38ce42396758d94bb4114a2b69123dc118b69728
[releases.git] / 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 #define EXT4FS_DEBUG
14
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17 #include <linux/jiffies.h>
18
19 #include "ext4_jbd2.h"
20
21 struct ext4_rcu_ptr {
22         struct rcu_head rcu;
23         void *ptr;
24 };
25
26 static void ext4_rcu_ptr_callback(struct rcu_head *head)
27 {
28         struct ext4_rcu_ptr *ptr;
29
30         ptr = container_of(head, struct ext4_rcu_ptr, rcu);
31         kvfree(ptr->ptr);
32         kfree(ptr);
33 }
34
35 void ext4_kvfree_array_rcu(void *to_free)
36 {
37         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
38
39         if (ptr) {
40                 ptr->ptr = to_free;
41                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
42                 return;
43         }
44         synchronize_rcu();
45         kvfree(to_free);
46 }
47
48 int ext4_resize_begin(struct super_block *sb)
49 {
50         struct ext4_sb_info *sbi = EXT4_SB(sb);
51         int ret = 0;
52
53         if (!capable(CAP_SYS_RESOURCE))
54                 return -EPERM;
55
56         /*
57          * If the reserved GDT blocks is non-zero, the resize_inode feature
58          * should always be set.
59          */
60         if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
61             !ext4_has_feature_resize_inode(sb)) {
62                 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
63                 return -EFSCORRUPTED;
64         }
65
66         /*
67          * If we are not using the primary superblock/GDT copy don't resize,
68          * because the user tools have no way of handling this.  Probably a
69          * bad time to do it anyways.
70          */
71         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
72             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
73                 ext4_warning(sb, "won't resize using backup superblock at %llu",
74                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
75                 return -EPERM;
76         }
77
78         /*
79          * We are not allowed to do online-resizing on a filesystem mounted
80          * with error, because it can destroy the filesystem easily.
81          */
82         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
83                 ext4_warning(sb, "There are errors in the filesystem, "
84                              "so online resizing is not allowed");
85                 return -EPERM;
86         }
87
88         if (ext4_has_feature_sparse_super2(sb)) {
89                 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
90                 return -EOPNOTSUPP;
91         }
92
93         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
94                                   &EXT4_SB(sb)->s_ext4_flags))
95                 ret = -EBUSY;
96
97         return ret;
98 }
99
100 int ext4_resize_end(struct super_block *sb, bool update_backups)
101 {
102         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
103         smp_mb__after_atomic();
104         if (update_backups)
105                 return ext4_update_overhead(sb, true);
106         return 0;
107 }
108
109 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
110                                              ext4_group_t group) {
111         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
112                EXT4_DESC_PER_BLOCK_BITS(sb);
113 }
114
115 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
116                                              ext4_group_t group) {
117         group = ext4_meta_bg_first_group(sb, group);
118         return ext4_group_first_block_no(sb, group);
119 }
120
121 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
122                                                 ext4_group_t group) {
123         ext4_grpblk_t overhead;
124         overhead = ext4_bg_num_gdb(sb, group);
125         if (ext4_bg_has_super(sb, group))
126                 overhead += 1 +
127                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
128         return overhead;
129 }
130
131 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
132 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
133
134 static int verify_group_input(struct super_block *sb,
135                               struct ext4_new_group_data *input)
136 {
137         struct ext4_sb_info *sbi = EXT4_SB(sb);
138         struct ext4_super_block *es = sbi->s_es;
139         ext4_fsblk_t start = ext4_blocks_count(es);
140         ext4_fsblk_t end = start + input->blocks_count;
141         ext4_group_t group = input->group;
142         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
143         unsigned overhead;
144         ext4_fsblk_t metaend;
145         struct buffer_head *bh = NULL;
146         ext4_grpblk_t free_blocks_count, offset;
147         int err = -EINVAL;
148
149         if (group != sbi->s_groups_count) {
150                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
151                              input->group, sbi->s_groups_count);
152                 return -EINVAL;
153         }
154
155         overhead = ext4_group_overhead_blocks(sb, group);
156         metaend = start + overhead;
157         input->free_clusters_count = free_blocks_count =
158                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
159
160         if (test_opt(sb, DEBUG))
161                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
162                        "(%d free, %u reserved)\n",
163                        ext4_bg_has_super(sb, input->group) ? "normal" :
164                        "no-super", input->group, input->blocks_count,
165                        free_blocks_count, input->reserved_blocks);
166
167         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
168         if (offset != 0)
169                         ext4_warning(sb, "Last group not full");
170         else if (input->reserved_blocks > input->blocks_count / 5)
171                 ext4_warning(sb, "Reserved blocks too high (%u)",
172                              input->reserved_blocks);
173         else if (free_blocks_count < 0)
174                 ext4_warning(sb, "Bad blocks count %u",
175                              input->blocks_count);
176         else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
177                 err = PTR_ERR(bh);
178                 bh = NULL;
179                 ext4_warning(sb, "Cannot read last block (%llu)",
180                              end - 1);
181         } else if (outside(input->block_bitmap, start, end))
182                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
183                              (unsigned long long)input->block_bitmap);
184         else if (outside(input->inode_bitmap, start, end))
185                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
186                              (unsigned long long)input->inode_bitmap);
187         else if (outside(input->inode_table, start, end) ||
188                  outside(itend - 1, start, end))
189                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
190                              (unsigned long long)input->inode_table, itend - 1);
191         else if (input->inode_bitmap == input->block_bitmap)
192                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
193                              (unsigned long long)input->block_bitmap);
194         else if (inside(input->block_bitmap, input->inode_table, itend))
195                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
196                              "(%llu-%llu)",
197                              (unsigned long long)input->block_bitmap,
198                              (unsigned long long)input->inode_table, itend - 1);
199         else if (inside(input->inode_bitmap, input->inode_table, itend))
200                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
201                              "(%llu-%llu)",
202                              (unsigned long long)input->inode_bitmap,
203                              (unsigned long long)input->inode_table, itend - 1);
204         else if (inside(input->block_bitmap, start, metaend))
205                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
206                              (unsigned long long)input->block_bitmap,
207                              start, metaend - 1);
208         else if (inside(input->inode_bitmap, start, metaend))
209                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
210                              (unsigned long long)input->inode_bitmap,
211                              start, metaend - 1);
212         else if (inside(input->inode_table, start, metaend) ||
213                  inside(itend - 1, start, metaend))
214                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
215                              "(%llu-%llu)",
216                              (unsigned long long)input->inode_table,
217                              itend - 1, start, metaend - 1);
218         else
219                 err = 0;
220         brelse(bh);
221
222         return err;
223 }
224
225 /*
226  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
227  * group each time.
228  */
229 struct ext4_new_flex_group_data {
230         struct ext4_new_group_data *groups;     /* new_group_data for groups
231                                                    in the flex group */
232         __u16 *bg_flags;                        /* block group flags of groups
233                                                    in @groups */
234         ext4_group_t resize_bg;                 /* number of allocated
235                                                    new_group_data */
236         ext4_group_t count;                     /* number of groups in @groups
237                                                  */
238 };
239
240 /*
241  * Avoiding memory allocation failures due to too many groups added each time.
242  */
243 #define MAX_RESIZE_BG                           16384
244
245 /*
246  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
247  * @flexbg_size.
248  *
249  * Returns NULL on failure otherwise address of the allocated structure.
250  */
251 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned int flexbg_size)
252 {
253         struct ext4_new_flex_group_data *flex_gd;
254
255         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
256         if (flex_gd == NULL)
257                 goto out3;
258
259         if (unlikely(flexbg_size > MAX_RESIZE_BG))
260                 flex_gd->resize_bg = MAX_RESIZE_BG;
261         else
262                 flex_gd->resize_bg = flexbg_size;
263
264         flex_gd->groups = kmalloc_array(flex_gd->resize_bg,
265                                         sizeof(struct ext4_new_group_data),
266                                         GFP_NOFS);
267         if (flex_gd->groups == NULL)
268                 goto out2;
269
270         flex_gd->bg_flags = kmalloc_array(flex_gd->resize_bg, sizeof(__u16),
271                                           GFP_NOFS);
272         if (flex_gd->bg_flags == NULL)
273                 goto out1;
274
275         return flex_gd;
276
277 out1:
278         kfree(flex_gd->groups);
279 out2:
280         kfree(flex_gd);
281 out3:
282         return NULL;
283 }
284
285 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
286 {
287         kfree(flex_gd->bg_flags);
288         kfree(flex_gd->groups);
289         kfree(flex_gd);
290 }
291
292 /*
293  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
294  * and inode tables for a flex group.
