GNU Linux-libre 6.9-gnu
[releases.git] / fs / ext4 / resize.c
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
19 struct ext4_rcu_ptr {
20         struct rcu_head rcu;
21         void *ptr;
22 };
23
24 static 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(ptr->ptr);
30         kfree(ptr);
31 }
32
33 void ext4_kvfree_array_rcu(void *to_free)
34 {
35         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
36
37         if (ptr) {
38                 ptr->ptr = to_free;
39                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
40                 return;
41         }
42         synchronize_rcu();
43         kvfree(to_free);
44 }
45
46 int 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                                   &sbi->s_ext4_flags))
93                 ret = -EBUSY;
94
95         return ret;
96 }
97
98 int ext4_resize_end(struct super_block *sb, bool update_backups)
99 {
100         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
101         smp_mb__after_atomic();
102         if (update_backups)
103                 return ext4_update_overhead(sb, true);
104         return 0;
105 }
106
107 static 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
120 static 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, start, NULL, &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(bh = ext4_sb_bread(sb, end - 1, 0))) {
164                 err = PTR_ERR(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  */
216 struct 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 resize_bg;                 /* number of allocated
222                                                    new_group_data */
223         ext4_group_t count;                     /* number of groups in @groups
224                                                  */
225 };
226
227 /*
228  * Avoiding memory allocation failures due to too many groups added each time.
229  */
230 #define MAX_RESIZE_BG                           16384
231
232 /*
233  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
234  * @flexbg_size.
235  *
236  * Returns NULL on failure otherwise address of the allocated structure.
237  */
238 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size,
239                                 ext4_group_t o_group, ext4_group_t n_group)
240 {
241         ext4_group_t last_group;
242         struct ext4_new_flex_group_data *flex_gd;
243
244         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
245         if (flex_gd == NULL)
246                 goto out3;
247
248         if (unlikely(flexbg_size > MAX_RESIZE_BG))
249                 flex_gd->resize_bg = MAX_RESIZE_BG;
250         else
251                 flex_gd->resize_bg = flexbg_size;
252
253         /* Avoid allocating large 'groups' array if not needed */
254         last_group = o_group | (flex_gd->resize_bg - 1);
255         if (n_group <= last_group)
256                 flex_gd->resize_bg = 1 << fls(n_group - o_group + 1);
257         else if (n_group - last_group < flex_gd->resize_bg)
258                 flex_gd->resize_bg = 1 << max(fls(last_group - o_group + 1),
259                                               fls(n_group - last_group));
260
261         flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
262                                         sizeof(struct ext4_new_group_data),
263                                         GFP_NOFS);
264         if (flex_gd->groups == NULL)
265                 goto out2;
266
267         flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
268                                           GFP_NOFS);
269         if (flex_gd->bg_flags == NULL)
270                 goto out1;
271
272         return flex_gd;
273
274 out1:
275         kfree(flex_gd->groups);
276 out2:
277         kfree(flex_gd);
278 out3:
279         return NULL;
280 }
281
282 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
283 {
284         kfree(flex_gd->bg_flags);
285         kfree(flex_gd->groups);
286         kfree(flex_gd);
287 }
288
289 /*
290  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
291  * and inode tables for a flex group.
292  *
293  * This function is used by 64bit-resize.  Note that this function allocates
294  * group tables from the 1st group of groups contained by @flexgd, which may
295  * be a partial of a flex group.
296  *
297  * @sb: super block of fs to which the groups belongs
298  *
299  * Returns 0 on a successful allocation of the metadata blocks in the
300  * block group.
301  */
302 static int ext4_alloc_group_tables(struct super_block *sb,
303                                 struct ext4_new_flex_group_data *flex_gd,
304                                 unsigned int flexbg_size)
305 {
306         struct ext4_new_group_data *group_data = flex_gd->groups;
307         ext4_fsblk_t start_blk;
308         ext4_fsblk_t last_blk;
309         ext4_group_t src_group;
310         ext4_group_t bb_index = 0;
311         ext4_group_t ib_index = 0;
312         ext4_group_t it_index = 0;
313         ext4_group_t group;
314         ext4_group_t last_group;
315         unsigned overhead;
316         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
317         int i;
318
319         BUG_ON(flex_gd->count == 0 || group_data == NULL);
320
321         src_group = group_data[0].group;
322         last_group  = src_group + flex_gd->count - 1;
323
324         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
325                (last_group & ~(flexbg_size - 1))));
326 next_group:
327         group = group_data[0].group;
328         if (src_group >= group_data[0].group + flex_gd->count)
329                 return -ENOSPC;
330         start_blk = ext4_group_first_block_no(sb, src_group);
331         last_blk = start_blk + group_data[src_group - group].blocks_count;
332
333         overhead = ext4_group_overhead_blocks(sb, src_group);
334
335         start_blk += overhead;
336
337         /* We collect contiguous blocks as much as possible. */
338         src_group++;
339         for (; src_group <= last_group; src_group++) {
340                 overhead = ext4_group_overhead_blocks(sb, src_group);
341                 if (overhead == 0)
342                         last_blk += group_data[src_group - group].blocks_count;
343                 else
344                         break;
345         }
346
347         /* Allocate block bitmaps */
348         for (; bb_index < flex_gd->count; bb_index++) {
349                 if (start_blk >= last_blk)
350                         goto next_group;
351                 group_data[bb_index].block_bitmap = start_blk++;
352                 group = ext4_get_group_number(sb, start_blk - 1);
353                 group -= group_data[0].group;
354                 group_data[group].mdata_blocks++;
355                 flex_gd->bg_flags[group] &= uninit_mask;
356         }
357
358         /* Allocate inode bitmaps */
359         for (; ib_index < flex_gd->count; ib_index++) {
360                 if (start_blk >= last_blk)
361                         goto next_group;
362                 group_data[ib_index].inode_bitmap = start_blk++;
363                 group = ext4_get_group_number(sb, start_blk - 1);
364                 group -= group_data[0].group;
365                 group_data[group].mdata_blocks++;
366                 flex_gd->bg_flags[group] &= uninit_mask;
367         }
368
369         /* Allocate inode tables */
370         for (; it_index < flex_gd->count; it_index++) {
371                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
372                 ext4_fsblk_t next_group_start;
373
374                 if (start_blk + itb > last_blk)
375                         goto next_group;
376                 group_data[it_index].inode_table = start_blk;
377                 group = ext4_get_group_number(sb, start_blk);
378                 next_group_start = ext4_group_first_block_no(sb, group + 1);
379                 group -= group_data[0].group;
380
381                 if (start_blk + itb > next_group_start) {
382                         flex_gd->bg_flags[group + 1] &= uninit_mask;
383                         overhead = start_blk + itb - next_group_start;
384                         group_data[group + 1].mdata_blocks += overhead;
385                         itb -= overhead;
386                 }
387
388                 group_data[group].mdata_blocks += itb;
389                 flex_gd->bg_flags[group] &= uninit_mask;
390                 start_blk += EXT4_SB(sb)->s_itb_per_group;
391         }
392
393         /* Update free clusters count to exclude metadata blocks */
394         for (i = 0; i < flex_gd->count; i++) {
395                 group_data[i].free_clusters_count -=
396                                 EXT4_NUM_B2C(EXT4_SB(sb),
397                                              group_data[i].mdata_blocks);
398         }
399
400         if (test_opt(sb, DEBUG)) {
401                 int i;
402                 group = group_data[0].group;
403
404                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
405                        "%u groups, flexbg size is %u:\n", flex_gd->count,
406                        flexbg_size);
407
408                 for (i = 0; i < flex_gd->count; i++) {
409                         ext4_debug(
410                                "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
411                                ext4_bg_has_super(sb, group + i) ? "normal" :
412                                "no-super", group + i,
413                                group_data[i].blocks_count,
414                                group_data[i].free_clusters_count,
415                                group_data[i].mdata_blocks);
416                 }
417         }
418         return 0;
419 }
420
421 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
422                                   ext4_fsblk_t blk)
423 {
424         struct buffer_head *bh;
425         int err;
426
427         bh = sb_getblk(sb, blk);
428         if (unlikely(!bh))
429                 return ERR_PTR(-ENOMEM);
430         BUFFER_TRACE(bh, "get_write_access");
431         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
432         if (err) {
433                 brelse(bh);
434                 bh = ERR_PTR(err);
435         } else {
436                 memset(bh->b_data, 0, sb->s_blocksize);
437                 set_buffer_uptodate(bh);
438         }
439
440         return bh;
441 }
442
443 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
444 {
445         return ext4_journal_ensure_credits_fn(handle, credits,
446                 EXT4_MAX_TRANS_DATA, 0, 0);
447 }
448
449 /*
450  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
451  *
452  * Helper function for ext4_setup_new_group_blocks() which set .
