb_bh);
859 if (unlikely(err))
860 goto errout;
861
862 BUFFER_TRACE(dind, "get_write_access");
863 err = ext4_journal_get_write_access(handle, dind);
864 if (unlikely(err)) {
865 ext4_std_error(sb, err);
866 goto errout;
867 }
868
869 /* ext4_reserve_inode_write() gets a reference on the iloc */
870 err = ext4_reserve_inode_write(handle, inode, &iloc);
871 if (unlikely(err))
872 goto errout;
873
874 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
875 GFP_KERNEL);
876 if (!n_group_desc) {
877 err = -ENOMEM;
878 ext4_warning(sb, "not enough memory for %lu groups",
879 gdb_num + 1);
880 goto errout;
881 }
882
883 /*
884 * Finally, we have all of the possible failures behind us...
885 *
886 * Remove new GDT block from inode double-indirect block and clear out
887 * the new GDT block for use (which also "frees" the backup GDT blocks
888 * from the reserved inode). We don't need to change the bitmaps for
889 * these blocks, because they are marked as in-use from being in the
890 * reserved inode, and will become GDT blocks (primary and backup).
891 */
892 data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
893 err = ext4_handle_dirty_metadata(handle, NULL, dind);
894 if (unlikely(err)) {
895 ext4_std_error(sb, err);
896 goto errout;
897 }
898 inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
899 (9 - EXT4_SB(sb)->s_cluster_bits);
900 ext4_mark_iloc_dirty(handle, inode, &iloc);
901 memset(gdb_bh->b_data, 0, sb->s_blocksize);
902 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
903 if (unlikely(err)) {
904 ext4_std_error(sb, err);
905 iloc.bh = NULL;
906 goto errout;
907 }
908 brelse(dind);
909
910 rcu_read_lock();
911 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
912 memcpy(n_group_desc, o_group_desc,
913 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
914 rcu_read_unlock();
915 n_group_desc[gdb_num] = gdb_bh;
916 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
917 EXT4_SB(sb)->s_gdb_count++;
918 ext4_kvfree_array_rcu(o_group_desc);
919
920 le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
921 err = ext4_handle_dirty_super(handle, sb);
922 if (err)
923 ext4_std_error(sb, err);
924 return err;
925 errout:
926 kvfree(n_group_desc);
927 brelse(iloc.bh);
928 brelse(dind);
929 brelse(gdb_bh);
930
931 ext4_debug("leaving with error %d\n", err);
932 return err;
933 }
934
935 /*
936 * add_new_gdb_meta_bg is the sister of add_new_gdb.
937 */
add_new_gdb_meta_bg(struct super_block * sb,handle_t * handle,ext4_group_t group)938 static int add_new_gdb_meta_bg(struct super_block *sb,
939 handle_t *handle, ext4_group_t group) {
940 ext4_fsblk_t gdblock;
941 struct buffer_head *gdb_bh;
942 struct buffer_head **o_group_desc, **n_group_desc;
943 unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
944 int err;
945
946 gdblock = ext4_meta_bg_first_block_no(sb, group) +
947 ext4_bg_has_super(sb, group);
948 gdb_bh = ext4_sb_bread(sb, gdblock, 0);
949 if (IS_ERR(gdb_bh))
950 return PTR_ERR(gdb_bh);
951 n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
952 GFP_KERNEL);
953 if (!n_group_desc) {
954 brelse(gdb_bh);
955 err = -ENOMEM;
956 ext4_warning(sb, "not enough memory for %lu groups",
957 gdb_num + 1);
958 return err;
959 }
960
961 rcu_read_lock();
962 o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
963 memcpy(n_group_desc, o_group_desc,
964 EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
965 rcu_read_unlock();
966 n_group_desc[gdb_num] = gdb_bh;
967
968 BUFFER_TRACE(gdb_bh, "get_write_access");
969 err = ext4_journal_get_write_access(handle, gdb_bh);
970 if (err) {
971 kvfree(n_group_desc);
972 brelse(gdb_bh);
973 return err;
974 }
975
976 rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
977 EXT4_SB(sb)->s_gdb_count++;
978 ext4_kvfree_array_rcu(o_group_desc);
979 return err;
980 }
981
982 /*
983 * Called when we are adding a new group which has a backup copy of each of
984 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
985 * We need to add these reserved backup GDT blocks to the resize inode, so
986 * that they are kept for future resizing and not allocated to files.
987 *
988 * Each reserved backup GDT block will go into a different indirect block.
989 * The indirect blocks are actually the primary reserved GDT blocks,
990 * so we know in advance what their block numbers are. We only get the
991 * double-indirect block to verify it is pointing to the primary reserved
992 * GDT blocks so we don't overwrite a data block by accident. The reserved
993 * backup GDT blocks are stored in their reserved primary GDT block.