295  *
296  * This function is used by 64bit-resize.  Note that this function allocates
297  * group tables from the 1st group of groups contained by @flexgd, which may
298  * be a partial of a flex group.
299  *
300  * @sb: super block of fs to which the groups belongs
301  *
302  * Returns 0 on a successful allocation of the metadata blocks in the
303  * block group.
304  */
305 static int ext4_alloc_group_tables(struct super_block *sb,
306                                 struct ext4_new_flex_group_data *flex_gd,
307                                 unsigned int flexbg_size)
308 {
309         struct ext4_new_group_data *group_data = flex_gd->groups;
310         ext4_fsblk_t start_blk;
311         ext4_fsblk_t last_blk;
312         ext4_group_t src_group;
313         ext4_group_t bb_index = 0;
314         ext4_group_t ib_index = 0;
315         ext4_group_t it_index = 0;
316         ext4_group_t group;
317         ext4_group_t last_group;
318         unsigned overhead;
319         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
320         int i;
321
322         BUG_ON(flex_gd->count == 0 || group_data == NULL);
323
324         src_group = group_data[0].group;
325         last_group  = src_group + flex_gd->count - 1;
326
327         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
328                (last_group & ~(flexbg_size - 1))));
329 next_group:
330         group = group_data[0].group;
331         if (src_group >= group_data[0].group + flex_gd->count)
332                 return -ENOSPC;
333         start_blk = ext4_group_first_block_no(sb, src_group);
334         last_blk = start_blk + group_data[src_group - group].blocks_count;
335
336         overhead = ext4_group_overhead_blocks(sb, src_group);
337
338         start_blk += overhead;
339
340         /* We collect contiguous blocks as much as possible. */
341         src_group++;
342         for (; src_group <= last_group; src_group++) {
343                 overhead = ext4_group_overhead_blocks(sb, src_group);
344                 if (overhead == 0)
345                         last_blk += group_data[src_group - group].blocks_count;
346                 else
347                         break;
348         }
349
350         /* Allocate block bitmaps */
351         for (; bb_index < flex_gd->count; bb_index++) {
352                 if (start_blk >= last_blk)
353                         goto next_group;
354                 group_data[bb_index].block_bitmap = start_blk++;
355                 group = ext4_get_group_number(sb, start_blk - 1);
356                 group -= group_data[0].group;
357                 group_data[group].mdata_blocks++;
358                 flex_gd->bg_flags[group] &= uninit_mask;
359         }
360
361         /* Allocate inode bitmaps */
362         for (; ib_index < flex_gd->count; ib_index++) {
363                 if (start_blk >= last_blk)
364                         goto next_group;
365                 group_data[ib_index].inode_bitmap = start_blk++;
366                 group = ext4_get_group_number(sb, start_blk - 1);
367                 group -= group_data[0].group;
368                 group_data[group].mdata_blocks++;
369                 flex_gd->bg_flags[group] &= uninit_mask;
370         }
371
372         /* Allocate inode tables */
373         for (; it_index < flex_gd->count; it_index++) {
374                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
375                 ext4_fsblk_t next_group_start;
376
377                 if (start_blk + itb > last_blk)
378                         goto next_group;
379                 group_data[it_index].inode_table = start_blk;
380                 group = ext4_get_group_number(sb, start_blk);
381                 next_group_start = ext4_group_first_block_no(sb, group + 1);
382                 group -= group_data[0].group;
383
384                 if (start_blk + itb > next_group_start) {
385                         flex_gd->bg_flags[group + 1] &= uninit_mask;
386                         overhead = start_blk + itb - next_group_start;
387                         group_data[group + 1].mdata_blocks += overhead;
388                         itb -= overhead;
389                 }
390
391                 group_data[group].mdata_blocks += itb;
392                 flex_gd->bg_flags[group] &= uninit_mask;
393                 start_blk += EXT4_SB(sb)->s_itb_per_group;
394         }
395
396         /* Update free clusters count to exclude metadata blocks */
397         for (i = 0; i < flex_gd->count; i++) {
398                 group_data[i].free_clusters_count -=
399                                 EXT4_NUM_B2C(EXT4_SB(sb),
400                                              group_data[i].mdata_blocks);
401         }
402
403         if (test_opt(sb, DEBUG)) {
404                 int i;
405                 group = group_data[0].group;
406
407                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
408                        "%u groups, flexbg size is %u:\n", flex_gd->count,
409                        flexbg_size);
410
411                 for (i = 0; i < flex_gd->count; i++) {
412                         ext4_debug(
413                                "adding %s group %u: %u blocks (%u free, %u mdata blocks)\n",
414                                ext4_bg_has_super(sb, group + i) ? "normal" :
415                                "no-super", group + i,
416                                group_data[i].blocks_count,
417                                group_data[i].free_clusters_count,
418                                group_data[i].mdata_blocks);
419                 }
420         }
421         return 0;
422 }
423
424 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
425                                   ext4_fsblk_t blk)
426 {
427         struct buffer_head *bh;
428         int err;
429
430         bh = sb_getblk(sb, blk);
431         if (unlikely(!bh))
432                 return ERR_PTR(-ENOMEM);
433         BUFFER_TRACE(bh, "get_write_access");
434         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
435         if (err) {
436                 brelse(bh);
437                 bh = ERR_PTR(err);
438         } else {
439                 memset(bh->b_data, 0, sb->s_blocksize);
440                 set_buffer_uptodate(bh);
441         }
442
443         return bh;
444 }
445
446 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
447 {
448         return ext4_journal_ensure_credits_fn(handle, credits,
449                 EXT4_MAX_TRANS_DATA, 0, 0);
450 }
451
452 /*
453  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
454  *
455  * Helper function for ext4_setup_new_group_blocks() which set .
456  *
457  * @sb: super block
458  * @handle: journal handle
459  * @flex_gd: flex group data
460  */
461 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
462                         struct ext4_new_flex_group_data *flex_gd,
463                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
464 {
465         struct ext4_sb_info *sbi = EXT4_SB(sb);
466         ext4_group_t count = last_cluster - first_cluster + 1;
467         ext4_group_t count2;
468
469         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
470                    last_cluster);
471         for (count2 = count; count > 0;
472              count -= count2, first_cluster += count2) {
473                 ext4_fsblk_t start;
474                 struct buffer_head *bh;
475                 ext4_group_t group;
476                 int err;
477
478                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
479                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
480                 group -= flex_gd->groups[0].group;
481
482                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
483                 if (count2 > count)
484                         count2 = count;
485
486                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
487                         BUG_ON(flex_gd->count > 1);
488                         continue;
489                 }
490
491                 err = ext4_resize_ensure_credits_batch(handle, 1);
492                 if (err < 0)
493                         return err;
494
495                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
496                 if (unlikely(!bh))
497                         return -ENOMEM;
498
499                 BUFFER_TRACE(bh, "get_write_access");
500                 err = ext4_journal_get_write_access(handle, sb, bh,
501                                                     EXT4_JTR_NONE);
502                 if (err) {
503                         brelse(bh);
504                         return err;
505                 }
506                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
507                            first_cluster, first_cluster - start, count2);
508                 mb_set_bits(bh->b_data, first_cluster - start, count2);
509
510                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
511                 brelse(bh);
512                 if (unlikely(err))
513                         return err;
514         }
515
516         return 0;
517 }
518
519 /*
520  * Set up the block and inode bitmaps, and the inode table for the new groups.
521  * This doesn't need to be part of the main transaction, since we are only
522  * changing blocks outside the actual filesystem.  We still do journaling to
523  * ensure the recovery is correct in case of a failure just after resize.
524  * If any part of this fails, we simply abort the resize.
