GNU Linux-libre 4.19.211-gnu1
[releases.git] / fs / ext4 / resize.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/resize.c
4  *
5  * Support for resizing an ext4 filesystem while it is mounted.
6  *
7  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
8  *
9  * This could probably be made into a module, because it is not often in use.
10  */
11
12
13 #define EXT4FS_DEBUG
14
15 #include <linux/errno.h>
16 #include <linux/slab.h>
17
18 #include "ext4_jbd2.h"
19
20 struct ext4_rcu_ptr {
21         struct rcu_head rcu;
22         void *ptr;
23 };
24
25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
26 {
27         struct ext4_rcu_ptr *ptr;
28
29         ptr = container_of(head, struct ext4_rcu_ptr, rcu);
30         kvfree(ptr->ptr);
31         kfree(ptr);
32 }
33
34 void ext4_kvfree_array_rcu(void *to_free)
35 {
36         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
37
38         if (ptr) {
39                 ptr->ptr = to_free;
40                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
41                 return;
42         }
43         synchronize_rcu();
44         kvfree(to_free);
45 }
46
47 int ext4_resize_begin(struct super_block *sb)
48 {
49         struct ext4_sb_info *sbi = EXT4_SB(sb);
50         int ret = 0;
51
52         if (!capable(CAP_SYS_RESOURCE))
53                 return -EPERM;
54
55         /*
56          * If we are not using the primary superblock/GDT copy don't resize,
57          * because the user tools have no way of handling this.  Probably a
58          * bad time to do it anyways.
59          */
60         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
61             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
62                 ext4_warning(sb, "won't resize using backup superblock at %llu",
63                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
64                 return -EPERM;
65         }
66
67         /*
68          * We are not allowed to do online-resizing on a filesystem mounted
69          * with error, because it can destroy the filesystem easily.
70          */
71         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
72                 ext4_warning(sb, "There are errors in the filesystem, "
73                              "so online resizing is not allowed");
74                 return -EPERM;
75         }
76
77         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
78                                   &EXT4_SB(sb)->s_ext4_flags))
79                 ret = -EBUSY;
80
81         return ret;
82 }
83
84 void ext4_resize_end(struct super_block *sb)
85 {
86         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
87         smp_mb__after_atomic();
88 }
89
90 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
91                                              ext4_group_t group) {
92         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
93                EXT4_DESC_PER_BLOCK_BITS(sb);
94 }
95
96 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
97                                              ext4_group_t group) {
98         group = ext4_meta_bg_first_group(sb, group);
99         return ext4_group_first_block_no(sb, group);
100 }
101
102 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
103                                                 ext4_group_t group) {
104         ext4_grpblk_t overhead;
105         overhead = ext4_bg_num_gdb(sb, group);
106         if (ext4_bg_has_super(sb, group))
107                 overhead += 1 +
108                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
109         return overhead;
110 }
111
112 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
113 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
114
115 static int verify_group_input(struct super_block *sb,
116                               struct ext4_new_group_data *input)
117 {
118         struct ext4_sb_info *sbi = EXT4_SB(sb);
119         struct ext4_super_block *es = sbi->s_es;
120         ext4_fsblk_t start = ext4_blocks_count(es);
121         ext4_fsblk_t end = start + input->blocks_count;
122         ext4_group_t group = input->group;
123         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
124         unsigned overhead;
125         ext4_fsblk_t metaend;
126         struct buffer_head *bh = NULL;
127         ext4_grpblk_t free_blocks_count, offset;
128         int err = -EINVAL;
129
130         if (group != sbi->s_groups_count) {
131                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
132                              input->group, sbi->s_groups_count);
133                 return -EINVAL;
134         }
135
136         overhead = ext4_group_overhead_blocks(sb, group);
137         metaend = start + overhead;
138         input->free_clusters_count = free_blocks_count =
139                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
140
141         if (test_opt(sb, DEBUG))
142                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
143                        "(%d free, %u reserved)\n",
144                        ext4_bg_has_super(sb, input->group) ? "normal" :
145                        "no-super", input->group, input->blocks_count,
146                        free_blocks_count, input->reserved_blocks);
147
148         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
149         if (offset != 0)