453  *
454  * @sb: super block
455  * @handle: journal handle
456  * @flex_gd: flex group data
457  */
458 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
459                         struct ext4_new_flex_group_data *flex_gd,
460                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
461 {
462         struct ext4_sb_info *sbi = EXT4_SB(sb);
463         ext4_group_t count = last_cluster - first_cluster + 1;
464         ext4_group_t count2;
465
466         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
467                    last_cluster);
468         for (; count > 0; count -= count2, first_cluster += count2) {
469                 ext4_fsblk_t start;
470                 struct buffer_head *bh;
471                 ext4_group_t group;
472                 int err;
473
474                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
475                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
476                 group -= flex_gd->groups[0].group;
477
478                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
479                 if (count2 > count)
480                         count2 = count;
481
482                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
483                         BUG_ON(flex_gd->count > 1);
484                         continue;
485                 }
486
487                 err = ext4_resize_ensure_credits_batch(handle, 1);
488                 if (err < 0)
489                         return err;
490
491                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
492                 if (unlikely(!bh))
493                         return -ENOMEM;
494
495                 BUFFER_TRACE(bh, "get_write_access");
496                 err = ext4_journal_get_write_access(handle, sb, bh,
497                                                     EXT4_JTR_NONE);
498                 if (err) {
499                         brelse(bh);
500                         return err;
501                 }
502                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
503                            first_cluster, first_cluster - start, count2);
504                 mb_set_bits(bh->b_data, first_cluster - start, count2);
505
506                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
507                 brelse(bh);
508                 if (unlikely(err))
509                         return err;
510         }
511
512         return 0;
513 }
514
515 /*
516  * Set up the block and inode bitmaps, and the inode table for the new groups.
517  * This doesn't need to be part of the main transaction, since we are only
518  * changing blocks outside the actual filesystem.  We still do journaling to
519  * ensure the recovery is correct in case of a failure just after resize.
520  * If any part of this fails, we simply abort the resize.
521  *
522  * setup_new_flex_group_blocks handles a flex group as follow:
523  *  1. copy super block and GDT, and initialize group tables if necessary.
524  *     In this step, we only set bits in blocks bitmaps for blocks taken by
525  *     super block and GDT.
526  *  2. allocate group tables in block bitmaps, that is, set bits in block
527  *     bitmap for blocks taken by group tables.
528  */
529 static int setup_new_flex_group_blocks(struct super_block *sb,
530                                 struct ext4_new_flex_group_data *flex_gd)
531 {
532         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
533         ext4_fsblk_t start;
534         ext4_fsblk_t block;
535         struct ext4_sb_info *sbi = EXT4_SB(sb);
536         struct ext4_super_block *es = sbi->s_es;
537         struct ext4_new_group_data *group_data = flex_gd->groups;
538         __u16 *bg_flags = flex_gd->bg_flags;
539         handle_t *handle;
540         ext4_group_t group, count;
541         struct buffer_head *bh = NULL;
542         int reserved_gdb, i, j, err = 0, err2;
543         int meta_bg;
544
545         BUG_ON(!flex_gd->count || !group_data ||
546                group_data[0].group != sbi->s_groups_count);
547
548         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
549         meta_bg = ext4_has_feature_meta_bg(sb);
550
551         /* This transaction may be extended/restarted along the way */
552         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
553         if (IS_ERR(handle))
554                 return PTR_ERR(handle);
555
556         group = group_data[0].group;
557         for (i = 0; i < flex_gd->count; i++, group++) {
558                 unsigned long gdblocks;
559                 ext4_grpblk_t overhead;
560
561                 gdblocks = ext4_bg_num_gdb(sb, group);
562                 start = ext4_group_first_block_no(sb, group);
563
564                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
565                         goto handle_itb;
566
567                 if (meta_bg == 1)
568                         goto handle_itb;
569
570                 block = start + ext4_bg_has_super(sb, group);
571                 /* Copy all of the GDT blocks into the backup in this group */
572                 for (j = 0; j < gdblocks; j++, block++) {
573                         struct buffer_head *gdb;
574
575                         ext4_debug("update backup group %#04llx\n", block);
576                         err = ext4_resize_ensure_credits_batch(handle, 1);
577                         if (err < 0)
578                                 goto out;
579
580                         gdb = sb_getblk(sb, block);
581                         if (unlikely(!gdb)) {
582                                 err = -ENOMEM;
583                                 goto out;
584                         }
585
586                         BUFFER_TRACE(gdb, "get_write_access");
587                         err = ext4_journal_get_write_access(handle, sb, gdb,
588                                                             EXT4_JTR_NONE);
589                         if (err) {
590                                 brelse(gdb);
591                                 goto out;
592                         }
593                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
594                                 s_group_desc, j)->b_data, gdb->b_size);
595                         set_buffer_uptodate(gdb);
596
597                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
598                         if (unlikely(err)) {
599                                 brelse(gdb);
600                                 goto out;
601                         }
602                         brelse(gdb);
603                 }
604
605                 /* Zero out all of the reserved backup group descriptor
606                  * table blocks
607                  */
608                 if (ext4_bg_has_super(sb, group)) {
609                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
610                                         reserved_gdb, GFP_NOFS);
611                         if (err)
612                                 goto out;
613                 }
614
615 handle_itb:
616                 /* Initialize group tables of the group @group */
617                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
618                         goto handle_bb;
619
620                 /* Zero out all of the inode table blocks */
621                 block = group_data[i].inode_table;
622                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
623                            block, sbi->s_itb_per_group);
624                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
625                                        GFP_NOFS);
626                 if (err)
627                         goto out;
628
629 handle_bb:
630                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
631                         goto handle_ib;
632
633                 /* Initialize block bitmap of the @group */
634                 block = group_data[i].block_bitmap;
635                 err = ext4_resize_ensure_credits_batch(handle, 1);
636                 if (err < 0)
637                         goto out;
638
639                 bh = bclean(handle, sb, block);
640                 if (IS_ERR(bh)) {
641                         err = PTR_ERR(bh);
642                         goto out;
643                 }
644                 overhead = ext4_group_overhead_blocks(sb, group);
645                 if (overhead != 0) {
646                         ext4_debug("mark backup superblock %#04llx (+0)\n",
647                                    start);
648                         mb_set_bits(bh->b_data, 0,
649                                       EXT4_NUM_B2C(sbi, overhead));
650                 }
651                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
652                                      sb->s_blocksize * 8, bh->b_data);
653                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
654                 brelse(bh);
655                 if (err)
656                         goto out;
657
658 handle_ib:
659                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
660                         continue;
661
662                 /* Initialize inode bitmap of the @group */
663                 block = group_data[i].inode_bitmap;
664                 err = ext4_resize_ensure_credits_batch(handle, 1);
665                 if (err < 0)
666                         goto out;
667                 /* Mark unused entries in inode bitmap used */
668                 bh = bclean(handle, sb, block);
669                 if (IS_ERR(bh)) {
670                         err = PTR_ERR(bh);
671                         goto out;
672                 }
673
674                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
675                                      sb->s_blocksize * 8, bh->b_data);
676                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
677                 brelse(bh);
678                 if (err)
679                         goto out;
680         }
681
682         /* Mark group tables in block bitmap */
683         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
684                 count = group_table_count[j];
685                 start = (&group_data[0].block_bitmap)[j];
686                 block = start;
687                 for (i = 1; i < flex_gd->count; i++) {
688                         block += group_table_count[j];
689                         if (block == (&group_data[i].block_bitmap)[j]) {
690                                 count += group_table_count[j];
691                                 continue;
692                         }
693                         err = set_flexbg_block_bitmap(sb, handle,
694                                                       flex_gd,
695                                                       EXT4_B2C(sbi, start),
696                                                       EXT4_B2C(sbi,
697                                                                start + count
698                                                                - 1));
699                         if (err)
700                                 goto out;
701                         count = group_table_count[j];
702                         start = (&group_data[i].block_bitmap)[j];
703                         block = start;
704                 }
705
706                 err = set_flexbg_block_bitmap(sb, handle,
707                                 flex_gd,
708                                 EXT4_B2C(sbi, start),
709                                 EXT4_B2C(sbi,
710                                         start + count
711                                         - 1));
712                 if (err)
713                         goto out;
714         }
715
716 out:
717         err2 = ext4_journal_stop(handle);
718         if (err2 && !err)
719                 err = err2;
720
721         return err;
722 }
723
724 /*
725  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
726  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
727  * calling this for the first time.  In a sparse filesystem it will be the
728  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
729  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
730  */
731 unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
732                                unsigned int *five, unsigned int *seven)
733 {
734         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
735         unsigned int *min = three;
736         int mult = 3;
737         unsigned int ret;
738
739         if (ext4_has_feature_sparse_super2(sb)) {
740                 do {
741                         if (*min > 2)
742                                 return UINT_MAX;
743                         ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
744                         *min += 1;
745                 } while (!ret);
746                 return ret;
747         }
748
749         if (!ext4_has_feature_sparse_super(sb)) {
750                 ret = *min;
751                 *min += 1;
752                 return ret;
753         }
754
755         if (*five < *min) {
756                 min = five;
757                 mult = 5;
758         }
759         if (*seven < *min) {
760                 min = seven;
761                 mult = 7;
762         }
763
764         ret = *min;
765         *min *= mult;
766
767         return ret;
768 }
769
770 /*
771  * Check that all of the backup GDT blocks are held in the primary GDT block.