525  *
526  * setup_new_flex_group_blocks handles a flex group as follow:
527  *  1. copy super block and GDT, and initialize group tables if necessary.
528  *     In this step, we only set bits in blocks bitmaps for blocks taken by
529  *     super block and GDT.
530  *  2. allocate group tables in block bitmaps, that is, set bits in block
531  *     bitmap for blocks taken by group tables.
532  */
533 static int setup_new_flex_group_blocks(struct super_block *sb,
534                                 struct ext4_new_flex_group_data *flex_gd)
535 {
536         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
537         ext4_fsblk_t start;
538         ext4_fsblk_t block;
539         struct ext4_sb_info *sbi = EXT4_SB(sb);
540         struct ext4_super_block *es = sbi->s_es;
541         struct ext4_new_group_data *group_data = flex_gd->groups;
542         __u16 *bg_flags = flex_gd->bg_flags;
543         handle_t *handle;
544         ext4_group_t group, count;
545         struct buffer_head *bh = NULL;
546         int reserved_gdb, i, j, err = 0, err2;
547         int meta_bg;
548
549         BUG_ON(!flex_gd->count || !group_data ||
550                group_data[0].group != sbi->s_groups_count);
551
552         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
553         meta_bg = ext4_has_feature_meta_bg(sb);
554
555         /* This transaction may be extended/restarted along the way */
556         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
557         if (IS_ERR(handle))
558                 return PTR_ERR(handle);
559
560         group = group_data[0].group;
561         for (i = 0; i < flex_gd->count; i++, group++) {
562                 unsigned long gdblocks;
563                 ext4_grpblk_t overhead;
564
565                 gdblocks = ext4_bg_num_gdb(sb, group);
566                 start = ext4_group_first_block_no(sb, group);
567
568                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
569                         goto handle_itb;
570
571                 if (meta_bg == 1)
572                         goto handle_itb;
573
574                 block = start + ext4_bg_has_super(sb, group);
575                 /* Copy all of the GDT blocks into the backup in this group */
576                 for (j = 0; j < gdblocks; j++, block++) {
577                         struct buffer_head *gdb;
578
579                         ext4_debug("update backup group %#04llx\n", block);
580                         err = ext4_resize_ensure_credits_batch(handle, 1);
581                         if (err < 0)
582                                 goto out;
583
584                         gdb = sb_getblk(sb, block);
585                         if (unlikely(!gdb)) {
586                                 err = -ENOMEM;
587                                 goto out;
588                         }
589
590                         BUFFER_TRACE(gdb, "get_write_access");
591                         err = ext4_journal_get_write_access(handle, sb, gdb,
592                                                             EXT4_JTR_NONE);
593                         if (err) {
594                                 brelse(gdb);
595                                 goto out;
596                         }
597                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
598                                 s_group_desc, j)->b_data, gdb->b_size);
599                         set_buffer_uptodate(gdb);
600
601                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
602                         if (unlikely(err)) {
603                                 brelse(gdb);
604                                 goto out;
605                         }
606                         brelse(gdb);
607                 }
608
609                 /* Zero out all of the reserved backup group descriptor
610                  * table blocks
611                  */
612                 if (ext4_bg_has_super(sb, group)) {
613                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
614                                         reserved_gdb, GFP_NOFS);
615                         if (err)
616                                 goto out;
617                 }
618
619 handle_itb:
620                 /* Initialize group tables of the grop @group */
621                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
622                         goto handle_bb;
623
624                 /* Zero out all of the inode table blocks */
625                 block = group_data[i].inode_table;
626                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
627                            block, sbi->s_itb_per_group);
628                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
629                                        GFP_NOFS);
630                 if (err)
631                         goto out;
632
633 handle_bb:
634                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
635                         goto handle_ib;
636
637                 /* Initialize block bitmap of the @group */
638                 block = group_data[i].block_bitmap;
639                 err = ext4_resize_ensure_credits_batch(handle, 1);
640                 if (err < 0)
641                         goto out;
642
643                 bh = bclean(handle, sb, block);
644                 if (IS_ERR(bh)) {
645                         err = PTR_ERR(bh);
646                         goto out;
647                 }
648                 overhead = ext4_group_overhead_blocks(sb, group);
649                 if (overhead != 0) {
650                         ext4_debug("mark backup superblock %#04llx (+0)\n",
651                                    start);
652                         mb_set_bits(bh->b_data, 0,
653                                       EXT4_NUM_B2C(sbi, overhead));
654                 }
655                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
656                                      sb->s_blocksize * 8, bh->b_data);
657                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
658                 brelse(bh);
659                 if (err)
660                         goto out;
661
662 handle_ib:
663                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
664                         continue;
665
666                 /* Initialize inode bitmap of the @group */
667                 block = group_data[i].inode_bitmap;
668                 err = ext4_resize_ensure_credits_batch(handle, 1);
669                 if (err < 0)
670                         goto out;
671                 /* Mark unused entries in inode bitmap used */
672                 bh = bclean(handle, sb, block);
673                 if (IS_ERR(bh)) {
674                         err = PTR_ERR(bh);
675                         goto out;
676                 }
677
678                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
679                                      sb->s_blocksize * 8, bh->b_data);
680                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
681                 brelse(bh);
682                 if (err)
683                         goto out;
684         }
685
686         /* Mark group tables in block bitmap */
687         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
688                 count = group_table_count[j];
689                 start = (&group_data[0].block_bitmap)[j];
690                 block = start;
691                 for (i = 1; i < flex_gd->count; i++) {
692                         block += group_table_count[j];
693                         if (block == (&group_data[i].block_bitmap)[j]) {
694                                 count += group_table_count[j];
695                                 continue;
696                         }
697                         err = set_flexbg_block_bitmap(sb, handle,
698                                                       flex_gd,
699                                                       EXT4_B2C(sbi, start),
700                                                       EXT4_B2C(sbi,
701                                                                start + count
702                                                                - 1));
703                         if (err)
704                                 goto out;
705                         count = group_table_count[j];
706                         start = (&group_data[i].block_bitmap)[j];
707                         block = start;
708                 }
709
710                 if (count) {
711                         err = set_flexbg_block_bitmap(sb, handle,
712                                                       flex_gd,
713                                                       EXT4_B2C(sbi, start),
714                                                       EXT4_B2C(sbi,
715                                                                start + count
716                                                                - 1));
717                         if (err)
718                                 goto out;
719                 }
720         }
721
722 out:
723         err2 = ext4_journal_stop(handle);
724         if (err2 && !err)
725                 err = err2;
726
727         return err;
728 }
729
730 /*
731  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
732  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
733  * calling this for the first time.  In a sparse filesystem it will be the
734  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
735  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
736  */
737 unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
738                                unsigned int *five, unsigned int *seven)
739 {
740         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
741         unsigned int *min = three;
742         int mult = 3;
743         unsigned int ret;
744
745         if (ext4_has_feature_sparse_super2(sb)) {
746                 do {
747                         if (*min > 2)
748                                 return UINT_MAX;
749                         ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
750                         *min += 1;
751                 } while (!ret);
752                 return ret;
753         }
754
755         if (!ext4_has_feature_sparse_super(sb)) {
756                 ret = *min;
757                 *min += 1;
758                 return ret;
759         }
760
761         if (*five < *min) {
762                 min = five;
763                 mult = 5;
764         }
765         if (*seven < *min) {
766                 min = seven;
767                 mult = 7;
768         }
769
770         ret = *min;
771         *min *= mult;
772
773         return ret;
774 }
775
776 /*
777  * Check that all of the backup GDT blocks are held in the primary GDT block.
778  * It is assumed that they are stored in group order.  Returns the number of
779  * groups in current filesystem that have BACKUPS, or -ve error code.