150                         ext4_warning(sb, "Last group not full");
151         else if (input->reserved_blocks > input->blocks_count / 5)
152                 ext4_warning(sb, "Reserved blocks too high (%u)",
153                              input->reserved_blocks);
154         else if (free_blocks_count < 0)
155                 ext4_warning(sb, "Bad blocks count %u",
156                              input->blocks_count);
157         else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
158                 err = PTR_ERR(bh);
159                 bh = NULL;
160                 ext4_warning(sb, "Cannot read last block (%llu)",
161                              end - 1);
162         } else if (outside(input->block_bitmap, start, end))
163                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
164                              (unsigned long long)input->block_bitmap);
165         else if (outside(input->inode_bitmap, start, end))
166                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
167                              (unsigned long long)input->inode_bitmap);
168         else if (outside(input->inode_table, start, end) ||
169                  outside(itend - 1, start, end))
170                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
171                              (unsigned long long)input->inode_table, itend - 1);
172         else if (input->inode_bitmap == input->block_bitmap)
173                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
174                              (unsigned long long)input->block_bitmap);
175         else if (inside(input->block_bitmap, input->inode_table, itend))
176                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
177                              "(%llu-%llu)",
178                              (unsigned long long)input->block_bitmap,
179                              (unsigned long long)input->inode_table, itend - 1);
180         else if (inside(input->inode_bitmap, input->inode_table, itend))
181                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
182                              "(%llu-%llu)",
183                              (unsigned long long)input->inode_bitmap,
184                              (unsigned long long)input->inode_table, itend - 1);
185         else if (inside(input->block_bitmap, start, metaend))
186                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
187                              (unsigned long long)input->block_bitmap,
188                              start, metaend - 1);
189         else if (inside(input->inode_bitmap, start, metaend))
190                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
191                              (unsigned long long)input->inode_bitmap,
192                              start, metaend - 1);
193         else if (inside(input->inode_table, start, metaend) ||
194                  inside(itend - 1, start, metaend))
195                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
196                              "(%llu-%llu)",
197                              (unsigned long long)input->inode_table,
198                              itend - 1, start, metaend - 1);
199         else
200                 err = 0;
201         brelse(bh);
202
203         return err;
204 }
205
206 /*
207  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
208  * group each time.
209  */
210 struct ext4_new_flex_group_data {
211         struct ext4_new_group_data *groups;     /* new_group_data for groups
212                                                    in the flex group */
213         __u16 *bg_flags;                        /* block group flags of groups
214                                                    in @groups */
215         ext4_group_t count;                     /* number of groups in @groups
216                                                  */
217 };
218
219 /*
220  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
221  * @flexbg_size.
222  *
223  * Returns NULL on failure otherwise address of the allocated structure.
224  */
225 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
226 {
227         struct ext4_new_flex_group_data *flex_gd;
228
229         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
230         if (flex_gd == NULL)
231                 goto out3;
232
233         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
234                 goto out2;
235         flex_gd->count = flexbg_size;
236
237         flex_gd->groups = kmalloc_array(flexbg_size,
238                                         sizeof(struct ext4_new_group_data),
239                                         GFP_NOFS);
240         if (flex_gd->groups == NULL)
241                 goto out2;
242
243         flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
244                                           GFP_NOFS);
245         if (flex_gd->bg_flags == NULL)
246                 goto out1;
247
248         return flex_gd;
249
250 out1:
251         kfree(flex_gd->groups);
252 out2:
253         kfree(flex_gd);
254 out3:
255         return NULL;
256 }
257
258 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
259 {
260         kfree(flex_gd->bg_flags);
261         kfree(flex_gd->groups);
262         kfree(flex_gd);
263 }
264
265 /*
266  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
267  * and inode tables for a flex group.