772  * It is assumed that they are stored in group order.  Returns the number of
773  * groups in current filesystem that have BACKUPS, or -ve error code.
774  */
775 static int verify_reserved_gdb(struct super_block *sb,
776                                ext4_group_t end,
777                                struct buffer_head *primary)
778 {
779         const ext4_fsblk_t blk = primary->b_blocknr;
780         unsigned three = 1;
781         unsigned five = 5;
782         unsigned seven = 7;
783         unsigned grp;
784         __le32 *p = (__le32 *)primary->b_data;
785         int gdbackups = 0;
786
787         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
788                 if (le32_to_cpu(*p++) !=
789                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
790                         ext4_warning(sb, "reserved GDT %llu"
791                                      " missing grp %d (%llu)",
792                                      blk, grp,
793                                      grp *
794                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
795                                      blk);
796                         return -EINVAL;
797                 }
798                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
799                         return -EFBIG;
800         }
801
802         return gdbackups;
803 }
804
805 /*
806  * Called when we need to bring a reserved group descriptor table block into
807  * use from the resize inode.  The primary copy of the new GDT block currently
808  * is an indirect block (under the double indirect block in the resize inode).
809  * The new backup GDT blocks will be stored as leaf blocks in this indirect
810  * block, in group order.  Even though we know all the block numbers we need,
811  * we check to ensure that the resize inode has actually reserved these blocks.
812  *
813  * Don't need to update the block bitmaps because the blocks are still in use.
814  *
815  * We get all of the error cases out of the way, so that we are sure to not
816  * fail once we start modifying the data on disk, because JBD has no rollback.
817  */
818 static int add_new_gdb(handle_t *handle, struct inode *inode,
819                        ext4_group_t group)
820 {
821         struct super_block *sb = inode->i_sb;
822         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
823         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
824         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
825         struct buffer_head **o_group_desc, **n_group_desc = NULL;
826         struct buffer_head *dind = NULL;
827         struct buffer_head *gdb_bh = NULL;
828         int gdbackups;
829         struct ext4_iloc iloc = { .bh = NULL };
830         __le32 *data;
831         int err;
832
833         if (test_opt(sb, DEBUG))
834                 printk(KERN_DEBUG
835                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
836                        gdb_num);
837
838         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
839         if (IS_ERR(gdb_bh))
840                 return PTR_ERR(gdb_bh);
841
842         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
843         if (gdbackups < 0) {
844                 err = gdbackups;
845                 goto errout;
846         }
847
848         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
849         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
850         if (IS_ERR(dind)) {
851                 err = PTR_ERR(dind);
852                 dind = NULL;
853                 goto errout;
854         }
855
856         data = (__le32 *)dind->b_data;
857         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
858                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
859                              group, gdblock);
860                 err = -EINVAL;
861                 goto errout;
862         }
863
864         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
865         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
866                                             EXT4_JTR_NONE);
867         if (unlikely(err))
868                 goto errout;
869
870         BUFFER_TRACE(gdb_bh, "get_write_access");
871         err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
872         if (unlikely(err))
873                 goto errout;
874
875         BUFFER_TRACE(dind, "get_write_access");
876         err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
877         if (unlikely(err)) {
878                 ext4_std_error(sb, err);
879                 goto errout;
880         }
881
882         /* ext4_reserve_inode_write() gets a reference on the iloc */
883         err = ext4_reserve_inode_write(handle, inode, &iloc);
884         if (unlikely(err))
885                 goto errout;
886
887         n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
888                                 GFP_KERNEL);
889         if (!n_group_desc) {
890                 err = -ENOMEM;
891                 ext4_warning(sb, "not enough memory for %lu groups",
892                              gdb_num + 1);
893                 goto errout;
894         }
895
896         /*
897          * Finally, we have all of the possible failures behind us...
898          *
899          * Remove new GDT block from inode double-indirect block and clear out
900          * the new GDT block for use (which also "frees" the backup GDT blocks
901          * from the reserved inode).  We don't need to change the bitmaps for
902          * these blocks, because they are marked as in-use from being in the
903          * reserved inode, and will become GDT blocks (primary and backup).
904          */
905         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
906         err = ext4_handle_dirty_metadata(handle, NULL, dind);
907         if (unlikely(err)) {
908                 ext4_std_error(sb, err);
909                 goto errout;
910         }
911         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
912                            (9 - EXT4_SB(sb)->s_cluster_bits);
913         ext4_mark_iloc_dirty(handle, inode, &iloc);
914         memset(gdb_bh->b_data, 0, sb->s_blocksize);
915         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
916         if (unlikely(err)) {
917                 ext4_std_error(sb, err);
918                 iloc.bh = NULL;
919                 goto errout;
920         }
921         brelse(dind);
922
923         rcu_read_lock();
924         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
925         memcpy(n_group_desc, o_group_desc,
926                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
927         rcu_read_unlock();
928         n_group_desc[gdb_num] = gdb_bh;
929         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
930         EXT4_SB(sb)->s_gdb_count++;
931         ext4_kvfree_array_rcu(o_group_desc);
932
933         lock_buffer(EXT4_SB(sb)->s_sbh);
934         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
935         ext4_superblock_csum_set(sb);
936         unlock_buffer(EXT4_SB(sb)->s_sbh);
937         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
938         if (err)
939                 ext4_std_error(sb, err);
940         return err;
941 errout:
942         kvfree(n_group_desc);
943         brelse(iloc.bh);
944         brelse(dind);
945         brelse(gdb_bh);
946
947         ext4_debug("leaving with error %d\n", err);
948         return err;
949 }
950
951 /*
952  * If there is no available space in the existing block group descriptors for
953  * the new block group and there are no reserved block group descriptors, then
954  * the meta_bg feature will get enabled, and es->s_first_meta_bg will get set
955  * to the first block group that is managed using meta_bg and s_first_meta_bg
956  * must be a multiple of EXT4_DESC_PER_BLOCK(sb).