780  */
781 static int verify_reserved_gdb(struct super_block *sb,
782                                ext4_group_t end,
783                                struct buffer_head *primary)
784 {
785         const ext4_fsblk_t blk = primary->b_blocknr;
786         unsigned three = 1;
787         unsigned five = 5;
788         unsigned seven = 7;
789         unsigned grp;
790         __le32 *p = (__le32 *)primary->b_data;
791         int gdbackups = 0;
792
793         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
794                 if (le32_to_cpu(*p++) !=
795                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
796                         ext4_warning(sb, "reserved GDT %llu"
797                                      " missing grp %d (%llu)",
798                                      blk, grp,
799                                      grp *
800                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
801                                      blk);
802                         return -EINVAL;
803                 }
804                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
805                         return -EFBIG;
806         }
807
808         return gdbackups;
809 }
810
811 /*
812  * Called when we need to bring a reserved group descriptor table block into
813  * use from the resize inode.  The primary copy of the new GDT block currently
814  * is an indirect block (under the double indirect block in the resize inode).
815  * The new backup GDT blocks will be stored as leaf blocks in this indirect
816  * block, in group order.  Even though we know all the block numbers we need,
817  * we check to ensure that the resize inode has actually reserved these blocks.
818  *
819  * Don't need to update the block bitmaps because the blocks are still in use.
820  *
821  * We get all of the error cases out of the way, so that we are sure to not
822  * fail once we start modifying the data on disk, because JBD has no rollback.
823  */
824 static int add_new_gdb(handle_t *handle, struct inode *inode,
825                        ext4_group_t group)
826 {
827         struct super_block *sb = inode->i_sb;
828         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
829         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
830         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
831         struct buffer_head **o_group_desc, **n_group_desc = NULL;
832         struct buffer_head *dind = NULL;
833         struct buffer_head *gdb_bh = NULL;
834         int gdbackups;
835         struct ext4_iloc iloc = { .bh = NULL };
836         __le32 *data;
837         int err;
838
839         if (test_opt(sb, DEBUG))
840                 printk(KERN_DEBUG
841                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
842                        gdb_num);
843
844         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
845         if (IS_ERR(gdb_bh))
846                 return PTR_ERR(gdb_bh);
847
848         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
849         if (gdbackups < 0) {
850                 err = gdbackups;
851                 goto errout;
852         }
853
854         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
855         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
856         if (IS_ERR(dind)) {
857                 err = PTR_ERR(dind);
858                 dind = NULL;
859                 goto errout;
860         }
861
862         data = (__le32 *)dind->b_data;
863         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
864                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
865                              group, gdblock);
866                 err = -EINVAL;
867                 goto errout;
868         }
869
870         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
871         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
872                                             EXT4_JTR_NONE);
873         if (unlikely(err))
874                 goto errout;
875
876         BUFFER_TRACE(gdb_bh, "get_write_access");
877         err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
878         if (unlikely(err))
879                 goto errout;
880
881         BUFFER_TRACE(dind, "get_write_access");
882         err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
883         if (unlikely(err)) {
884                 ext4_std_error(sb, err);
885                 goto errout;
886         }
887
888         /* ext4_reserve_inode_write() gets a reference on the iloc */
889         err = ext4_reserve_inode_write(handle, inode, &iloc);
890         if (unlikely(err))
891                 goto errout;
892
893         n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
894                                 GFP_KERNEL);
895         if (!n_group_desc) {
896                 err = -ENOMEM;
897                 ext4_warning(sb, "not enough memory for %lu groups",
898                              gdb_num + 1);
899                 goto errout;
900         }
901
902         /*
903          * Finally, we have all of the possible failures behind us...
904          *
905          * Remove new GDT block from inode double-indirect block and clear out
906          * the new GDT block for use (which also "frees" the backup GDT blocks
907          * from the reserved inode).  We don't need to change the bitmaps for
908          * these blocks, because they are marked as in-use from being in the
909          * reserved inode, and will become GDT blocks (primary and backup).
910          */
911         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
912         err = ext4_handle_dirty_metadata(handle, NULL, dind);
913         if (unlikely(err)) {
914                 ext4_std_error(sb, err);
915                 goto errout;
916         }
917         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
918                            (9 - EXT4_SB(sb)->s_cluster_bits);
919         ext4_mark_iloc_dirty(handle, inode, &iloc);
920         memset(gdb_bh->b_data, 0, sb->s_blocksize);
921         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
922         if (unlikely(err)) {
923                 ext4_std_error(sb, err);
924                 iloc.bh = NULL;
925                 goto errout;
926         }
927         brelse(dind);
928
929         rcu_read_lock();
930         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
931         memcpy(n_group_desc, o_group_desc,
932                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
933         rcu_read_unlock();
934         n_group_desc[gdb_num] = gdb_bh;
935         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
936         EXT4_SB(sb)->s_gdb_count++;
937         ext4_kvfree_array_rcu(o_group_desc);
938
939         lock_buffer(EXT4_SB(sb)->s_sbh);
940         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
941         ext4_superblock_csum_set(sb);
942         unlock_buffer(EXT4_SB(sb)->s_sbh);
943         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
944         if (err)
945                 ext4_std_error(sb, err);
946         return err;
947 errout:
948         kvfree(n_group_desc);
949         brelse(iloc.bh);
950         brelse(dind);
951         brelse(gdb_bh);
952
953         ext4_debug("leaving with error %d\n", err);
954         return err;
955 }
956
957 /*
958  * add_new_gdb_meta_bg is the sister of add_new_gdb.
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_meta_bg_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                 /* printk("reserving backup %lu[%u] = %lu\n",
1094                        primary[i]->b_blocknr, gdbackups,
1095                        blk + primary[i]->b_blocknr); */
1096                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1097                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1098                 if (!err)
1099                         err = err2;
1100         }
1101
1102         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1103         ext4_mark_iloc_dirty(handle, inode, &iloc);
1104
1105 exit_bh:
1106         while (--res >= 0)
1107                 brelse(primary[res]);
1108         brelse(dind);
1109
1110 exit_free:
1111         kfree(primary);
1112
1113         return err;
1114 }
1115
1116 static inline void ext4_set_block_group_nr(struct super_block *sb, char *data,
1117                                            ext4_group_t group)
1118 {
1119         struct ext4_super_block *es = (struct ext4_super_block *) data;
1120
1121         es->s_block_group_nr = cpu_to_le16(group);
1122         if (ext4_has_metadata_csum(sb))
1123                 es->s_checksum = ext4_superblock_csum(sb, es);
1124 }
1125
1126 /*
1127  * Update the backup copies of the ext4 metadata.  These don't need to be part
1128  * of the main resize transaction, because e2fsck will re-write them if there
1129  * is a problem (basically only OOM will cause a problem).  However, we
1130  * _should_ update the backups if possible, in case the primary gets trashed
1131  * for some reason and we need to run e2fsck from a backup superblock.  The
1132  * important part is that the new block and inode counts are in the backup
1133  * superblocks, and the location of the new group metadata in the GDT backups.
1134  *
1135  * We do not need take the s_resize_lock for this, because these
1136  * blocks are not otherwise touched by the filesystem code when it is
1137  * mounted.  We don't need to worry about last changing from
1138  * sbi->s_groups_count, because the worst that can happen is that we
1139  * do not copy the full number of backups at this time.  The resize
1140  * which changed s_groups_count will backup again.