994 */
reserve_backup_gdb(handle_t * handle,struct inode * inode,ext4_group_t group)995 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
996 ext4_group_t group)
997 {
998 struct super_block *sb = inode->i_sb;
999 int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1000 int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1001 struct buffer_head **primary;
1002 struct buffer_head *dind;
1003 struct ext4_iloc iloc;
1004 ext4_fsblk_t blk;
1005 __le32 *data, *end;
1006 int gdbackups = 0;
1007 int res, i;
1008 int err;
1009
1010 primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1011 if (!primary)
1012 return -ENOMEM;
1013
1014 data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1015 dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1016 if (IS_ERR(dind)) {
1017 err = PTR_ERR(dind);
1018 dind = NULL;
1019 goto exit_free;
1020 }
1021
1022 blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1023 data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1024 EXT4_ADDR_PER_BLOCK(sb));
1025 end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1026
1027 /* Get each reserved primary GDT block and verify it holds backups */
1028 for (res = 0; res < reserved_gdb; res++, blk++) {
1029 if (le32_to_cpu(*data) != blk) {
1030 ext4_warning(sb, "reserved block %llu"
1031 " not at offset %ld",
1032 blk,
1033 (long)(data - (__le32 *)dind->b_data));
1034 err = -EINVAL;
1035 goto exit_bh;
1036 }
1037 primary[res] = ext4_sb_bread(sb, blk, 0);
1038 if (IS_ERR(primary[res])) {
1039 err = PTR_ERR(primary[res]);
1040 primary[res] = NULL;
1041 goto exit_bh;
1042 }
1043 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1044 if (gdbackups < 0) {
1045 brelse(primary[res]);
1046 err = gdbackups;
1047 goto exit_bh;
1048 }
1049 if (++data >= end)
1050 data = (__le32 *)dind->b_data;
1051 }
1052
1053 for (i = 0; i < reserved_gdb; i++) {
1054 BUFFER_TRACE(primary[i], "get_write_access");
1055 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1056 goto exit_bh;
1057 }
1058
1059 if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1060 goto exit_bh;
1061
1062 /*
1063 * Finally we can add each of the reserved backup GDT blocks from
1064 * the new group to its reserved primary GDT block.
1065 */
1066 blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1067 for (i = 0; i < reserved_gdb; i++) {
1068 int err2;
1069 data = (__le32 *)primary[i]->b_data;
1070 /* printk("reserving backup %lu[%u] = %lu\n",
1071 primary[i]->b_blocknr, gdbackups,
1072 blk + primary[i]->b_blocknr); */
1073 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1074 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1075 if (!err)
1076 err = err2;
1077 }
1078
1079 inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1080 ext4_mark_iloc_dirty(handle, inode, &iloc);
1081
1082 exit_bh:
1083 while (--res >= 0)
1084 brelse(primary[res]);
1085 brelse(dind);
1086
1087 exit_free:
1088 kfree(primary);
1089
1090 return err;
1091 }
1092
1093 /*
1094 * Update the backup copies of the ext4 metadata. These don't need to be part
1095 * of the main resize transaction, because e2fsck will re-write them if there
1096 * is a problem (basically only OOM will cause a problem). However, we
1097 * _should_ update the backups if possible, in case the primary gets trashed
1098 * for some reason and we need to run e2fsck from a backup superblock. The
1099 * important part is that the new block and inode counts are in the backup
1100 * superblocks, and the location of the new group metadata in the GDT backups.
1101 *
1102 * We do not need take the s_resize_lock for this, because these
1103 * blocks are not otherwise touched by the filesystem code when it is
1104 * mounted. We don't need to worry about last changing from
1105 * sbi->s_groups_count, because the worst that can happen is that we
1106 * do not copy the full number of backups at this time. The resize
1107 * which changed s_groups_count will backup again.
1108 */
update_backups(struct super_block * sb,sector_t blk_off,char * data,int size,int meta_bg)1109 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1110 int size, int meta_bg)
1111 {
1112 struct ext4_sb_info *sbi = EXT4_SB(sb);
1113 ext4_group_t last;
1114 const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1115 unsigned three = 1;
1116 unsigned five = 5;
1117 unsigned seven = 7;
1118 ext4_group_t group = 0;
1119 int rest = sb->s_blocksize - size;
1120 handle_t *handle;
1121 int err = 0, err2;
1122
1123 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1124 if (IS_ERR(handle)) {
1125 group = 1;
1126 err = PTR_ERR(handle);
1127 goto exit_err;
1128 }
1129
1130 if (meta_bg == 0) {
1131 group = ext4_list_backups(sb, &three, &five, &seven);
1132 last = sbi->s_groups_count;
1133 } else {
1134 group = ext4_get_group_number(sb, blk_off) + 1;
1135 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1136 }
1137
1138 while (group < sbi->s_groups_count) {
1139 struct buffer_head *bh;
1140 ext4_fsblk_t backup_block;
1141
1142 /* Out of journal space, and can't get more - abort - so sad */
1143 err = ext4_resize_ensure_credits_batch(handle, 1);
1144 if (err < 0)
1145 break;
1146
1147 if (meta_bg == 0)
1148 backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1149 else
1150 backup_block = (ext4_group_first_block_no(sb, group) +
1151 ext4_bg_has_super(sb, group));
1152
1153 bh = sb_getblk(sb, backup_block);
1154 if (unlikely(!bh)) {
1155 err = -ENOMEM;
1156 break;
1157 }
1158 ext4_debug("update metadata backup %llu(+%llu)\n",
1159 backup_block, backup_block -
1160 ext4_group_first_block_no(sb, group));
1161 BUFFER_TRACE(bh, "get_write_access");
1162 if ((err = ext4_journal_get_write_access(handle, bh))) {
1163 brelse(bh);
1164 break;
1165 }
1166 lock_buffer(bh);
1167 memcpy(bh->b_data, data, size);
1168 if (rest)
1169 memset(bh->b_data + size, 0, rest);
1170 set_buffer_uptodate(bh);
1171 unlock_buffer(bh);
1172 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1173 if (unlikely(err))
1174 ext4_std_error(sb, err);
1175 brelse(bh);
1176
1177 if (meta_bg == 0)
1178 group = ext4_list_backups(sb, &three, &five, &seven);
1179 else if (group == last)
1180 break;
1181 else
1182 group = last;
1183 }
1184 if ((err2 = ext4_journal_stop(handle)) && !err)
1185 err = err2;
1186
1187 /*
1188 * Ugh! Need to have e2fsck write the backup copies. It is too
1189 * late to revert the resize, we shouldn't fail just because of
1190 * the backup copies (they are only needed in case of corruption).