957  * This function will be called when first group of meta_bg is added to bring
958  * new group descriptors block of new added meta_bg.
959  */
960 static int add_new_gdb_meta_bg(struct super_block *sb,
961                                handle_t *handle, ext4_group_t group) {
962         ext4_fsblk_t gdblock;
963         struct buffer_head *gdb_bh;
964         struct buffer_head **o_group_desc, **n_group_desc;
965         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
966         int err;
967
968         gdblock = ext4_group_first_block_no(sb, group) +
969                   ext4_bg_has_super(sb, group);
970         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
971         if (IS_ERR(gdb_bh))
972                 return PTR_ERR(gdb_bh);
973         n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
974                                 GFP_KERNEL);
975         if (!n_group_desc) {
976                 brelse(gdb_bh);
977                 err = -ENOMEM;
978                 ext4_warning(sb, "not enough memory for %lu groups",
979                              gdb_num + 1);
980                 return err;
981         }
982
983         rcu_read_lock();
984         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
985         memcpy(n_group_desc, o_group_desc,
986                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
987         rcu_read_unlock();
988         n_group_desc[gdb_num] = gdb_bh;
989
990         BUFFER_TRACE(gdb_bh, "get_write_access");
991         err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
992         if (err) {
993                 kvfree(n_group_desc);
994                 brelse(gdb_bh);
995                 return err;
996         }
997
998         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
999         EXT4_SB(sb)->s_gdb_count++;
1000         ext4_kvfree_array_rcu(o_group_desc);
1001         return err;
1002 }
1003
1004 /*
1005  * Called when we are adding a new group which has a backup copy of each of
1006  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
1007  * We need to add these reserved backup GDT blocks to the resize inode, so
1008  * that they are kept for future resizing and not allocated to files.
1009  *
1010  * Each reserved backup GDT block will go into a different indirect block.
1011  * The indirect blocks are actually the primary reserved GDT blocks,
1012  * so we know in advance what their block numbers are.  We only get the
1013  * double-indirect block to verify it is pointing to the primary reserved
1014  * GDT blocks so we don't overwrite a data block by accident.  The reserved
1015  * backup GDT blocks are stored in their reserved primary GDT block.
1016  */
1017 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1018                               ext4_group_t group)
1019 {
1020         struct super_block *sb = inode->i_sb;
1021         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1022         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1023         struct buffer_head **primary;
1024         struct buffer_head *dind;
1025         struct ext4_iloc iloc;
1026         ext4_fsblk_t blk;
1027         __le32 *data, *end;
1028         int gdbackups = 0;
1029         int res, i;
1030         int err;
1031
1032         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1033         if (!primary)
1034                 return -ENOMEM;
1035
1036         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1037         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1038         if (IS_ERR(dind)) {
1039                 err = PTR_ERR(dind);
1040                 dind = NULL;
1041                 goto exit_free;
1042         }
1043
1044         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1045         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1046                                          EXT4_ADDR_PER_BLOCK(sb));
1047         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1048
1049         /* Get each reserved primary GDT block and verify it holds backups */
1050         for (res = 0; res < reserved_gdb; res++, blk++) {
1051                 if (le32_to_cpu(*data) != blk) {
1052                         ext4_warning(sb, "reserved block %llu"
1053                                      " not at offset %ld",
1054                                      blk,
1055                                      (long)(data - (__le32 *)dind->b_data));
1056                         err = -EINVAL;
1057                         goto exit_bh;
1058                 }
1059                 primary[res] = ext4_sb_bread(sb, blk, 0);
1060                 if (IS_ERR(primary[res])) {
1061                         err = PTR_ERR(primary[res]);
1062                         primary[res] = NULL;
1063                         goto exit_bh;
1064                 }
1065                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1066                 if (gdbackups < 0) {
1067                         brelse(primary[res]);
1068                         err = gdbackups;
1069                         goto exit_bh;
1070                 }
1071                 if (++data >= end)
1072                         data = (__le32 *)dind->b_data;
1073         }
1074
1075         for (i = 0; i < reserved_gdb; i++) {
1076                 BUFFER_TRACE(primary[i], "get_write_access");
1077                 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1078                                                          EXT4_JTR_NONE)))
1079                         goto exit_bh;
1080         }
1081
1082         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1083                 goto exit_bh;
1084
1085         /*
1086          * Finally we can add each of the reserved backup GDT blocks from
1087          * the new group to its reserved primary GDT block.
1088          */
1089         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1090         for (i = 0; i < reserved_gdb; i++) {
1091                 int err2;
1092                 data = (__le32 *)primary[i]->b_data;
1093                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1094                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1095                 if (!err)
1096                         err = err2;
1097         }
1098
1099         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1100         ext4_mark_iloc_dirty(handle, inode, &iloc);
1101
1102 exit_bh:
1103         while (--res >= 0)
1104                 brelse(primary[res]);
1105         brelse(dind);
1106
1107 exit_free:
1108         kfree(primary);
1109
1110         return err;
1111 }
1112
1113 static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1114                                            ext4_group_t group)
1115 {
1116         struct ext4_super_block *es = (struct ext4_super_block *) data;
1117
1118         es->s_block_group_nr = cpu_to_le16(group);
1119         if (ext4_has_metadata_csum(sb))
1120                 es->s_checksum = ext4_superblock_csum(sb, es);
1121 }
1122
1123 /*
1124  * Update the backup copies of the ext4 metadata.  These don't need to be part
1125  * of the main resize transaction, because e2fsck will re-write them if there
1126  * is a problem (basically only OOM will cause a problem).  However, we
1127  * _should_ update the backups if possible, in case the primary gets trashed
1128  * for some reason and we need to run e2fsck from a backup superblock.  The
1129  * important part is that the new block and inode counts are in the backup
1130  * superblocks, and the location of the new group metadata in the GDT backups.
1131  *
1132  * We do not need take the s_resize_lock for this, because these
1133  * blocks are not otherwise touched by the filesystem code when it is
1134  * mounted.  We don't need to worry about last changing from
1135  * sbi->s_groups_count, because the worst that can happen is that we
1136  * do not copy the full number of backups at this time.  The resize
1137  * which changed s_groups_count will backup again.
1138  */
1139 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1140                            int size, int meta_bg)
1141 {
1142         struct ext4_sb_info *sbi = EXT4_SB(sb);
1143         ext4_group_t last;
1144         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1145         unsigned three = 1;
1146         unsigned five = 5;
1147         unsigned seven = 7;
1148         ext4_group_t group = 0;
1149         int rest = sb->s_blocksize - size;
1150         handle_t *handle;
1151         int err = 0, err2;
1152
1153         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1154         if (IS_ERR(handle)) {
1155                 group = 1;
1156                 err = PTR_ERR(handle);
1157                 goto exit_err;
1158         }
1159
1160         if (meta_bg == 0) {
1161                 group = ext4_list_backups(sb, &three, &five, &seven);
1162                 last = sbi->s_groups_count;
1163         } else {
1164                 group = ext4_get_group_number(sb, blk_off) + 1;
1165                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1166         }
1167
1168         while (group < sbi->s_groups_count) {
1169                 struct buffer_head *bh;
1170                 ext4_fsblk_t backup_block;
1171                 int has_super = ext4_bg_has_super(sb, group);
1172                 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1173
1174                 /* Out of journal space, and can't get more - abort - so sad */
1175                 err = ext4_resize_ensure_credits_batch(handle, 1);
1176                 if (err < 0)
1177                         break;
1178
1179                 if (meta_bg == 0)
1180                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1181                 else
1182                         backup_block = first_block + has_super;
1183
1184                 bh = sb_getblk(sb, backup_block);
1185                 if (unlikely(!bh)) {
1186                         err = -ENOMEM;
1187                         break;
1188                 }
1189                 ext4_debug("update metadata backup %llu(+%llu)\n",
1190                            backup_block, backup_block -
1191                            ext4_group_first_block_no(sb, group));
1192                 BUFFER_TRACE(bh, "get_write_access");
1193                 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1194                                                          EXT4_JTR_NONE))) {
1195                         brelse(bh);
1196                         break;
1197                 }
1198                 lock_buffer(bh);
1199                 memcpy(bh->b_data, data, size);
1200                 if (rest)
1201                         memset(bh->b_data + size, 0, rest);
1202                 if (has_super && (backup_block == first_block))
1203                         ext4_set_block_group_nr(sb, bh->b_data, group);
1204                 set_buffer_uptodate(bh);
1205                 unlock_buffer(bh);
1206                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1207                 if (unlikely(err))
1208                         ext4_std_error(sb, err);
1209                 brelse(bh);
1210
1211                 if (meta_bg == 0)
1212                         group = ext4_list_backups(sb, &three, &five, &seven);
1213                 else if (group == last)
1214                         break;
1215                 else
1216                         group = last;
1217         }
1218         if ((err2 = ext4_journal_stop(handle)) && !err)
1219                 err = err2;
1220
1221         /*
1222          * Ugh! Need to have e2fsck write the backup copies.  It is too
1223          * late to revert the resize, we shouldn't fail just because of
1224          * the backup copies (they are only needed in case of corruption).