1141  */
1142 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1143                            int size, int meta_bg)
1144 {
1145         struct ext4_sb_info *sbi = EXT4_SB(sb);
1146         ext4_group_t last;
1147         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1148         unsigned three = 1;
1149         unsigned five = 5;
1150         unsigned seven = 7;
1151         ext4_group_t group = 0;
1152         int rest = sb->s_blocksize - size;
1153         handle_t *handle;
1154         int err = 0, err2;
1155
1156         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1157         if (IS_ERR(handle)) {
1158                 group = 1;
1159                 err = PTR_ERR(handle);
1160                 goto exit_err;
1161         }
1162
1163         if (meta_bg == 0) {
1164                 group = ext4_list_backups(sb, &three, &five, &seven);
1165                 last = sbi->s_groups_count;
1166         } else {
1167                 group = ext4_get_group_number(sb, blk_off) + 1;
1168                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1169         }
1170
1171         while (group < sbi->s_groups_count) {
1172                 struct buffer_head *bh;
1173                 ext4_fsblk_t backup_block;
1174                 int has_super = ext4_bg_has_super(sb, group);
1175                 ext4_fsblk_t first_block = ext4_group_first_block_no(sb, group);
1176
1177                 /* Out of journal space, and can't get more - abort - so sad */
1178                 err = ext4_resize_ensure_credits_batch(handle, 1);
1179                 if (err < 0)
1180                         break;
1181
1182                 if (meta_bg == 0)
1183                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1184                 else
1185                         backup_block = first_block + has_super;
1186
1187                 bh = sb_getblk(sb, backup_block);
1188                 if (unlikely(!bh)) {
1189                         err = -ENOMEM;
1190                         break;
1191                 }
1192                 ext4_debug("update metadata backup %llu(+%llu)\n",
1193                            backup_block, backup_block -
1194                            ext4_group_first_block_no(sb, group));
1195                 BUFFER_TRACE(bh, "get_write_access");
1196                 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1197                                                          EXT4_JTR_NONE))) {
1198                         brelse(bh);
1199                         break;
1200                 }
1201                 lock_buffer(bh);
1202                 memcpy(bh->b_data, data, size);
1203                 if (rest)
1204                         memset(bh->b_data + size, 0, rest);
1205                 if (has_super && (backup_block == first_block))
1206                         ext4_set_block_group_nr(sb, bh->b_data, group);
1207                 set_buffer_uptodate(bh);
1208                 unlock_buffer(bh);
1209                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1210                 if (unlikely(err))
1211                         ext4_std_error(sb, err);
1212                 brelse(bh);
1213
1214                 if (meta_bg == 0)
1215                         group = ext4_list_backups(sb, &three, &five, &seven);
1216                 else if (group == last)
1217                         break;
1218                 else
1219                         group = last;
1220         }
1221         if ((err2 = ext4_journal_stop(handle)) && !err)
1222                 err = err2;
1223
1224         /*
1225          * Ugh! Need to have e2fsck write the backup copies.  It is too
1226          * late to revert the resize, we shouldn't fail just because of
1227          * the backup copies (they are only needed in case of corruption).
1228          *
1229          * However, if we got here we have a journal problem too, so we
1230          * can't really start a transaction to mark the superblock.
1231          * Chicken out and just set the flag on the hope it will be written
1232          * to disk, and if not - we will simply wait until next fsck.
1233          */
1234 exit_err:
1235         if (err) {
1236                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1237                              "forcing fsck on next reboot", group, err);
1238                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1239                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1240                 mark_buffer_dirty(sbi->s_sbh);
1241         }
1242 }
1243
1244 /*
1245  * ext4_add_new_descs() adds @count group descriptor of groups
1246  * starting at @group
1247  *
1248  * @handle: journal handle
1249  * @sb: super block
1250  * @group: the group no. of the first group desc to be added
1251  * @resize_inode: the resize inode
1252  * @count: number of group descriptors to be added
1253  */
1254 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1255                               ext4_group_t group, struct inode *resize_inode,
1256                               ext4_group_t count)
1257 {
1258         struct ext4_sb_info *sbi = EXT4_SB(sb);
1259         struct ext4_super_block *es = sbi->s_es;
1260         struct buffer_head *gdb_bh;
1261         int i, gdb_off, gdb_num, err = 0;
1262         int meta_bg;
1263
1264         meta_bg = ext4_has_feature_meta_bg(sb);
1265         for (i = 0; i < count; i++, group++) {
1266                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1267                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1268
1269                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1270                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1271
1272                 /*
1273                  * We will only either add reserved group blocks to a backup group
1274                  * or remove reserved blocks for the first group in a new group block.
1275                  * Doing both would be mean more complex code, and sane people don't
1276                  * use non-sparse filesystems anymore.  This is already checked above.
1277                  */
1278                 if (gdb_off) {
1279                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1280                                                      gdb_num);
1281                         BUFFER_TRACE(gdb_bh, "get_write_access");
1282                         err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1283                                                             EXT4_JTR_NONE);
1284
1285                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1286                                 err = reserve_backup_gdb(handle, resize_inode, group);
1287                 } else if (meta_bg != 0) {
1288                         err = add_new_gdb_meta_bg(sb, handle, group);
1289                 } else {
1290                         err = add_new_gdb(handle, resize_inode, group);
1291                 }
1292                 if (err)
1293                         break;
1294         }
1295         return err;
1296 }
1297
1298 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1299 {
1300         struct buffer_head *bh = sb_getblk(sb, block);
1301         if (unlikely(!bh))
1302                 return NULL;
1303         if (!bh_uptodate_or_lock(bh)) {
1304                 if (ext4_read_bh(bh, 0, NULL) < 0) {
1305                         brelse(bh);
1306                         return NULL;
1307                 }
1308         }
1309
1310         return bh;
1311 }
1312
1313 static int ext4_set_bitmap_checksums(struct super_block *sb,
1314                                      ext4_group_t group,
1315                                      struct ext4_group_desc *gdp,
1316                                      struct ext4_new_group_data *group_data)
1317 {
1318         struct buffer_head *bh;
1319
1320         if (!ext4_has_metadata_csum(sb))
1321                 return 0;
1322
1323         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1324         if (!bh)
1325                 return -EIO;
1326         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1327                                    EXT4_INODES_PER_GROUP(sb) / 8);
1328         brelse(bh);
1329
1330         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1331         if (!bh)
1332                 return -EIO;
1333         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1334         brelse(bh);
1335
1336         return 0;
1337 }
1338
1339 /*
1340  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1341  */
1342 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1343                                 struct ext4_new_flex_group_data *flex_gd)
1344 {
1345         struct ext4_new_group_data      *group_data = flex_gd->groups;
1346         struct ext4_group_desc          *gdp;
1347         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1348         struct buffer_head              *gdb_bh;
1349         ext4_group_t                    group;
1350         __u16                           *bg_flags = flex_gd->bg_flags;
1351         int                             i, gdb_off, gdb_num, err = 0;
1352
1353
1354         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1355                 group = group_data->group;
1356
1357                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1358                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1359
1360                 /*
1361                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1362                  */
1363                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1364                 /* Update group descriptor block for new group */
1365                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1366                                                  gdb_off * EXT4_DESC_SIZE(sb));
1367
1368                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1369                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1370                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1371                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1372                 if (err) {
1373                         ext4_std_error(sb, err);
1374                         break;
1375                 }
1376
1377                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1378                 ext4_free_group_clusters_set(sb, gdp,
1379                                              group_data->free_clusters_count);
1380                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1381                 if (ext4_has_group_desc_csum(sb))
1382                         ext4_itable_unused_set(sb, gdp,
1383                                                EXT4_INODES_PER_GROUP(sb));
1384                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1385                 ext4_group_desc_csum_set(sb, group, gdp);
1386
1387                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1388                 if (unlikely(err)) {
1389                         ext4_std_error(sb, err);
1390                         break;
1391                 }
1392
1393                 /*
1394                  * We can allocate memory for mb_alloc based on the new group
1395                  * descriptor
1396                  */
1397                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1398                 if (err)
1399                         break;
1400         }
1401         return err;
1402 }
1403
1404 static void ext4_add_overhead(struct super_block *sb,
1405                               const ext4_fsblk_t overhead)
1406 {
1407        struct ext4_sb_info *sbi = EXT4_SB(sb);
1408        struct ext4_super_block *es = sbi->s_es;
1409
1410        sbi->s_overhead += overhead;
1411        es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1412        smp_wmb();
1413 }
1414
1415 /*
1416  * ext4_update_super() updates the super block so that the newly added
1417  * groups can be seen by the filesystem.
1418  *
1419  * @sb: super block
1420  * @flex_gd: new added groups
1421  */
1422 static void ext4_update_super(struct super_block *sb,
1423                              struct ext4_new_flex_group_data *flex_gd)
1424 {
1425         ext4_fsblk_t blocks_count = 0;
1426         ext4_fsblk_t free_blocks = 0;
1427         ext4_fsblk_t reserved_blocks = 0;
1428         struct ext4_new_group_data *group_data = flex_gd->groups;
1429         struct ext4_sb_info *sbi = EXT4_SB(sb);
1430         struct ext4_super_block *es = sbi->s_es;
1431         int i;
1432
1433         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1434         /*
1435          * Make the new blocks and inodes valid next.  We do this before
1436          * increasing the group count so that once the group is enabled,
1437          * all of its blocks and inodes are already valid.