1191 *
1192 * However, if we got here we have a journal problem too, so we
1193 * can't really start a transaction to mark the superblock.
1194 * Chicken out and just set the flag on the hope it will be written
1195 * to disk, and if not - we will simply wait until next fsck.
1196 */
1197 exit_err:
1198 if (err) {
1199 ext4_warning(sb, "can't update backup for group %u (err %d), "
1200 "forcing fsck on next reboot", group, err);
1201 sbi->s_mount_state &= ~EXT4_VALID_FS;
1202 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1203 mark_buffer_dirty(sbi->s_sbh);
1204 }
1205 }
1206
1207 /*
1208 * ext4_add_new_descs() adds @count group descriptor of groups
1209 * starting at @group
1210 *
1211 * @handle: journal handle
1212 * @sb: super block
1213 * @group: the group no. of the first group desc to be added
1214 * @resize_inode: the resize inode
1215 * @count: number of group descriptors to be added
1216 */
ext4_add_new_descs(handle_t * handle,struct super_block * sb,ext4_group_t group,struct inode * resize_inode,ext4_group_t count)1217 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1218 ext4_group_t group, struct inode *resize_inode,
1219 ext4_group_t count)
1220 {
1221 struct ext4_sb_info *sbi = EXT4_SB(sb);
1222 struct ext4_super_block *es = sbi->s_es;
1223 struct buffer_head *gdb_bh;
1224 int i, gdb_off, gdb_num, err = 0;
1225 int meta_bg;
1226
1227 meta_bg = ext4_has_feature_meta_bg(sb);
1228 for (i = 0; i < count; i++, group++) {
1229 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1230 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1231
1232 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1233 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1234
1235 /*
1236 * We will only either add reserved group blocks to a backup group
1237 * or remove reserved blocks for the first group in a new group block.
1238 * Doing both would be mean more complex code, and sane people don't
1239 * use non-sparse filesystems anymore. This is already checked above.
1240 */
1241 if (gdb_off) {
1242 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1243 gdb_num);
1244 BUFFER_TRACE(gdb_bh, "get_write_access");
1245 err = ext4_journal_get_write_access(handle, gdb_bh);
1246
1247 if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1248 err = reserve_backup_gdb(handle, resize_inode, group);
1249 } else if (meta_bg != 0) {
1250 err = add_new_gdb_meta_bg(sb, handle, group);
1251 } else {
1252 err = add_new_gdb(handle, resize_inode, group);
1253 }
1254 if (err)
1255 break;
1256 }
1257 return err;
1258 }
1259
ext4_get_bitmap(struct super_block * sb,__u64 block)1260 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1261 {
1262 struct buffer_head *bh = sb_getblk(sb, block);
1263 if (unlikely(!bh))
1264 return NULL;
1265 if (!bh_uptodate_or_lock(bh)) {
1266 if (ext4_read_bh(bh, 0, NULL) < 0) {
1267 brelse(bh);
1268 return NULL;
1269 }
1270 }
1271
1272 return bh;
1273 }
1274
ext4_set_bitmap_checksums(struct super_block * sb,ext4_group_t group,struct ext4_group_desc * gdp,struct ext4_new_group_data * group_data)1275 static int ext4_set_bitmap_checksums(struct super_block *sb,
1276 ext4_group_t group,
1277 struct ext4_group_desc *gdp,
1278 struct ext4_new_group_data *group_data)
1279 {
1280 struct buffer_head *bh;
1281
1282 if (!ext4_has_metadata_csum(sb))
1283 return 0;
1284
1285 bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1286 if (!bh)
1287 return -EIO;
1288 ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1289 EXT4_INODES_PER_GROUP(sb) / 8);
1290 brelse(bh);
1291
1292 bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1293 if (!bh)
1294 return -EIO;
1295 ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1296 brelse(bh);
1297
1298 return 0;
1299 }
1300
1301 /*
1302 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1303 */
ext4_setup_new_descs(handle_t * handle,struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1304 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1305 struct ext4_new_flex_group_data *flex_gd)
1306 {
1307 struct ext4_new_group_data *group_data = flex_gd->groups;
1308 struct ext4_group_desc *gdp;
1309 struct ext4_sb_info *sbi = EXT4_SB(sb);
1310 struct buffer_head *gdb_bh;
1311 ext4_group_t group;
1312 __u16 *bg_flags = flex_gd->bg_flags;
1313 int i, gdb_off, gdb_num, err = 0;
1314
1315
1316 for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1317 group = group_data->group;
1318
1319 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1320 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1321
1322 /*
1323 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1324 */
1325 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1326 /* Update group descriptor block for new group */
1327 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1328 gdb_off * EXT4_DESC_SIZE(sb));
1329
1330 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1331 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1332 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1333 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1334 if (err) {
1335 ext4_std_error(sb, err);
1336 break;
1337 }
1338
1339 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1340 ext4_free_group_clusters_set(sb, gdp,
1341 group_data->free_clusters_count);
1342 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1343 if (ext4_has_group_desc_csum(sb))
1344 ext4_itable_unused_set(sb, gdp,
1345 EXT4_INODES_PER_GROUP(sb));
1346 gdp->bg_flags = cpu_to_le16(*bg_flags);
1347 ext4_group_desc_csum_set(sb, group, gdp);
1348
1349 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1350 if (unlikely(err)) {
1351 ext4_std_error(sb, err);
1352 break;
1353 }
1354
1355 /*
1356 * We can allocate memory for mb_alloc based on the new group
1357 * descriptor
1358 */
1359 err = ext4_mb_add_groupinfo(sb, group, gdp);
1360 if (err)
1361 break;
1362 }
1363 return err;
1364 }
1365
1366 /*
1367 * ext4_update_super() updates the super block so that the newly added
1368 * groups can be seen by the filesystem.