1225          *
1226          * However, if we got here we have a journal problem too, so we
1227          * can't really start a transaction to mark the superblock.
1228          * Chicken out and just set the flag on the hope it will be written
1229          * to disk, and if not - we will simply wait until next fsck.
1230          */
1231 exit_err:
1232         if (err) {
1233                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1234                              "forcing fsck on next reboot", group, err);
1235                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1236                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1237                 mark_buffer_dirty(sbi->s_sbh);
1238         }
1239 }
1240
1241 /*
1242  * ext4_add_new_descs() adds @count group descriptor of groups
1243  * starting at @group
1244  *
1245  * @handle: journal handle
1246  * @sb: super block
1247  * @group: the group no. of the first group desc to be added
1248  * @resize_inode: the resize inode
1249  * @count: number of group descriptors to be added
1250  */
1251 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1252                               ext4_group_t group, struct inode *resize_inode,
1253                               ext4_group_t count)
1254 {
1255         struct ext4_sb_info *sbi = EXT4_SB(sb);
1256         struct ext4_super_block *es = sbi->s_es;
1257         struct buffer_head *gdb_bh;
1258         int i, gdb_off, gdb_num, err = 0;
1259         int meta_bg;
1260
1261         meta_bg = ext4_has_feature_meta_bg(sb);
1262         for (i = 0; i < count; i++, group++) {
1263                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1264                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1265
1266                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1267                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1268
1269                 /*
1270                  * We will only either add reserved group blocks to a backup group
1271                  * or remove reserved blocks for the first group in a new group block.
1272                  * Doing both would be mean more complex code, and sane people don't
1273                  * use non-sparse filesystems anymore.  This is already checked above.
1274                  */
1275                 if (gdb_off) {
1276                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1277                                                      gdb_num);
1278                         BUFFER_TRACE(gdb_bh, "get_write_access");
1279                         err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1280                                                             EXT4_JTR_NONE);
1281
1282                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1283                                 err = reserve_backup_gdb(handle, resize_inode, group);
1284                 } else if (meta_bg != 0) {
1285                         err = add_new_gdb_meta_bg(sb, handle, group);
1286                 } else {
1287                         err = add_new_gdb(handle, resize_inode, group);
1288                 }
1289                 if (err)
1290                         break;
1291         }
1292         return err;
1293 }
1294
1295 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1296 {
1297         struct buffer_head *bh = sb_getblk(sb, block);
1298         if (unlikely(!bh))
1299                 return NULL;
1300         if (!bh_uptodate_or_lock(bh)) {
1301                 if (ext4_read_bh(bh, 0, NULL) < 0) {
1302                         brelse(bh);
1303                         return NULL;
1304                 }
1305         }
1306
1307         return bh;
1308 }
1309
1310 static int ext4_set_bitmap_checksums(struct super_block *sb,
1311                                      struct ext4_group_desc *gdp,
1312                                      struct ext4_new_group_data *group_data)
1313 {
1314         struct buffer_head *bh;
1315
1316         if (!ext4_has_metadata_csum(sb))
1317                 return 0;
1318
1319         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1320         if (!bh)
1321                 return -EIO;
1322         ext4_inode_bitmap_csum_set(sb, gdp, bh,
1323                                    EXT4_INODES_PER_GROUP(sb) / 8);
1324         brelse(bh);
1325
1326         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1327         if (!bh)
1328                 return -EIO;
1329         ext4_block_bitmap_csum_set(sb, gdp, bh);
1330         brelse(bh);
1331
1332         return 0;
1333 }
1334
1335 /*
1336  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1337  */
1338 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1339                                 struct ext4_new_flex_group_data *flex_gd)
1340 {
1341         struct ext4_new_group_data      *group_data = flex_gd->groups;
1342         struct ext4_group_desc          *gdp;
1343         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1344         struct buffer_head              *gdb_bh;
1345         ext4_group_t                    group;
1346         __u16                           *bg_flags = flex_gd->bg_flags;
1347         int                             i, gdb_off, gdb_num, err = 0;
1348
1349
1350         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1351                 group = group_data->group;
1352
1353                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1354                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1355
1356                 /*
1357                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1358                  */
1359                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1360                 /* Update group descriptor block for new group */
1361                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1362                                                  gdb_off * EXT4_DESC_SIZE(sb));
1363
1364                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1365                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1366                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1367                 err = ext4_set_bitmap_checksums(sb, gdp, group_data);
1368                 if (err) {
1369                         ext4_std_error(sb, err);
1370                         break;
1371                 }
1372
1373                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1374                 ext4_free_group_clusters_set(sb, gdp,
1375                                              group_data->free_clusters_count);
1376                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1377                 if (ext4_has_group_desc_csum(sb))
1378                         ext4_itable_unused_set(sb, gdp,
1379                                                EXT4_INODES_PER_GROUP(sb));
1380                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1381                 ext4_group_desc_csum_set(sb, group, gdp);
1382
1383                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1384                 if (unlikely(err)) {
1385                         ext4_std_error(sb, err);
1386                         break;
1387                 }
1388
1389                 /*
1390                  * We can allocate memory for mb_alloc based on the new group
1391                  * descriptor
1392                  */
1393                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1394                 if (err)
1395                         break;
1396         }
1397         return err;
1398 }
1399
1400 static void ext4_add_overhead(struct super_block *sb,
1401                               const ext4_fsblk_t overhead)
1402 {
1403        struct ext4_sb_info *sbi = EXT4_SB(sb);
1404        struct ext4_super_block *es = sbi->s_es;
1405
1406        sbi->s_overhead += overhead;
1407        es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1408        smp_wmb();
1409 }
1410
1411 /*
1412  * ext4_update_super() updates the super block so that the newly added
1413  * groups can be seen by the filesystem.
1414  *
1415  * @sb: super block
1416  * @flex_gd: new added groups
1417  */
1418 static void ext4_update_super(struct super_block *sb,
1419                              struct ext4_new_flex_group_data *flex_gd)
1420 {
1421         ext4_fsblk_t blocks_count = 0;
1422         ext4_fsblk_t free_blocks = 0;
1423         ext4_fsblk_t reserved_blocks = 0;
1424         struct ext4_new_group_data *group_data = flex_gd->groups;
1425         struct ext4_sb_info *sbi = EXT4_SB(sb);
1426         struct ext4_super_block *es = sbi->s_es;
1427         int i;
1428
1429         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1430         /*
1431          * Make the new blocks and inodes valid next.  We do this before
1432          * increasing the group count so that once the group is enabled,
1433          * all of its blocks and inodes are already valid.