1438          *
1439          * We always allocate group-by-group, then block-by-block or
1440          * inode-by-inode within a group, so enabling these
1441          * blocks/inodes before the group is live won't actually let us
1442          * allocate the new space yet.
1443          */
1444         for (i = 0; i < flex_gd->count; i++) {
1445                 blocks_count += group_data[i].blocks_count;
1446                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1447         }
1448
1449         reserved_blocks = ext4_r_blocks_count(es) * 100;
1450         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1451         reserved_blocks *= blocks_count;
1452         do_div(reserved_blocks, 100);
1453
1454         lock_buffer(sbi->s_sbh);
1455         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1456         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1457         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1458                      flex_gd->count);
1459         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1460                      flex_gd->count);
1461
1462         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1463         /*
1464          * We need to protect s_groups_count against other CPUs seeing
1465          * inconsistent state in the superblock.
1466          *
1467          * The precise rules we use are:
1468          *
1469          * * Writers must perform a smp_wmb() after updating all
1470          *   dependent data and before modifying the groups count
1471          *
1472          * * Readers must perform an smp_rmb() after reading the groups
1473          *   count and before reading any dependent data.
1474          *
1475          * NB. These rules can be relaxed when checking the group count
1476          * while freeing data, as we can only allocate from a block
1477          * group after serialising against the group count, and we can
1478          * only then free after serialising in turn against that
1479          * allocation.
1480          */
1481         smp_wmb();
1482
1483         /* Update the global fs size fields */
1484         sbi->s_groups_count += flex_gd->count;
1485         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1486                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1487
1488         /* Update the reserved block counts only once the new group is
1489          * active. */
1490         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1491                                 reserved_blocks);
1492
1493         /* Update the free space counts */
1494         percpu_counter_add(&sbi->s_freeclusters_counter,
1495                            EXT4_NUM_B2C(sbi, free_blocks));
1496         percpu_counter_add(&sbi->s_freeinodes_counter,
1497                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1498
1499         ext4_debug("free blocks count %llu",
1500                    percpu_counter_read(&sbi->s_freeclusters_counter));
1501         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1502                 ext4_group_t flex_group;
1503                 struct flex_groups *fg;
1504
1505                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1506                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1507                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1508                              &fg->free_clusters);
1509                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1510                            &fg->free_inodes);
1511         }
1512
1513         /*
1514          * Update the fs overhead information.
1515          *
1516          * For bigalloc, if the superblock already has a properly calculated
1517          * overhead, update it with a value based on numbers already computed
1518          * above for the newly allocated capacity.
1519          */
1520         if (ext4_has_feature_bigalloc(sb) && (sbi->s_overhead != 0))
1521                 ext4_add_overhead(sb,
1522                         EXT4_NUM_B2C(sbi, blocks_count - free_blocks));
1523         else
1524                 ext4_calculate_overhead(sb);
1525         es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1526
1527         ext4_superblock_csum_set(sb);
1528         unlock_buffer(sbi->s_sbh);
1529         if (test_opt(sb, DEBUG))
1530                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1531                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1532                        blocks_count, free_blocks, reserved_blocks);
1533 }
1534
1535 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1536  * _before_ we start modifying the filesystem, because we cannot abort the
1537  * transaction and not have it write the data to disk.
1538  */
1539 static int ext4_flex_group_add(struct super_block *sb,
1540                                struct inode *resize_inode,
1541                                struct ext4_new_flex_group_data *flex_gd)
1542 {
1543         struct ext4_sb_info *sbi = EXT4_SB(sb);
1544         struct ext4_super_block *es = sbi->s_es;
1545         ext4_fsblk_t o_blocks_count;
1546         ext4_grpblk_t last;
1547         ext4_group_t group;
1548         handle_t *handle;
1549         unsigned reserved_gdb;
1550         int err = 0, err2 = 0, credit;
1551
1552         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1553
1554         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1555         o_blocks_count = ext4_blocks_count(es);
1556         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1557         BUG_ON(last);
1558
1559         err = setup_new_flex_group_blocks(sb, flex_gd);
1560         if (err)
1561                 goto exit;
1562         /*
1563          * We will always be modifying at least the superblock and  GDT
1564          * blocks.  If we are adding a group past the last current GDT block,
1565          * we will also modify the inode and the dindirect block.  If we
1566          * are adding a group with superblock/GDT backups  we will also
1567          * modify each of the reserved GDT dindirect blocks.
1568          */
1569         credit = 3;     /* sb, resize inode, resize inode dindirect */
1570         /* GDT blocks */
1571         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1572         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1573         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1574         if (IS_ERR(handle)) {
1575                 err = PTR_ERR(handle);
1576                 goto exit;
1577         }
1578
1579         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1580         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1581                                             EXT4_JTR_NONE);
1582         if (err)
1583                 goto exit_journal;
1584
1585         group = flex_gd->groups[0].group;
1586         BUG_ON(group != sbi->s_groups_count);
1587         err = ext4_add_new_descs(handle, sb, group,
1588                                 resize_inode, flex_gd->count);
1589         if (err)
1590                 goto exit_journal;
1591
1592         err = ext4_setup_new_descs(handle, sb, flex_gd);
1593         if (err)
1594                 goto exit_journal;
1595
1596         ext4_update_super(sb, flex_gd);
1597
1598         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1599
1600 exit_journal:
1601         err2 = ext4_journal_stop(handle);
1602         if (!err)
1603                 err = err2;
1604
1605         if (!err) {
1606                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1607                 int gdb_num_end = ((group + flex_gd->count - 1) /
1608                                    EXT4_DESC_PER_BLOCK(sb));
1609                 int meta_bg = ext4_has_feature_meta_bg(sb) &&
1610                               gdb_num >= le32_to_cpu(es->s_first_meta_bg);
1611                 sector_t padding_blocks = meta_bg ? 0 : sbi->s_sbh->b_blocknr -
1612                                          ext4_group_first_block_no(sb, 0);
1613                 sector_t old_gdb = 0;
1614
1615                 update_backups(sb, ext4_group_first_block_no(sb, 0),
1616                                (char *)es, sizeof(struct ext4_super_block), 0);
1617                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1618                         struct buffer_head *gdb_bh;
1619
1620                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1621                                                      gdb_num);
1622                         if (old_gdb == gdb_bh->b_blocknr)
1623                                 continue;
1624                         update_backups(sb, gdb_bh->b_blocknr - padding_blocks,
1625                                        gdb_bh->b_data, gdb_bh->b_size, meta_bg);
1626                         old_gdb = gdb_bh->b_blocknr;
1627                 }
1628         }
1629 exit:
1630         return err;
1631 }
1632
1633 static int ext4_setup_next_flex_gd(struct super_block *sb,
1634                                     struct ext4_new_flex_group_data *flex_gd,
1635                                     ext4_fsblk_t n_blocks_count)
1636 {
1637         struct ext4_sb_info *sbi = EXT4_SB(sb);
1638         struct ext4_super_block *es = sbi->s_es;
1639         struct ext4_new_group_data *group_data = flex_gd->groups;
1640         ext4_fsblk_t o_blocks_count;
1641         ext4_group_t n_group;
1642         ext4_group_t group;
1643         ext4_group_t last_group;
1644         ext4_grpblk_t last;
1645         ext4_grpblk_t clusters_per_group;
1646         unsigned long i;
1647
1648         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1649
1650         o_blocks_count = ext4_blocks_count(es);
1651
1652         if (o_blocks_count == n_blocks_count)
1653                 return 0;
1654
1655         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1656         BUG_ON(last);
1657         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1658
1659         last_group = group | (flex_gd->resize_bg - 1);
1660         if (last_group > n_group)
1661                 last_group = n_group;
1662
1663         flex_gd->count = last_group - group + 1;
1664
1665         for (i = 0; i < flex_gd->count; i++) {
1666                 int overhead;
1667
1668                 group_data[i].group = group + i;
1669                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1670                 overhead = ext4_group_overhead_blocks(sb, group + i);
1671                 group_data[i].mdata_blocks = overhead;
1672                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1673                 if (ext4_has_group_desc_csum(sb)) {
1674                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1675                                                EXT4_BG_INODE_UNINIT;
1676                         if (!test_opt(sb, INIT_INODE_TABLE))
1677                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1678                 } else
1679                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1680         }
1681
1682         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1683                 /* We need to initialize block bitmap of last group. */
1684                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1685
1686         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1687                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1688                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1689                                                        last - 1;
1690         }
1691
1692         return 1;
1693 }
1694
1695 /* Add group descriptor data to an existing or new group descriptor block.