1369 *
1370 * @sb: super block
1371 * @flex_gd: new added groups
1372 */
ext4_update_super(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd)1373 static void ext4_update_super(struct super_block *sb,
1374 struct ext4_new_flex_group_data *flex_gd)
1375 {
1376 ext4_fsblk_t blocks_count = 0;
1377 ext4_fsblk_t free_blocks = 0;
1378 ext4_fsblk_t reserved_blocks = 0;
1379 struct ext4_new_group_data *group_data = flex_gd->groups;
1380 struct ext4_sb_info *sbi = EXT4_SB(sb);
1381 struct ext4_super_block *es = sbi->s_es;
1382 int i;
1383
1384 BUG_ON(flex_gd->count == 0 || group_data == NULL);
1385 /*
1386 * Make the new blocks and inodes valid next. We do this before
1387 * increasing the group count so that once the group is enabled,
1388 * all of its blocks and inodes are already valid.
1389 *
1390 * We always allocate group-by-group, then block-by-block or
1391 * inode-by-inode within a group, so enabling these
1392 * blocks/inodes before the group is live won't actually let us
1393 * allocate the new space yet.
1394 */
1395 for (i = 0; i < flex_gd->count; i++) {
1396 blocks_count += group_data[i].blocks_count;
1397 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1398 }
1399
1400 reserved_blocks = ext4_r_blocks_count(es) * 100;
1401 reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1402 reserved_blocks *= blocks_count;
1403 do_div(reserved_blocks, 100);
1404
1405 ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1406 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1407 le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1408 flex_gd->count);
1409 le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1410 flex_gd->count);
1411
1412 ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1413 /*
1414 * We need to protect s_groups_count against other CPUs seeing
1415 * inconsistent state in the superblock.
1416 *
1417 * The precise rules we use are:
1418 *
1419 * * Writers must perform a smp_wmb() after updating all
1420 * dependent data and before modifying the groups count
1421 *
1422 * * Readers must perform an smp_rmb() after reading the groups
1423 * count and before reading any dependent data.
1424 *
1425 * NB. These rules can be relaxed when checking the group count
1426 * while freeing data, as we can only allocate from a block
1427 * group after serialising against the group count, and we can
1428 * only then free after serialising in turn against that
1429 * allocation.
1430 */
1431 smp_wmb();
1432
1433 /* Update the global fs size fields */
1434 sbi->s_groups_count += flex_gd->count;
1435 sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1436 (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1437
1438 /* Update the reserved block counts only once the new group is
1439 * active. */
1440 ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1441 reserved_blocks);
1442
1443 /* Update the free space counts */
1444 percpu_counter_add(&sbi->s_freeclusters_counter,
1445 EXT4_NUM_B2C(sbi, free_blocks));
1446 percpu_counter_add(&sbi->s_freeinodes_counter,
1447 EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1448
1449 ext4_debug("free blocks count %llu",
1450 percpu_counter_read(&sbi->s_freeclusters_counter));
1451 if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1452 ext4_group_t flex_group;
1453 struct flex_groups *fg;
1454
1455 flex_group = ext4_flex_group(sbi, group_data[0].group);
1456 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1457 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1458 &fg->free_clusters);
1459 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1460 &fg->free_inodes);
1461 }
1462
1463 /*
1464 * Update the fs overhead information
1465 */
1466 ext4_calculate_overhead(sb);
1467 es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1468
1469 if (test_opt(sb, DEBUG))
1470 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1471 "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1472 blocks_count, free_blocks, reserved_blocks);
1473 }
1474
1475 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1476 * _before_ we start modifying the filesystem, because we cannot abort the
1477 * transaction and not have it write the data to disk.