1434          *
1435          * We always allocate group-by-group, then block-by-block or
1436          * inode-by-inode within a group, so enabling these
1437          * blocks/inodes before the group is live won't actually let us
1438          * allocate the new space yet.
1439          */
1440         for (i = 0; i < flex_gd->count; i++) {
1441                 blocks_count += group_data[i].blocks_count;
1442                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1443         }
1444
1445         reserved_blocks = ext4_r_blocks_count(es) * 100;
1446         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1447         reserved_blocks *= blocks_count;
1448         do_div(reserved_blocks, 100);
1449
1450         lock_buffer(sbi->s_sbh);
1451         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1452         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1453         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1454                      flex_gd->count);
1455         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1456                      flex_gd->count);
1457
1458         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1459         /*
1460          * We need to protect s_groups_count against other CPUs seeing
1461          * inconsistent state in the superblock.
1462          *
1463          * The precise rules we use are:
1464          *
1465          * * Writers must perform a smp_wmb() after updating all
1466          *   dependent data and before modifying the groups count
1467          *
1468          * * Readers must perform an smp_rmb() after reading the groups
1469          *   count and before reading any dependent data.
1470          *
1471          * NB. These rules can be relaxed when checking the group count
1472          * while freeing data, as we can only allocate from a block
1473          * group after serialising against the group count, and we can
1474          * only then free after serialising in turn against that
1475          * allocation.
1476          */
1477         smp_wmb();
1478
1479         /* Update the global fs size fields */
1480         sbi->s_groups_count += flex_gd->count;
1481         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1482                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1483
1484         /* Update the reserved block counts only once the new group is
1485          * active. */
1486         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1487                                 reserved_blocks);
1488
1489         /* Update the free space counts */
1490         percpu_counter_add(&sbi->s_freeclusters_counter,
1491                            EXT4_NUM_B2C(sbi, free_blocks));
1492         percpu_counter_add(&sbi->s_freeinodes_counter,
1493                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1494
1495         ext4_debug("free blocks count %llu",
1496                    percpu_counter_read(&sbi->s_freeclusters_counter));
1497         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1498                 ext4_group_t flex_group;
1499                 struct flex_groups *fg;
1500
1501                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1502                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1503                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1504                              &fg->free_clusters);
1505                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1506                            &fg->free_inodes);
1507         }
1508
1509         /*
1510          * Update the fs overhead information.
1511          *
1512          * For bigalloc, if the superblock already has a properly calculated
1513          * overhead, update it with a value based on numbers already computed
1514          * above for the newly allocated capacity.
1515          */
1516         if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1517                 ext4_add_overhead(sb,
1518                         EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1519         else
1520                 ext4_calculate_overhead(sb);
1521         es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1522
1523         ext4_superblock_csum_set(sb);
1524         unlock_buffer(sbi->s_sbh);
1525         if (test_opt(sb, DEBUG))
1526                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1527                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1528                        blocks_count, free_blocks, reserved_blocks);
1529 }
1530
1531 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1532  * _before_ we start modifying the filesystem, because we cannot abort the
1533  * transaction and not have it write the data to disk.
1534  */
1535 static int ext4_flex_group_add(struct super_block *sb,
1536                                struct inode *resize_inode,
1537                                struct ext4_new_flex_group_data *flex_gd)
1538 {
1539         struct ext4_sb_info *sbi = EXT4_SB(sb);
1540         struct ext4_super_block *es = sbi->s_es;
1541         ext4_fsblk_t o_blocks_count;
1542         ext4_grpblk_t last;
1543         ext4_group_t group;
1544         handle_t *handle;
1545         unsigned reserved_gdb;
1546         int err = 0, err2 = 0, credit;
1547
1548         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1549
1550         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1551         o_blocks_count = ext4_blocks_count(es);
1552         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1553         BUG_ON(last);
1554
1555         err = setup_new_flex_group_blocks(sb, flex_gd);
1556         if (err)
1557                 goto exit;
1558         /*
1559          * We will always be modifying at least the superblock and  GDT
1560          * blocks.  If we are adding a group past the last current GDT block,
1561          * we will also modify the inode and the dindirect block.  If we
1562          * are adding a group with superblock/GDT backups  we will also
1563          * modify each of the reserved GDT dindirect blocks.
1564          */
1565         credit = 3;     /* sb, resize inode, resize inode dindirect */
1566         /* GDT blocks */
1567         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1568         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1569         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1570         if (IS_ERR(handle)) {
1571                 err = PTR_ERR(handle);
1572                 goto exit;
1573         }
1574
1575         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1576         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1577                                             EXT4_JTR_NONE);
1578         if (err)
1579                 goto exit_journal;
1580
1581         group = flex_gd->groups[0].group;
1582         BUG_ON(group != sbi->s_groups_count);
1583         err = ext4_add_new_descs(handle, sb, group,
1584                                 resize_inode, flex_gd->count);
1585         if (err)
1586                 goto exit_journal;
1587
1588         err = ext4_setup_new_descs(handle, sb, flex_gd);
1589         if (err)
1590                 goto exit_journal;
1591
1592         ext4_update_super(sb, flex_gd);
1593
1594         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1595
1596 exit_journal:
1597         err2 = ext4_journal_stop(handle);
1598         if (!err)
1599                 err = err2;
1600
1601         if (!err) {
1602                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1603                 int gdb_num_end = ((group + flex_gd->count - 1) /
1604                                    EXT4_DESC_PER_BLOCK(sb));
1605                 int meta_bg = ext4_has_feature_meta_bg(sb) &&
1606                               gdb_num >= le32_to_cpu(es->s_first_meta_bg);
1607                 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1608                                          ext4_group_first_block_no(sb, 0);
1609
1610                 update_backups(sb, ext4_group_first_block_no(sb, 0),
1611                                (char *)es, sizeof(struct ext4_super_block), 0);
1612                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1613                         struct buffer_head *gdb_bh;
1614
1615                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1616                                                      gdb_num);
1617                         update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1618                                        gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1619                 }
1620         }
1621 exit:
1622         return err;
1623 }
1624
1625 static int ext4_setup_next_flex_gd(struct super_block *sb,
1626                                     struct ext4_new_flex_group_data *flex_gd,
1627                                     ext4_fsblk_t n_blocks_count)
1628 {
1629         struct ext4_sb_info *sbi = EXT4_SB(sb);
1630         struct ext4_super_block *es = sbi->s_es;
1631         struct ext4_new_group_data *group_data = flex_gd->groups;
1632         ext4_fsblk_t o_blocks_count;
1633         ext4_group_t n_group;
1634         ext4_group_t group;
1635         ext4_group_t last_group;
1636         ext4_grpblk_t last;
1637         ext4_grpblk_t clusters_per_group;
1638         unsigned long i;
1639
1640         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1641
1642         o_blocks_count = ext4_blocks_count(es);
1643
1644         if (o_blocks_count == n_blocks_count)
1645                 return 0;
1646
1647         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1648         BUG_ON(last);
1649         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1650
1651         last_group = group | (flex_gd->resize_bg - 1);
1652         if (last_group > n_group)
1653                 last_group = n_group;
1654
1655         flex_gd->count = last_group - group + 1;
1656
1657         for (i = 0; i < flex_gd->count; i++) {
1658                 int overhead;
1659
1660                 group_data[i].group = group + i;
1661                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1662                 overhead = ext4_group_overhead_blocks(sb, group + i);
1663                 group_data[i].mdata_blocks = overhead;
1664                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1665                 if (ext4_has_group_desc_csum(sb)) {
1666                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1667                                                EXT4_BG_INODE_UNINIT;
1668                         if (!test_opt(sb, INIT_INODE_TABLE))
1669                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1670                 } else
1671                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1672         }
1673
1674         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1675                 /* We need to initialize block bitmap of last group. */
1676                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1677
1678         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1679                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1680                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1681                                                        last - 1;
1682         }
1683
1684         return 1;
1685 }
1686
1687 /* Add group descriptor data to an existing or new group descriptor block.