1696  * Ensure we handle all possible error conditions _before_ we start modifying
1697  * the filesystem, because we cannot abort the transaction and not have it
1698  * write the data to disk.
1699  *
1700  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1701  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1702  *
1703  * We only need to hold the superblock lock while we are actually adding
1704  * in the new group's counts to the superblock.  Prior to that we have
1705  * not really "added" the group at all.  We re-check that we are still
1706  * adding in the last group in case things have changed since verifying.
1707  */
1708 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1709 {
1710         struct ext4_new_flex_group_data flex_gd;
1711         struct ext4_sb_info *sbi = EXT4_SB(sb);
1712         struct ext4_super_block *es = sbi->s_es;
1713         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1714                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1715         struct inode *inode = NULL;
1716         int gdb_off;
1717         int err;
1718         __u16 bg_flags = 0;
1719
1720         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1721
1722         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1723                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1724                 return -EPERM;
1725         }
1726
1727         if (ext4_blocks_count(es) + input->blocks_count <
1728             ext4_blocks_count(es)) {
1729                 ext4_warning(sb, "blocks_count overflow");
1730                 return -EINVAL;
1731         }
1732
1733         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1734             le32_to_cpu(es->s_inodes_count)) {
1735                 ext4_warning(sb, "inodes_count overflow");
1736                 return -EINVAL;
1737         }
1738
1739         if (reserved_gdb || gdb_off == 0) {
1740                 if (!ext4_has_feature_resize_inode(sb) ||
1741                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1742                         ext4_warning(sb,
1743                                      "No reserved GDT blocks, can't resize");
1744                         return -EPERM;
1745                 }
1746                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1747                 if (IS_ERR(inode)) {
1748                         ext4_warning(sb, "Error opening resize inode");
1749                         return PTR_ERR(inode);
1750                 }
1751         }
1752
1753
1754         err = verify_group_input(sb, input);
1755         if (err)
1756                 goto out;
1757
1758         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1759         if (err)
1760                 goto out;
1761
1762         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1763         if (err)
1764                 goto out;
1765
1766         flex_gd.count = 1;
1767         flex_gd.groups = input;
1768         flex_gd.bg_flags = &bg_flags;
1769         err = ext4_flex_group_add(sb, inode, &flex_gd);
1770 out:
1771         iput(inode);
1772         return err;
1773 } /* ext4_group_add */
1774
1775 /*
1776  * extend a group without checking assuming that checking has been done.
1777  */
1778 static int ext4_group_extend_no_check(struct super_block *sb,
1779                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1780 {
1781         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1782         handle_t *handle;
1783         int err = 0, err2;
1784
1785         /* We will update the superblock, one block bitmap, and
1786          * one group descriptor via ext4_group_add_blocks().
1787          */
1788         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1789         if (IS_ERR(handle)) {
1790                 err = PTR_ERR(handle);
1791                 ext4_warning(sb, "error %d on journal start", err);
1792                 return err;
1793         }
1794
1795         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1796         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1797                                             EXT4_JTR_NONE);
1798         if (err) {
1799                 ext4_warning(sb, "error %d on journal write access", err);
1800                 goto errout;
1801         }
1802
1803         lock_buffer(EXT4_SB(sb)->s_sbh);
1804         ext4_blocks_count_set(es, o_blocks_count + add);
1805         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1806         ext4_superblock_csum_set(sb);
1807         unlock_buffer(EXT4_SB(sb)->s_sbh);
1808         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1809                    o_blocks_count + add);
1810         /* We add the blocks to the bitmap and set the group need init bit */
1811         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1812         if (err)
1813                 goto errout;
1814         ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1815         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1816                    o_blocks_count + add);
1817 errout:
1818         err2 = ext4_journal_stop(handle);
1819         if (err2 && !err)
1820                 err = err2;
1821
1822         if (!err) {
1823                 if (test_opt(sb, DEBUG))
1824                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1825                                "blocks\n", ext4_blocks_count(es));
1826                 update_backups(sb, ext4_group_first_block_no(sb, 0),
1827                                (char *)es, sizeof(struct ext4_super_block), 0);
1828         }
1829         return err;
1830 }
1831
1832 /*
1833  * Extend the filesystem to the new number of blocks specified.  This entry
1834  * point is only used to extend the current filesystem to the end of the last
1835  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1836  * for emergencies (because it has no dependencies on reserved blocks).
1837  *
1838  * If we _really_ wanted, we could use default values to call ext4_group_add()
1839  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1840  * GDT blocks are reserved to grow to the desired size.
1841  */
1842 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1843                       ext4_fsblk_t n_blocks_count)
1844 {
1845         ext4_fsblk_t o_blocks_count;
1846         ext4_grpblk_t last;
1847         ext4_grpblk_t add;
1848         struct buffer_head *bh;
1849         int err;
1850         ext4_group_t group;
1851
1852         o_blocks_count = ext4_blocks_count(es);
1853
1854         if (test_opt(sb, DEBUG))
1855                 ext4_msg(sb, KERN_DEBUG,
1856                          "extending last group from %llu to %llu blocks",
1857                          o_blocks_count, n_blocks_count);
1858
1859         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1860                 return 0;
1861
1862         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1863                 ext4_msg(sb, KERN_ERR,
1864                          "filesystem too large to resize to %llu blocks safely",
1865                          n_blocks_count);
1866                 return -EINVAL;
1867         }
1868
1869         if (n_blocks_count < o_blocks_count) {
1870                 ext4_warning(sb, "can't shrink FS - resize aborted");
1871                 return -EINVAL;
1872         }
1873
1874         /* Handle the remaining blocks in the last group only. */
1875         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1876
1877         if (last == 0) {
1878                 ext4_warning(sb, "need to use ext2online to resize further");
1879                 return -EPERM;
1880         }
1881
1882         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1883
1884         if (o_blocks_count + add < o_blocks_count) {
1885                 ext4_warning(sb, "blocks_count overflow");
1886                 return -EINVAL;
1887         }
1888
1889         if (o_blocks_count + add > n_blocks_count)
1890                 add = n_blocks_count - o_blocks_count;
1891
1892         if (o_blocks_count + add < n_blocks_count)
1893                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1894                              o_blocks_count + add, add);
1895
1896         /* See if the device is actually as big as what was requested */
1897         bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1898         if (IS_ERR(bh)) {
1899                 ext4_warning(sb, "can't read last block, resize aborted");
1900                 return -ENOSPC;
1901         }
1902         brelse(bh);
1903
1904         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1905         return err;
1906 } /* ext4_group_extend */
1907
1908
1909 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1910 {
1911         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1912 }
1913
1914 /*
1915  * Release the resize inode and drop the resize_inode feature if there
1916  * are no more reserved gdt blocks, and then convert the file system
1917  * to enable meta_bg
1918  */
1919 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1920 {
1921         handle_t *handle;
1922         struct ext4_sb_info *sbi = EXT4_SB(sb);
1923         struct ext4_super_block *es = sbi->s_es;
1924         struct ext4_inode_info *ei = EXT4_I(inode);
1925         ext4_fsblk_t nr;
1926         int i, ret, err = 0;
1927         int credits = 1;
1928
1929         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1930         if (inode) {
1931                 if (es->s_reserved_gdt_blocks) {
1932                         ext4_error(sb, "Unexpected non-zero "
1933                                    "s_reserved_gdt_blocks");
1934                         return -EPERM;
1935                 }
1936
1937                 /* Do a quick sanity check of the resize inode */
1938                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1939                                              (9 - sbi->s_cluster_bits)))
1940                         goto invalid_resize_inode;
1941                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1942                         if (i == EXT4_DIND_BLOCK) {
1943                                 if (ei->i_data[i])
1944                                         continue;
1945                                 else
1946                                         goto invalid_resize_inode;
1947                         }
1948                         if (ei->i_data[i])
1949                                 goto invalid_resize_inode;
1950                 }