268  *
269  * This function is used by 64bit-resize.  Note that this function allocates
270  * group tables from the 1st group of groups contained by @flexgd, which may
271  * be a partial of a flex group.
272  *
273  * @sb: super block of fs to which the groups belongs
274  *
275  * Returns 0 on a successful allocation of the metadata blocks in the
276  * block group.
277  */
278 static int ext4_alloc_group_tables(struct super_block *sb,
279                                 struct ext4_new_flex_group_data *flex_gd,
280                                 int flexbg_size)
281 {
282         struct ext4_new_group_data *group_data = flex_gd->groups;
283         ext4_fsblk_t start_blk;
284         ext4_fsblk_t last_blk;
285         ext4_group_t src_group;
286         ext4_group_t bb_index = 0;
287         ext4_group_t ib_index = 0;
288         ext4_group_t it_index = 0;
289         ext4_group_t group;
290         ext4_group_t last_group;
291         unsigned overhead;
292         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
293         int i;
294
295         BUG_ON(flex_gd->count == 0 || group_data == NULL);
296
297         src_group = group_data[0].group;
298         last_group  = src_group + flex_gd->count - 1;
299
300         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
301                (last_group & ~(flexbg_size - 1))));
302 next_group:
303         group = group_data[0].group;
304         if (src_group >= group_data[0].group + flex_gd->count)
305                 return -ENOSPC;
306         start_blk = ext4_group_first_block_no(sb, src_group);
307         last_blk = start_blk + group_data[src_group - group].blocks_count;
308
309         overhead = ext4_group_overhead_blocks(sb, src_group);
310
311         start_blk += overhead;
312
313         /* We collect contiguous blocks as much as possible. */
314         src_group++;
315         for (; src_group <= last_group; src_group++) {
316                 overhead = ext4_group_overhead_blocks(sb, src_group);
317                 if (overhead == 0)
318                         last_blk += group_data[src_group - group].blocks_count;
319                 else
320                         break;
321         }
322
323         /* Allocate block bitmaps */
324         for (; bb_index < flex_gd->count; bb_index++) {
325                 if (start_blk >= last_blk)
326                         goto next_group;
327                 group_data[bb_index].block_bitmap = start_blk++;
328                 group = ext4_get_group_number(sb, start_blk - 1);
329                 group -= group_data[0].group;
330                 group_data[group].mdata_blocks++;
331                 flex_gd->bg_flags[group] &= uninit_mask;
332         }
333
334         /* Allocate inode bitmaps */
335         for (; ib_index < flex_gd->count; ib_index++) {
336                 if (start_blk >= last_blk)
337                         goto next_group;
338                 group_data[ib_index].inode_bitmap = start_blk++;
339                 group = ext4_get_group_number(sb, start_blk - 1);
340                 group -= group_data[0].group;
341                 group_data[group].mdata_blocks++;
342                 flex_gd->bg_flags[group] &= uninit_mask;
343         }
344
345         /* Allocate inode tables */
346         for (; it_index < flex_gd->count; it_index++) {
347                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
348                 ext4_fsblk_t next_group_start;
349
350                 if (start_blk + itb > last_blk)
351                         goto next_group;
352                 group_data[it_index].inode_table = start_blk;
353                 group = ext4_get_group_number(sb, start_blk);
354                 next_group_start = ext4_group_first_block_no(sb, group + 1);
355                 group -= group_data[0].group;
356
357                 if (start_blk + itb > next_group_start) {
358                         flex_gd->bg_flags[group + 1] &= uninit_mask;
359                         overhead = start_blk + itb - next_group_start;
360                         group_data[group + 1].mdata_blocks += overhead;
361                         itb -= overhead;
362                 }
363
364                 group_data[group].mdata_blocks += itb;
365                 flex_gd->bg_flags[group] &= uninit_mask;
366                 start_blk += EXT4_SB(sb)->s_itb_per_group;
367         }
368
369         /* Update free clusters count to exclude metadata blocks */
370         for (i = 0; i < flex_gd->count; i++) {
371                 group_data[i].free_clusters_count -=
372                                 EXT4_NUM_B2C(EXT4_SB(sb),
373                                              group_data[i].mdata_blocks);
374         }
375
376         if (test_opt(sb, DEBUG)) {
377                 int i;
378                 group = group_data[0].group;
379
380                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
381                        "%d groups, flexbg size is %d:\n", flex_gd->count,