1688  * Ensure we handle all possible error conditions _before_ we start modifying
1689  * the filesystem, because we cannot abort the transaction and not have it
1690  * write the data to disk.
1691  *
1692  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1693  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1694  *
1695  * We only need to hold the superblock lock while we are actually adding
1696  * in the new group's counts to the superblock.  Prior to that we have
1697  * not really "added" the group at all.  We re-check that we are still
1698  * adding in the last group in case things have changed since verifying.
1699  */
1700 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1701 {
1702         struct ext4_new_flex_group_data flex_gd;
1703         struct ext4_sb_info *sbi = EXT4_SB(sb);
1704         struct ext4_super_block *es = sbi->s_es;
1705         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1706                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1707         struct inode *inode = NULL;
1708         int gdb_off;
1709         int err;
1710         __u16 bg_flags = 0;
1711
1712         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1713
1714         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1715                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1716                 return -EPERM;
1717         }
1718
1719         if (ext4_blocks_count(es) + input->blocks_count <
1720             ext4_blocks_count(es)) {
1721                 ext4_warning(sb, "blocks_count overflow");
1722                 return -EINVAL;
1723         }
1724
1725         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1726             le32_to_cpu(es->s_inodes_count)) {
1727                 ext4_warning(sb, "inodes_count overflow");
1728                 return -EINVAL;
1729         }
1730
1731         if (reserved_gdb || gdb_off == 0) {
1732                 if (!ext4_has_feature_resize_inode(sb) ||
1733                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1734                         ext4_warning(sb,
1735                                      "No reserved GDT blocks, can't resize");
1736                         return -EPERM;
1737                 }
1738                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1739                 if (IS_ERR(inode)) {
1740                         ext4_warning(sb, "Error opening resize inode");
1741                         return PTR_ERR(inode);
1742                 }
1743         }
1744
1745
1746         err = verify_group_input(sb, input);
1747         if (err)
1748                 goto out;
1749
1750         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1751         if (err)
1752                 goto out;
1753
1754         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1755         if (err)
1756                 goto out;
1757
1758         flex_gd.count = 1;
1759         flex_gd.groups = input;
1760         flex_gd.bg_flags = &bg_flags;
1761         err = ext4_flex_group_add(sb, inode, &flex_gd);
1762 out:
1763         iput(inode);
1764         return err;
1765 } /* ext4_group_add */
1766
1767 /*
1768  * extend a group without checking assuming that checking has been done.
1769  */
1770 static int ext4_group_extend_no_check(struct super_block *sb,
1771                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1772 {
1773         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1774         handle_t *handle;
1775         int err = 0, err2;
1776
1777         /* We will update the superblock, one block bitmap, and
1778          * one group descriptor via ext4_group_add_blocks().
1779          */
1780         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1781         if (IS_ERR(handle)) {
1782                 err = PTR_ERR(handle);
1783                 ext4_warning(sb, "error %d on journal start", err);
1784                 return err;
1785         }
1786
1787         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1788         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1789                                             EXT4_JTR_NONE);
1790         if (err) {
1791                 ext4_warning(sb, "error %d on journal write access", err);
1792                 goto errout;
1793         }
1794
1795         lock_buffer(EXT4_SB(sb)->s_sbh);
1796         ext4_blocks_count_set(es, o_blocks_count + add);
1797         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1798         ext4_superblock_csum_set(sb);
1799         unlock_buffer(EXT4_SB(sb)->s_sbh);
1800         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1801                    o_blocks_count + add);
1802         /* We add the blocks to the bitmap and set the group need init bit */
1803         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1804         if (err)
1805                 goto errout;
1806         ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1807         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1808                    o_blocks_count + add);
1809 errout:
1810         err2 = ext4_journal_stop(handle);
1811         if (err2 && !err)
1812                 err = err2;
1813
1814         if (!err) {
1815                 if (test_opt(sb, DEBUG))
1816                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1817                                "blocks\n", ext4_blocks_count(es));
1818                 update_backups(sb, ext4_group_first_block_no(sb, 0),
1819                                (char *)es, sizeof(struct ext4_super_block), 0);
1820         }
1821         return err;
1822 }
1823
1824 /*
1825  * Extend the filesystem to the new number of blocks specified.  This entry
1826  * point is only used to extend the current filesystem to the end of the last
1827  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1828  * for emergencies (because it has no dependencies on reserved blocks).
1829  *
1830  * If we _really_ wanted, we could use default values to call ext4_group_add()
1831  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1832  * GDT blocks are reserved to grow to the desired size.
1833  */
1834 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1835                       ext4_fsblk_t n_blocks_count)
1836 {
1837         ext4_fsblk_t o_blocks_count;
1838         ext4_grpblk_t last;
1839         ext4_grpblk_t add;
1840         struct buffer_head *bh;
1841         ext4_group_t group;
1842
1843         o_blocks_count = ext4_blocks_count(es);
1844
1845         if (test_opt(sb, DEBUG))
1846                 ext4_msg(sb, KERN_DEBUG,
1847                          "extending last group from %llu to %llu blocks",
1848                          o_blocks_count, n_blocks_count);
1849
1850         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1851                 return 0;
1852
1853         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1854                 ext4_msg(sb, KERN_ERR,
1855                          "filesystem too large to resize to %llu blocks safely",
1856                          n_blocks_count);
1857                 return -EINVAL;
1858         }
1859
1860         if (n_blocks_count < o_blocks_count) {
1861                 ext4_warning(sb, "can't shrink FS - resize aborted");
1862                 return -EINVAL;
1863         }
1864
1865         /* Handle the remaining blocks in the last group only. */
1866         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1867
1868         if (last == 0) {
1869                 ext4_warning(sb, "need to use ext2online to resize further");
1870                 return -EPERM;
1871         }
1872
1873         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1874
1875         if (o_blocks_count + add < o_blocks_count) {
1876                 ext4_warning(sb, "blocks_count overflow");
1877                 return -EINVAL;
1878         }
1879
1880         if (o_blocks_count + add > n_blocks_count)
1881                 add = n_blocks_count - o_blocks_count;
1882
1883         if (o_blocks_count + add < n_blocks_count)
1884                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1885                              o_blocks_count + add, add);
1886
1887         /* See if the device is actually as big as what was requested */
1888         bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1889         if (IS_ERR(bh)) {
1890                 ext4_warning(sb, "can't read last block, resize aborted");
1891                 return -ENOSPC;
1892         }
1893         brelse(bh);
1894
1895         return ext4_group_extend_no_check(sb, o_blocks_count, add);
1896 } /* ext4_group_extend */
1897
1898
1899 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1900 {
1901         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1902 }
1903
1904 /*
1905  * Release the resize inode and drop the resize_inode feature if there
1906  * are no more reserved gdt blocks, and then convert the file system
1907  * to enable meta_bg
1908  */
1909 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1910 {
1911         handle_t *handle;
1912         struct ext4_sb_info *sbi = EXT4_SB(sb);
1913         struct ext4_super_block *es = sbi->s_es;
1914         struct ext4_inode_info *ei = EXT4_I(inode);
1915         ext4_fsblk_t nr;
1916         int i, ret, err = 0;
1917         int credits = 1;
1918
1919         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1920         if (inode) {
1921                 if (es->s_reserved_gdt_blocks) {
1922                         ext4_error(sb, "Unexpected non-zero "
1923                                    "s_reserved_gdt_blocks");
1924                         return -EPERM;
1925                 }
1926
1927                 /* Do a quick sanity check of the resize inode */
1928                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1929                                              (9 - sbi->s_cluster_bits)))
1930                         goto invalid_resize_inode;
1931                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1932                         if (i == EXT4_DIND_BLOCK) {
1933                                 if (ei->i_data[i])
1934                                         continue;
1935                                 else
1936                                         goto invalid_resize_inode;
1937                         }
1938                         if (ei->i_data[i])
1939                                 goto invalid_resize_inode;
1940                 }