1478 */
ext4_flex_group_add(struct super_block * sb,struct inode * resize_inode,struct ext4_new_flex_group_data * flex_gd)1479 static int ext4_flex_group_add(struct super_block *sb,
1480 struct inode *resize_inode,
1481 struct ext4_new_flex_group_data *flex_gd)
1482 {
1483 struct ext4_sb_info *sbi = EXT4_SB(sb);
1484 struct ext4_super_block *es = sbi->s_es;
1485 ext4_fsblk_t o_blocks_count;
1486 ext4_grpblk_t last;
1487 ext4_group_t group;
1488 handle_t *handle;
1489 unsigned reserved_gdb;
1490 int err = 0, err2 = 0, credit;
1491
1492 BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1493
1494 reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1495 o_blocks_count = ext4_blocks_count(es);
1496 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1497 BUG_ON(last);
1498
1499 err = setup_new_flex_group_blocks(sb, flex_gd);
1500 if (err)
1501 goto exit;
1502 /*
1503 * We will always be modifying at least the superblock and GDT
1504 * blocks. If we are adding a group past the last current GDT block,
1505 * we will also modify the inode and the dindirect block. If we
1506 * are adding a group with superblock/GDT backups we will also
1507 * modify each of the reserved GDT dindirect blocks.
1508 */
1509 credit = 3; /* sb, resize inode, resize inode dindirect */
1510 /* GDT blocks */
1511 credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1512 credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1513 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1514 if (IS_ERR(handle)) {
1515 err = PTR_ERR(handle);
1516 goto exit;
1517 }
1518
1519 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1520 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1521 if (err)
1522 goto exit_journal;
1523
1524 group = flex_gd->groups[0].group;
1525 BUG_ON(group != sbi->s_groups_count);
1526 err = ext4_add_new_descs(handle, sb, group,
1527 resize_inode, flex_gd->count);
1528 if (err)
1529 goto exit_journal;
1530
1531 err = ext4_setup_new_descs(handle, sb, flex_gd);
1532 if (err)
1533 goto exit_journal;
1534
1535 ext4_update_super(sb, flex_gd);
1536
1537 err = ext4_handle_dirty_super(handle, sb);
1538
1539 exit_journal:
1540 err2 = ext4_journal_stop(handle);
1541 if (!err)
1542 err = err2;
1543
1544 if (!err) {
1545 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1546 int gdb_num_end = ((group + flex_gd->count - 1) /
1547 EXT4_DESC_PER_BLOCK(sb));
1548 int meta_bg = ext4_has_feature_meta_bg(sb);
1549 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1550 ext4_group_first_block_no(sb, 0);
1551 sector_t old_gdb = 0;
1552
1553 update_backups(sb, ext4_group_first_block_no(sb, 0),
1554 (char *)es, sizeof(struct ext4_super_block), 0);
1555 for (; gdb_num <= gdb_num_end; gdb_num++) {
1556 struct buffer_head *gdb_bh;
1557
1558 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1559 gdb_num);
1560 if (old_gdb == gdb_bh->b_blocknr)
1561 continue;
1562 update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1563 gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1564 old_gdb = gdb_bh->b_blocknr;
1565 }
1566 }
1567 exit:
1568 return err;
1569 }
1570
ext4_setup_next_flex_gd(struct super_block * sb,struct ext4_new_flex_group_data * flex_gd,ext4_fsblk_t n_blocks_count)1571 static int ext4_setup_next_flex_gd(struct super_block *sb,
1572 struct ext4_new_flex_group_data *flex_gd,
1573 ext4_fsblk_t n_blocks_count)
1574 {
1575 struct ext4_sb_info *sbi = EXT4_SB(sb);
1576 struct ext4_super_block *es = sbi->s_es;
1577 struct ext4_new_group_data *group_data = flex_gd->groups;
1578 ext4_fsblk_t o_blocks_count;
1579 ext4_group_t n_group;
1580 ext4_group_t group;
1581 ext4_group_t last_group;
1582 ext4_grpblk_t last;
1583 ext4_grpblk_t clusters_per_group;
1584 unsigned long i;
1585
1586 clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1587
1588 o_blocks_count = ext4_blocks_count(es);
1589
1590 if (o_blocks_count == n_blocks_count)
1591 return 0;
1592
1593 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1594 BUG_ON(last);
1595 ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1596
1597 last_group = group | (flex_gd->resize_bg - 1);
1598 if (last_group > n_group)
1599 last_group = n_group;
1600
1601 flex_gd->count = last_group - group + 1;
1602
1603 for (i = 0; i < flex_gd->count; i++) {
1604 int overhead;
1605
1606 group_data[i].group = group + i;
1607 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1608 overhead = ext4_group_overhead_blocks(sb, group + i);
1609 group_data[i].mdata_blocks = overhead;
1610 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1611 if (ext4_has_group_desc_csum(sb)) {
1612 flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1613 EXT4_BG_INODE_UNINIT;
1614 if (!test_opt(sb, INIT_INODE_TABLE))
1615 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1616 } else
1617 flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1618 }
1619
1620 if (last_group == n_group && ext4_has_group_desc_csum(sb))
1621 /* We need to initialize block bitmap of last group. */
1622 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1623
1624 if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1625 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1626 group_data[i - 1].free_clusters_count -= clusters_per_group -
1627 last - 1;
1628 }
1629
1630 return 1;
1631 }
1632
1633 /* Add group descriptor data to an existing or new group descriptor block.
1634 * Ensure we handle all possible error conditions _before_ we start modifying
1635 * the filesystem, because we cannot abort the transaction and not have it
1636 * write the data to disk.