1951                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1952         }
1953
1954         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1955         if (IS_ERR(handle))
1956                 return PTR_ERR(handle);
1957
1958         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1959         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1960                                             EXT4_JTR_NONE);
1961         if (err)
1962                 goto errout;
1963
1964         lock_buffer(sbi->s_sbh);
1965         ext4_clear_feature_resize_inode(sb);
1966         ext4_set_feature_meta_bg(sb);
1967         sbi->s_es->s_first_meta_bg =
1968                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1969         ext4_superblock_csum_set(sb);
1970         unlock_buffer(sbi->s_sbh);
1971
1972         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1973         if (err) {
1974                 ext4_std_error(sb, err);
1975                 goto errout;
1976         }
1977
1978         if (inode) {
1979                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1980                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1981                                  EXT4_FREE_BLOCKS_METADATA |
1982                                  EXT4_FREE_BLOCKS_FORGET);
1983                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1984                 inode->i_blocks = 0;
1985
1986                 err = ext4_mark_inode_dirty(handle, inode);
1987                 if (err)
1988                         ext4_std_error(sb, err);
1989         }
1990
1991 errout:
1992         ret = ext4_journal_stop(handle);
1993         return err ? err : ret;
1994
1995 invalid_resize_inode:
1996         ext4_error(sb, "corrupted/inconsistent resize inode");
1997         return -EINVAL;
1998 }
1999
2000 /*
2001  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
2002  *
2003  * @sb: super block of the fs to be resized
2004  * @n_blocks_count: the number of blocks resides in the resized fs
2005  */
2006 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
2007 {
2008         struct ext4_new_flex_group_data *flex_gd = NULL;
2009         struct ext4_sb_info *sbi = EXT4_SB(sb);
2010         struct ext4_super_block *es = sbi->s_es;
2011         struct buffer_head *bh;
2012         struct inode *resize_inode = NULL;
2013         ext4_grpblk_t add, offset;
2014         unsigned long n_desc_blocks;
2015         unsigned long o_desc_blocks;
2016         ext4_group_t o_group;
2017         ext4_group_t n_group;
2018         ext4_fsblk_t o_blocks_count;
2019         ext4_fsblk_t n_blocks_count_retry = 0;
2020         unsigned long last_update_time = 0;
2021         int err = 0;
2022         int meta_bg;
2023         unsigned int flexbg_size = ext4_flex_bg_size(sbi);
2024
2025         /* See if the device is actually as big as what was requested */
2026         bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
2027         if (IS_ERR(bh)) {
2028                 ext4_warning(sb, "can't read last block, resize aborted");
2029                 return -ENOSPC;
2030         }
2031         brelse(bh);
2032
2033         /*
2034          * For bigalloc, trim the requested size to the nearest cluster
2035          * boundary to avoid creating an unusable filesystem. We do this
2036          * silently, instead of returning an error, to avoid breaking
2037          * callers that blindly resize the filesystem to the full size of
2038          * the underlying block device.
2039          */
2040         if (ext4_has_feature_bigalloc(sb))
2041                 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
2042
2043 retry:
2044         o_blocks_count = ext4_blocks_count(es);
2045
2046         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
2047                  "to %llu blocks", o_blocks_count, n_blocks_count);
2048
2049         if (n_blocks_count < o_blocks_count) {
2050                 /* On-line shrinking not supported */
2051                 ext4_warning(sb, "can't shrink FS - resize aborted");
2052                 return -EINVAL;
2053         }
2054
2055         if (n_blocks_count == o_blocks_count)
2056                 /* Nothing need to do */
2057                 return 0;
2058
2059         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
2060         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
2061                 ext4_warning(sb, "resize would cause inodes_count overflow");
2062                 return -EINVAL;
2063         }
2064         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2065
2066         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2067         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2068
2069         meta_bg = ext4_has_feature_meta_bg(sb);
2070
2071         if (ext4_has_feature_resize_inode(sb)) {
2072                 if (meta_bg) {
2073                         ext4_error(sb, "resize_inode and meta_bg enabled "
2074                                    "simultaneously");
2075                         return -EINVAL;
2076                 }
2077                 if (n_desc_blocks > o_desc_blocks +
2078                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
2079                         n_blocks_count_retry = n_blocks_count;
2080                         n_desc_blocks = o_desc_blocks +
2081                                 le16_to_cpu(es->s_reserved_gdt_blocks);
2082                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2083                         n_blocks_count = (ext4_fsblk_t)n_group *
2084                                 EXT4_BLOCKS_PER_GROUP(sb) +
2085                                 le32_to_cpu(es->s_first_data_block);
2086                         n_group--; /* set to last group number */
2087                 }
2088
2089                 if (!resize_inode)
2090                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2091                                                  EXT4_IGET_SPECIAL);
2092                 if (IS_ERR(resize_inode)) {
2093                         ext4_warning(sb, "Error opening resize inode");
2094                         return PTR_ERR(resize_inode);
2095                 }
2096         }
2097
2098         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2099                 err = ext4_convert_meta_bg(sb, resize_inode);
2100                 if (err)
2101                         goto out;
2102                 if (resize_inode) {
2103                         iput(resize_inode);
2104                         resize_inode = NULL;
2105                 }
2106                 if (n_blocks_count_retry) {
2107                         n_blocks_count = n_blocks_count_retry;
2108                         n_blocks_count_retry = 0;
2109                         goto retry;
2110                 }
2111         }
2112
2113         /*
2114          * Make sure the last group has enough space so that it's
2115          * guaranteed to have enough space for all metadata blocks
2116          * that it might need to hold.  (We might not need to store
2117          * the inode table blocks in the last block group, but there
2118          * will be cases where this might be needed.)
2119          */
2120         if ((ext4_group_first_block_no(sb, n_group) +
2121              ext4_group_overhead_blocks(sb, n_group) + 2 +
2122              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2123                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2124                 n_group--;
2125                 n_blocks_count_retry = 0;
2126                 if (resize_inode) {
2127                         iput(resize_inode);
2128                         resize_inode = NULL;
2129                 }
2130                 goto retry;
2131         }
2132
2133         /* extend the last group */
2134         if (n_group == o_group)
2135                 add = n_blocks_count - o_blocks_count;
2136         else
2137                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2138         if (add > 0) {
2139                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2140                 if (err)
2141                         goto out;
2142         }
2143
2144         if (ext4_blocks_count(es) == n_blocks_count && n_blocks_count_retry == 0)
2145                 goto out;
2146
2147         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2148         if (err)
2149                 goto out;
2150
2151         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2152         if (err)
2153                 goto out;
2154
2155         flex_gd = alloc_flex_gd(flexbg_size);
2156         if (flex_gd == NULL) {
2157                 err = -ENOMEM;
2158                 goto out;
2159         }
2160
2161         /* Add flex groups. Note that a regular group is a
2162          * flex group with 1 group.
2163          */
2164         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count)) {
2165                 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2166                         if (last_update_time)
2167                                 ext4_msg(sb, KERN_INFO,
2168                                          "resized to %llu blocks",
2169                                          ext4_blocks_count(es));
2170                         last_update_time = jiffies;
2171                 }
2172                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2173                         break;
2174                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2175                 if (unlikely(err))
2176                         break;
2177         }
2178
2179         if (!err && n_blocks_count_retry) {
2180                 n_blocks_count = n_blocks_count_retry;
2181                 n_blocks_count_retry = 0;
2182                 free_flex_gd(flex_gd);
2183                 flex_gd = NULL;
2184                 if (resize_inode) {
2185                         iput(resize_inode);
2186                         resize_inode = NULL;
2187                 }
2188                 goto retry;
2189         }
2190
2191 out:
2192         if (flex_gd)
2193                 free_flex_gd(flex_gd);
2194         if (resize_inode != NULL)
2195                 iput(resize_inode);
2196         if (err)
2197                 ext4_warning(sb, "error (%d) occurred during "
2198                              "file system resize", err);
2199         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2200                  ext4_blocks_count(es));
2201         return err;
2202 }