1941                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1942         }
1943
1944         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1945         if (IS_ERR(handle))
1946                 return PTR_ERR(handle);
1947
1948         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1949         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1950                                             EXT4_JTR_NONE);
1951         if (err)
1952                 goto errout;
1953
1954         lock_buffer(sbi->s_sbh);
1955         ext4_clear_feature_resize_inode(sb);
1956         ext4_set_feature_meta_bg(sb);
1957         sbi->s_es->s_first_meta_bg =
1958                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1959         ext4_superblock_csum_set(sb);
1960         unlock_buffer(sbi->s_sbh);
1961
1962         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1963         if (err) {
1964                 ext4_std_error(sb, err);
1965                 goto errout;
1966         }
1967
1968         if (inode) {
1969                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1970                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1971                                  EXT4_FREE_BLOCKS_METADATA |
1972                                  EXT4_FREE_BLOCKS_FORGET);
1973                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1974                 inode->i_blocks = 0;
1975
1976                 err = ext4_mark_inode_dirty(handle, inode);
1977                 if (err)
1978                         ext4_std_error(sb, err);
1979         }
1980
1981 errout:
1982         ret = ext4_journal_stop(handle);
1983         return err ? err : ret;
1984
1985 invalid_resize_inode:
1986         ext4_error(sb, "corrupted/inconsistent resize inode");
1987         return -EINVAL;
1988 }
1989
1990 /*
1991  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1992  *
1993  * @sb: super block of the fs to be resized
1994  * @n_blocks_count: the number of blocks resides in the resized fs
1995  */
1996 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1997 {
1998         struct ext4_new_flex_group_data *flex_gd = NULL;
1999         struct ext4_sb_info *sbi = EXT4_SB(sb);
2000         struct ext4_super_block *es = sbi->s_es;
2001         struct buffer_head *bh;
2002         struct inode *resize_inode = NULL;
2003         ext4_grpblk_t add, offset;
2004         unsigned long n_desc_blocks;
2005         unsigned long o_desc_blocks;
2006         ext4_group_t o_group;
2007         ext4_group_t n_group;
2008         ext4_fsblk_t o_blocks_count;
2009         ext4_fsblk_t n_blocks_count_retry = 0;
2010         unsigned long last_update_time = 0;
2011         int err = 0;
2012         int meta_bg;
2013         unsigned int flexbg_size = ext4_flex_bg_size(sbi);
2014
2015         /* See if the device is actually as big as what was requested */
2016         bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2017         if (IS_ERR(bh)) {
2018                 ext4_warning(sb, "can't read last block, resize aborted");
2019                 return -ENOSPC;
2020         }
2021         brelse(bh);
2022
2023         /*
2024          * For bigalloc, trim the requested size to the nearest cluster
2025          * boundary to avoid creating an unusable filesystem. We do this
2026          * silently, instead of returning an error, to avoid breaking
2027          * callers that blindly resize the filesystem to the full size of
2028          * the underlying block device.
2029          */
2030         if (ext4_has_feature_bigalloc(sb))
2031                 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2032
2033 retry:
2034         o_blocks_count = ext4_blocks_count(es);
2035
2036         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2037                  "to %llu blocks", o_blocks_count, n_blocks_count);
2038
2039         if (n_blocks_count < o_blocks_count) {
2040                 /* On-line shrinking not supported */
2041                 ext4_warning(sb, "can't shrink FS - resize aborted");
2042                 return -EINVAL;
2043         }
2044
2045         if (n_blocks_count == o_blocks_count)
2046                 /* Nothing need to do */
2047                 return 0;
2048
2049         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2050         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2051                 ext4_warning(sb, "resize would cause inodes_count overflow");
2052                 return -EINVAL;
2053         }
2054         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2055
2056         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2057         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2058
2059         meta_bg = ext4_has_feature_meta_bg(sb);
2060
2061         if (ext4_has_feature_resize_inode(sb)) {
2062                 if (meta_bg) {
2063                         ext4_error(sb, "resize_inode and meta_bg enabled "
2064                                    "simultaneously");
2065                         return -EINVAL;
2066                 }
2067                 if (n_desc_blocks > o_desc_blocks +
2068                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
2069                         n_blocks_count_retry = n_blocks_count;
2070                         n_desc_blocks = o_desc_blocks +
2071                                 le16_to_cpu(es->s_reserved_gdt_blocks);
2072                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2073                         n_blocks_count = (ext4_fsblk_t)n_group *
2074                                 EXT4_BLOCKS_PER_GROUP(sb) +
2075                                 le32_to_cpu(es->s_first_data_block);
2076                         n_group--; /* set to last group number */
2077                 }
2078
2079                 if (!resize_inode)
2080                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2081                                                  EXT4_IGET_SPECIAL);
2082                 if (IS_ERR(resize_inode)) {
2083                         ext4_warning(sb, "Error opening resize inode");
2084                         return PTR_ERR(resize_inode);
2085                 }
2086         }
2087
2088         if ((!resize_inode && !meta_bg && n_desc_blocks > o_desc_blocks) || n_blocks_count == o_blocks_count) {
2089                 err = ext4_convert_meta_bg(sb, resize_inode);
2090                 if (err)
2091                         goto out;
2092                 if (resize_inode) {
2093                         iput(resize_inode);
2094                         resize_inode = NULL;
2095                 }
2096                 if (n_blocks_count_retry) {
2097                         n_blocks_count = n_blocks_count_retry;
2098                         n_blocks_count_retry = 0;
2099                         goto retry;
2100                 }
2101         }
2102
2103         /*
2104          * Make sure the last group has enough space so that it's
2105          * guaranteed to have enough space for all metadata blocks
2106          * that it might need to hold.  (We might not need to store
2107          * the inode table blocks in the last block group, but there
2108          * will be cases where this might be needed.)
2109          */
2110         if ((ext4_group_first_block_no(sb, n_group) +
2111              ext4_group_overhead_blocks(sb, n_group) + 2 +
2112              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2113                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2114                 n_group--;
2115                 n_blocks_count_retry = 0;
2116                 if (resize_inode) {
2117                         iput(resize_inode);
2118                         resize_inode = NULL;
2119                 }
2120                 goto retry;
2121         }
2122
2123         /* extend the last group */
2124         if (n_group == o_group)
2125                 add = n_blocks_count - o_blocks_count;
2126         else
2127                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2128         if (add > 0) {
2129                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2130                 if (err)
2131                         goto out;
2132         }
2133
2134         if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2135                 goto out;
2136
2137         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2138         if (err)
2139                 goto out;
2140
2141         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2142         if (err)
2143                 goto out;
2144
2145         flex_gd = alloc_flex_gd(flexbg_size, o_group, n_group);
2146         if (flex_gd == NULL) {
2147                 err = -ENOMEM;
2148                 goto out;
2149         }
2150
2151         /* Add flex groups. Note that a regular group is a
2152          * flex group with 1 group.
2153          */
2154         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2155                 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2156                         if (last_update_time)
2157                                 ext4_msg(sb, KERN_INFO,
2158                                          "resized to %llu blocks",
2159                                          ext4_blocks_count(es));
2160                         last_update_time = jiffies;
2161                 }
2162                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2163                         break;
2164                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2165                 if (unlikely(err))
2166                         break;
2167         }
2168
2169         if (!err && n_blocks_count_retry) {
2170                 n_blocks_count = n_blocks_count_retry;
2171                 n_blocks_count_retry = 0;
2172                 free_flex_gd(flex_gd);
2173                 flex_gd = NULL;
2174                 if (resize_inode) {
2175                         iput(resize_inode);
2176                         resize_inode = NULL;
2177                 }
2178                 goto retry;
2179         }
2180
2181 out:
2182         if (flex_gd)
2183                 free_flex_gd(flex_gd);
2184         if (resize_inode != NULL)
2185                 iput(resize_inode);
2186         if (err)
2187                 ext4_warning(sb, "error (%d) occurred during "
2188                              "file system resize", err);
2189         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2190                  ext4_blocks_count(es));
2191         return err;
2192 }