1637 *
1638 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1639 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1640 *
1641 * We only need to hold the superblock lock while we are actually adding
1642 * in the new group's counts to the superblock. Prior to that we have
1643 * not really "added" the group at all. We re-check that we are still
1644 * adding in the last group in case things have changed since verifying.
1645 */
ext4_group_add(struct super_block * sb,struct ext4_new_group_data * input)1646 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1647 {
1648 struct ext4_new_flex_group_data flex_gd;
1649 struct ext4_sb_info *sbi = EXT4_SB(sb);
1650 struct ext4_super_block *es = sbi->s_es;
1651 int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1652 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1653 struct inode *inode = NULL;
1654 int gdb_off;
1655 int err;
1656 __u16 bg_flags = 0;
1657
1658 gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1659
1660 if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1661 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1662 return -EPERM;
1663 }
1664
1665 if (ext4_blocks_count(es) + input->blocks_count <
1666 ext4_blocks_count(es)) {
1667 ext4_warning(sb, "blocks_count overflow");
1668 return -EINVAL;
1669 }
1670
1671 if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1672 le32_to_cpu(es->s_inodes_count)) {
1673 ext4_warning(sb, "inodes_count overflow");
1674 return -EINVAL;
1675 }
1676
1677 if (reserved_gdb || gdb_off == 0) {
1678 if (!ext4_has_feature_resize_inode(sb) ||
1679 !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1680 ext4_warning(sb,
1681 "No reserved GDT blocks, can't resize");
1682 return -EPERM;
1683 }
1684 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1685 if (IS_ERR(inode)) {
1686 ext4_warning(sb, "Error opening resize inode");
1687 return PTR_ERR(inode);
1688 }
1689 }
1690
1691
1692 err = verify_group_input(sb, input);
1693 if (err)
1694 goto out;
1695
1696 err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1697 if (err)
1698 goto out;
1699
1700 err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1701 if (err)
1702 goto out;
1703
1704 flex_gd.count = 1;
1705 flex_gd.groups = input;
1706 flex_gd.bg_flags = &bg_flags;
1707 err = ext4_flex_group_add(sb, inode, &flex_gd);
1708 out:
1709 iput(inode);
1710 return err;
1711 } /* ext4_group_add */
1712
1713 /*
1714 * extend a group without checking assuming that checking has been done.
1715 */
ext4_group_extend_no_check(struct super_block * sb,ext4_fsblk_t o_blocks_count,ext4_grpblk_t add)1716 static int ext4_group_extend_no_check(struct super_block *sb,
1717 ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1718 {
1719 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1720 handle_t *handle;
1721 int err = 0, err2;
1722
1723 /* We will update the superblock, one block bitmap, and
1724 * one group descriptor via ext4_group_add_blocks().
1725 */
1726 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1727 if (IS_ERR(handle)) {
1728 err = PTR_ERR(handle);
1729 ext4_warning(sb, "error %d on journal start", err);
1730 return err;
1731 }
1732
1733 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1734 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1735 if (err) {
1736 ext4_warning(sb, "error %d on journal write access", err);
1737 goto errout;
1738 }
1739
1740 ext4_blocks_count_set(es, o_blocks_count + add);
1741 ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1742 ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1743 o_blocks_count + add);
1744 /* We add the blocks to the bitmap and set the group need init bit */
1745 err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1746 if (err)
1747 goto errout;
1748 ext4_handle_dirty_super(handle, sb);
1749 ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1750 o_blocks_count + add);
1751 errout:
1752 err2 = ext4_journal_stop(handle);
1753 if (err2 && !err)
1754 err = err2;
1755
1756 if (!err) {
1757 if (test_opt(sb, DEBUG))
1758 printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1759 "blocks\n", ext4_blocks_count(es));
1760 update_backups(sb, ext4_group_first_block_no(sb, 0),
1761 (char *)es, sizeof(struct ext4_super_block), 0);
1762 }
1763 return err;
1764 }
1765
1766 /*
1767 * Extend the filesystem to the new number of blocks specified. This entry
1768 * point is only used to extend the current filesystem to the end of the last
1769 * existing group. It can be accessed via ioctl, or by "remount,resize=<size>"
1770 * for emergencies (because it has no dependencies on reserved blocks).
1771 *
1772 * If we _really_ wanted, we could use default values to call ext4_group_add()
1773 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1774 * GDT blocks are reserved to grow to the desired size.