382                        flexbg_size);
383
384                 for (i = 0; i < flex_gd->count; i++) {
385                         ext4_debug(
386                                "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
387                                ext4_bg_has_super(sb, group + i) ? "normal" :
388                                "no-super", group + i,
389                                group_data[i].blocks_count,
390                                group_data[i].free_clusters_count,
391                                group_data[i].mdata_blocks);
392                 }
393         }
394         return 0;
395 }
396
397 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
398                                   ext4_fsblk_t blk)
399 {
400         struct buffer_head *bh;
401         int err;
402
403         bh = sb_getblk(sb, blk);
404         if (unlikely(!bh))
405                 return ERR_PTR(-ENOMEM);
406         BUFFER_TRACE(bh, "get_write_access");
407         if ((err = ext4_journal_get_write_access(handle, bh))) {
408                 brelse(bh);
409                 bh = ERR_PTR(err);
410         } else {
411                 memset(bh->b_data, 0, sb->s_blocksize);
412                 set_buffer_uptodate(bh);
413         }
414
415         return bh;
416 }
417
418 /*
419  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
420  * If that fails, restart the transaction & regain write access for the
421  * buffer head which is used for block_bitmap modifications.
422  */
423 static int extend_or_restart_transaction(handle_t *handle, int thresh)
424 {
425         int err;
426
427         if (ext4_handle_has_enough_credits(handle, thresh))
428                 return 0;
429
430         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
431         if (err < 0)
432                 return err;
433         if (err) {
434                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
435                 if (err)
436                         return err;
437         }
438
439         return 0;
440 }
441
442 /*
443  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
444  *
445  * Helper function for ext4_setup_new_group_blocks() which set .
446  *
447  * @sb: super block
448  * @handle: journal handle
449  * @flex_gd: flex group data
450  */
451 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
452                         struct ext4_new_flex_group_data *flex_gd,
453                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
454 {
455         struct ext4_sb_info *sbi = EXT4_SB(sb);
456         ext4_group_t count = last_cluster - first_cluster + 1;
457         ext4_group_t count2;
458
459         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
460                    last_cluster);
461         for (count2 = count; count > 0;
462              count -= count2, first_cluster += count2) {
463                 ext4_fsblk_t start;
464                 struct buffer_head *bh;
465                 ext4_group_t group;
466                 int err;
467
468                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
469                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
470                 group -= flex_gd->groups[0].group;
471
472                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
473                 if (count2 > count)
474                         count2 = count;
475
476                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
477                         BUG_ON(flex_gd->count > 1);
478                         continue;
479                 }
480
481                 err = extend_or_restart_transaction(handle, 1);
482                 if (err)
483                         return err;
484
485                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
486                 if (unlikely(!bh))
487                         return -ENOMEM;
488
489                 BUFFER_TRACE(bh, "get_write_access");
490                 err = ext4_journal_get_write_access(handle, bh);
491                 if (err) {
492                         brelse(bh);
493                         return err;
494                 }
495                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
496                            first_cluster, first_cluster - start, count2);
497                 ext4_set_bits(bh->b_data, first_cluster - start, count2);
498
499                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
500                 brelse(bh);
501                 if (unlikely(err))
502                         return err;
503         }
504
505         return 0;
506 }
507
508 /*
509  * Set up the block and inode bitmaps, and the inode table for the new groups.
510  * This doesn't need to be part of the main transaction, since we are only
511  * changing blocks outside the actual filesystem.  We still do journaling to
512  * ensure the recovery is correct in case of a failure just after resize.
513  * If any part of this fails, we simply abort the resize.
514  *
515  * setup_new_flex_group_blocks handles a flex group as follow:
516  *  1. copy super block and GDT, and initialize group tables if necessary.
517  *     In this step, we only set bits in blocks bitmaps for blocks taken by
518  *     super block and GDT.
519  *  2. allocate group tables in block bitmaps, that is, set bits in block
520  *     bitmap for blocks taken by group tables.
521  */
522 static int setup_new_flex_group_blocks(struct super_block *sb,
523                                 struct ext4_new_flex_group_data *flex_gd)
524 {
525         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
526         ext4_fsblk_t start;
527         ext4_fsblk_t block;
528         struct ext4_sb_info *sbi = EXT4_SB(sb);
529         struct ext4_super_block *es = sbi->s_es;
530         struct ext4_new_group_data *group_data = flex_gd->groups;
531         __u16 *bg_flags = flex_gd->bg_flags;
532         handle_t *handle;
533         ext4_group_t group, count;
534         struct buffer_head *bh = NULL;
535         int reserved_gdb, i, j, err = 0, err2;
536         int meta_bg;
537
538         BUG_ON(!flex_gd->count || !group_data ||
539                group_data[0].group != sbi->s_groups_count);
540
541         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
542         meta_bg = ext4_has_feature_meta_bg(sb);
543
544         /* This transaction may be extended/restarted along the way */
545         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
546         if (IS_ERR(handle))
547                 return PTR_ERR(handle);
548
549         group = group_data[0].group;
550         for (i = 0; i < flex_gd->count; i++, group++) {
551                 unsigned long gdblocks;
552                 ext4_grpblk_t overhead;
553
554                 gdblocks = ext4_bg_num_gdb(sb, group);
555                 start = ext4_group_first_block_no(sb, group);
556
557                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
558                         goto handle_itb;
559
560                 if (meta_bg == 1) {
561                         ext4_group_t first_group;
562                         first_group = ext4_meta_bg_first_group(sb, group);
563                         if (first_group != group + 1 &&
564                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
565                                 goto handle_itb;
566                 }
567
568                 block = start + ext4_bg_has_super(sb, group);
569                 /* Copy all of the GDT blocks into the backup in this group */
570                 for (j = 0; j < gdblocks; j++, block++) {
571                         struct buffer_head *gdb;
572
573                         ext4_debug("update backup group %#04llx\n", block);
574                         err = extend_or_restart_transaction(handle, 1);
575                         if (err)
576                                 goto out;
577
578                         gdb = sb_getblk(sb, block);
579                         if (unlikely(!gdb)) {
580                                 err = -ENOMEM;
581                                 goto out;
582                         }
583
584                         BUFFER_TRACE(gdb, "get_write_access");
585                         err = ext4_journal_get_write_access(handle, gdb);
586                         if (err) {
587                                 brelse(gdb);
588                                 goto out;
589                         }
590                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
591                                 s_group_desc, j)->b_data, gdb->b_size);
592                         set_buffer_uptodate(gdb);
593
594                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
595                         if (unlikely(err)) {
596                                 brelse(gdb);
597                                 goto out;
598                         }
599                         brelse(gdb);
600                 }
601
602                 /* Zero out all of the reserved backup group descriptor
603                  * table blocks
604                  */
605                 if (ext4_bg_has_super(sb, group)) {
606                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
607                                         reserved_gdb, GFP_NOFS);
608                         if (err)
609                                 goto out;
610                 }
611
612 handle_itb:
613                 /* Initialize group tables of the grop @group */
614                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
615                         goto handle_bb;
616
617                 /* Zero out all of the inode table blocks */
618                 block = group_data[i].inode_table;
619                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
620                            block, sbi->s_itb_per_group);
621                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
622                                        GFP_NOFS);
623                 if (err)
624                         goto out;
625
626 handle_bb:
627                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
628                         goto handle_ib;
629
630                 /* Initialize block bitmap of the @group */
631                 block = group_data[i].block_bitmap;
632                 err = extend_or_restart_transaction(handle, 1);
633                 if (err)
634                         goto out;