1775 */
ext4_group_extend(struct super_block * sb,struct ext4_super_block * es,ext4_fsblk_t n_blocks_count)1776 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1777 ext4_fsblk_t n_blocks_count)
1778 {
1779 ext4_fsblk_t o_blocks_count;
1780 ext4_grpblk_t last;
1781 ext4_grpblk_t add;
1782 struct buffer_head *bh;
1783 int err;
1784 ext4_group_t group;
1785
1786 o_blocks_count = ext4_blocks_count(es);
1787
1788 if (test_opt(sb, DEBUG))
1789 ext4_msg(sb, KERN_DEBUG,
1790 "extending last group from %llu to %llu blocks",
1791 o_blocks_count, n_blocks_count);
1792
1793 if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1794 return 0;
1795
1796 if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1797 ext4_msg(sb, KERN_ERR,
1798 "filesystem too large to resize to %llu blocks safely",
1799 n_blocks_count);
1800 return -EINVAL;
1801 }
1802
1803 if (n_blocks_count < o_blocks_count) {
1804 ext4_warning(sb, "can't shrink FS - resize aborted");
1805 return -EINVAL;
1806 }
1807
1808 /* Handle the remaining blocks in the last group only. */
1809 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1810
1811 if (last == 0) {
1812 ext4_warning(sb, "need to use ext2online to resize further");
1813 return -EPERM;
1814 }
1815
1816 add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1817
1818 if (o_blocks_count + add < o_blocks_count) {
1819 ext4_warning(sb, "blocks_count overflow");
1820 return -EINVAL;
1821 }
1822
1823 if (o_blocks_count + add > n_blocks_count)
1824 add = n_blocks_count - o_blocks_count;
1825
1826 if (o_blocks_count + add < n_blocks_count)
1827 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1828 o_blocks_count + add, add);
1829
1830 /* See if the device is actually as big as what was requested */
1831 bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1832 if (IS_ERR(bh)) {
1833 ext4_warning(sb, "can't read last block, resize aborted");
1834 return -ENOSPC;
1835 }
1836 brelse(bh);
1837
1838 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1839 return err;
1840 } /* ext4_group_extend */
1841
1842
num_desc_blocks(struct super_block * sb,ext4_group_t groups)1843 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1844 {
1845 return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1846 }
1847
1848 /*
1849 * Release the resize inode and drop the resize_inode feature if there
1850 * are no more reserved gdt blocks, and then convert the file system
1851 * to enable meta_bg
1852 */
ext4_convert_meta_bg(struct super_block * sb,struct inode * inode)1853 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1854 {
1855 handle_t *handle;
1856 struct ext4_sb_info *sbi = EXT4_SB(sb);
1857 struct ext4_super_block *es = sbi->s_es;
1858 struct ext4_inode_info *ei = EXT4_I(inode);
1859 ext4_fsblk_t nr;
1860 int i, ret, err = 0;
1861 int credits = 1;
1862
1863 ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1864 if (inode) {
1865 if (es->s_reserved_gdt_blocks) {
1866 ext4_error(sb, "Unexpected non-zero "
1867 "s_reserved_gdt_blocks");
1868 return -EPERM;
1869 }
1870
1871 /* Do a quick sanity check of the resize inode */
1872 if (inode->i_blocks != 1 << (inode->i_blkbits -
1873 (9 - sbi->s_cluster_bits)))
1874 goto invalid_resize_inode;
1875 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1876 if (i == EXT4_DIND_BLOCK) {
1877 if (ei->i_data[i])
1878 continue;
1879 else
1880 goto invalid_resize_inode;
1881 }
1882 if (ei->i_data[i])
1883 goto invalid_resize_inode;
1884 }
1885 credits += 3; /* block bitmap, bg descriptor, resize inode */
1886 }
1887
1888 handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1889 if (IS_ERR(handle))
1890 return PTR_ERR(handle);
1891
1892 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1893 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1894 if (err)
1895 goto errout;
1896
1897 ext4_clear_feature_resize_inode(sb);
1898 ext4_set_feature_meta_bg(sb);
1899 sbi->s_es->s_first_meta_bg =
1900 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1901
1902 err = ext4_handle_dirty_super(handle, sb);
1903 if (err) {
1904 ext4_std_error(sb, err);
1905 goto errout;
1906 }
1907
1908 if (inode) {
1909 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1910 ext4_free_blocks(handle, inode, NULL, nr, 1,
1911 EXT4_FREE_BLOCKS_METADATA |
1912 EXT4_FREE_BLOCKS_FORGET);
1913 ei->i_data[EXT4_DIND_BLOCK] = 0;
1914 inode->i_blocks = 0;
1915
1916 err = ext4_mark_inode_dirty(handle, inode);
1917 if (err)
1918 ext4_std_error(sb, err);
1919 }
1920
1921 errout:
1922 ret = ext4_journal_stop(handle);
1923 return err ? err : ret;
1924
1925 invalid_resize_inode:
1926 ext4_error(sb, "corrupted/inconsistent resize inode");
1927 return -EINVAL;
1928 }
1929
1930 /*
1931 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1932 *
1933 * @sb: super block of the fs to be resized
1934 * @n_blocks_count: the number of blocks resides in the resized fs
1935 */
ext4_resize_fs(struct super_block * sb,ext4_fsblk_t n_blocks_count)1936 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1937 {
1938 struct ext4_new_flex_group_data *flex_gd = NULL;
1939 struct ext4_sb_info *sbi = EXT4_SB(sb);
1940 struct ext4_super_block *es = sbi->s_es;
1941 struct buffer_head *bh;
1942 struct inode *resize_inode = NULL;
1943 ext4_grpblk_t add, offset;
1944 unsigned long n_desc_blocks;
1945 unsigned long o_desc_blocks;
1946 ext4_group_t o_group;
1947 ext4_group_t n_group;
1948 ext4_fsblk_t o_blocks_count;
1949 ext4_fsblk_t n_blocks_count_retry = 0;
1950 unsigned long last_update_time = 0;
1951 int err = 0;
1952 int meta_bg;
1953 unsigned int flexbg_size = ext4_flex_bg_size(sbi);
1954
1955 /* See if the device is actually as big as what was requested */
1956 bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1957 if (IS_ERR(bh)) {
1958 ext4_warning(sb, "can't read last block, resize aborted");
1959 return -ENOSPC;
1960 }
1961 brelse(bh);
1962
1963 /*
1964 * For bigalloc, trim the requested size to the nearest cluster
1965 * boundary to avoid creating an unusable filesystem. We do this
1966 * silently, instead of returning an error, to avoid breaking
1967 * callers that blindly resize the filesystem to the full size of
1968 * the underlying block device.
1969 */
1970 if (ext4_has_feature_bigalloc(sb))
1971 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1972
1973 retry:
1974 o_blocks_count = ext4_blocks_count(es);
1975
1976 ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1977 "to %llu blocks", o_blocks_count, n_blocks_count);
1978
1979 if (n_blocks_count < o_blocks_count) {
1980 /* On-line shrinking not supported */
1981 ext4_warning(sb, "can't shrink FS - resize aborted");
1982 return -EINVAL;
1983 }
1984
1985 if (n_blocks_count == o_blocks_count)
1986 /* Nothing need to do */
1987 return 0;
1988
1989 n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1990 if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1991 ext4_warning(sb, "resize would cause inodes_count overflow");
1992 return -EINVAL;
1993 }
1994 ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1995
1996 n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1997 o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1998
1999 meta_bg = ext4_has_feature_meta_bg(sb);
2000
2001 if (ext4_has_feature_resize_inode(sb)) {
2002 if (meta_bg) {
2003 ext4_error(sb, "resize_inode and meta_bg enabled "
2004 "simultaneously");
2005 return -EINVAL;
2006 }
2007 if (n_desc_blocks > o_desc_blocks +
2008 le16_to_cpu(es->s_reserved_gdt_blocks)) {
2009 n_blocks_count_retry = n_blocks_count;
2010 n_desc_blocks = o_desc_blocks +
2011 le16_to_cpu(es->s_reserved_gdt_blocks);
2012 n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2013 n_blocks_count = (ext4_fsblk_t)n_group *
2014 EXT4_BLOCKS_PER_GROUP(sb) +
2015 le32_to_cpu(es->s_first_data_block);
2016 n_group--; /* set to last group number */
2017 }
2018
2019 if (!resize_inode)
2020 resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2021 EXT4_IGET_SPECIAL);
2022 if (IS_ERR(resize_inode)) {
2023 ext4_warning(sb, "Error opening resize inode");
2024 return PTR_ERR(resize_inode);
2025 }
2026 }
2027
2028 if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2029 err = ext4_convert_meta_bg(sb, resize_inode);
2030 if (err)
2031 goto out;
2032 if (resize_inode) {
2033 iput(resize_inode);
2034 resize_inode = NULL;
2035 }
2036 if (n_blocks_count_retry) {
2037 n_blocks_count = n_blocks_count_retry;
2038 n_blocks_count_retry = 0;
2039 goto retry;
2040 }
2041 }
2042
2043 /*
2044 * Make sure the last group has enough space so that it's
2045 * guaranteed to have enough space for all metadata blocks
2046 * that it might need to hold. (We might not need to store
2047 * the inode table blocks in the last block group, but there
2048 * will be cases where this might be needed.)
2049 */
2050 if ((ext4_group_first_block_no(sb, n_group) +
2051 ext4_group_overhead_blocks(sb, n_group) + 2 +
2052 sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2053 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2054 n_group--;
2055 n_blocks_count_retry = 0;
2056 if (resize_inode) {
2057 iput(resize_inode);
2058 resize_inode = NULL;
2059 }
2060 goto retry;
2061 }
2062
2063 /* extend the last group */
2064 if (n_group == o_group)
2065 add = n_blocks_count - o_blocks_count;
2066 else
2067 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2068 if (add > 0) {
2069 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2070 if (err)
2071 goto out;
2072 }
2073
2074 if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2075 goto out;
2076
2077 err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2078 if (err)
2079 goto out;
2080
2081 err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2082 if (err)
2083 goto out;
2084
2085 flex_gd = alloc_flex_gd(flexbg_size);
2086 if (flex_gd == NULL) {
2087 err = -ENOMEM;
2088 goto out;
2089 }
2090
2091 /* Add flex groups. Note that a regular group is a
2092 * flex group with 1 group.
2093 */
2094 while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2095 if (jiffies - last_update_time > HZ * 10) {
2096 if (last_update_time)
2097 ext4_msg(sb, KERN_INFO,
2098 "resized to %llu blocks",
2099 ext4_blocks_count(es));
2100 last_update_time = jiffies;
2101 }
2102 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2103 break;
2104 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2105 if (unlikely(err))
2106 break;
2107 }
2108
2109 if (!err && n_blocks_count_retry) {
2110 n_blocks_count = n_blocks_count_retry;
2111 n_blocks_count_retry = 0;
2112 free_flex_gd(flex_gd);
2113 flex_gd = NULL;
2114 if (resize_inode) {
2115 iput(resize_inode);
2116 resize_inode = NULL;
2117 }
2118 goto retry;
2119 }
2120
2121 out:
2122 if (flex_gd)
2123 free_flex_gd(flex_gd);
2124 if (resize_inode != NULL)
2125 iput(resize_inode);
2126 if (err)
2127 ext4_warning(sb, "error (%d) occurred during "
2128 "file system resize", err);
2129 ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2130 ext4_blocks_count(es));
2131 return err;
2132 }
2133