635
636                 bh = bclean(handle, sb, block);
637                 if (IS_ERR(bh)) {
638                         err = PTR_ERR(bh);
639                         goto out;
640                 }
641                 overhead = ext4_group_overhead_blocks(sb, group);
642                 if (overhead != 0) {
643                         ext4_debug("mark backup superblock %#04llx (+0)\n",
644                                    start);
645                         ext4_set_bits(bh->b_data, 0,
646                                       EXT4_NUM_B2C(sbi, overhead));
647                 }
648                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
649                                      sb->s_blocksize * 8, bh->b_data);
650                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
651                 brelse(bh);
652                 if (err)
653                         goto out;
654
655 handle_ib:
656                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
657                         continue;
658
659                 /* Initialize inode bitmap of the @group */
660                 block = group_data[i].inode_bitmap;
661                 err = extend_or_restart_transaction(handle, 1);
662                 if (err)
663                         goto out;
664                 /* Mark unused entries in inode bitmap used */
665                 bh = bclean(handle, sb, block);
666                 if (IS_ERR(bh)) {
667                         err = PTR_ERR(bh);
668                         goto out;
669                 }
670
671                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
672                                      sb->s_blocksize * 8, bh->b_data);
673                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
674                 brelse(bh);
675                 if (err)
676                         goto out;
677         }
678
679         /* Mark group tables in block bitmap */
680         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
681                 count = group_table_count[j];
682                 start = (&group_data[0].block_bitmap)[j];
683                 block = start;
684                 for (i = 1; i < flex_gd->count; i++) {
685                         block += group_table_count[j];
686                         if (block == (&group_data[i].block_bitmap)[j]) {
687                                 count += group_table_count[j];
688                                 continue;
689                         }
690                         err = set_flexbg_block_bitmap(sb, handle,
691                                                       flex_gd,
692                                                       EXT4_B2C(sbi, start),
693                                                       EXT4_B2C(sbi,
694                                                                start + count
695                                                                - 1));
696                         if (err)
697                                 goto out;
698                         count = group_table_count[j];
699                         start = (&group_data[i].block_bitmap)[j];
700                         block = start;
701                 }
702
703                 if (count) {
704                         err = set_flexbg_block_bitmap(sb, handle,
705                                                       flex_gd,
706                                                       EXT4_B2C(sbi, start),
707                                                       EXT4_B2C(sbi,
708                                                                start + count
709                                                                - 1));
710                         if (err)
711                                 goto out;
712                 }
713         }
714
715 out:
716         err2 = ext4_journal_stop(handle);
717         if (err2 && !err)
718                 err = err2;
719
720         return err;
721 }
722
723 /*
724  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
725  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
726  * calling this for the first time.  In a sparse filesystem it will be the
727  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
728  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
729  */
730 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
731                                   unsigned *five, unsigned *seven)
732 {
733         unsigned *min = three;
734         int mult = 3;
735         unsigned ret;
736
737         if (!ext4_has_feature_sparse_super(sb)) {
738                 ret = *min;
739                 *min += 1;
740                 return ret;
741         }
742
743         if (*five < *min) {
744                 min = five;
745                 mult = 5;
746         }
747         if (*seven < *min) {
748                 min = seven;
749                 mult = 7;
750         }
751
752         ret = *min;
753         *min *= mult;
754
755         return ret;
756 }
757
758 /*
759  * Check that all of the backup GDT blocks are held in the primary GDT block.
760  * It is assumed that they are stored in group order.  Returns the number of
761  * groups in current filesystem that have BACKUPS, or -ve error code.
762  */
763 static int verify_reserved_gdb(struct super_block *sb,
764                                ext4_group_t end,
765                                struct buffer_head *primary)
766 {
767         const ext4_fsblk_t blk = primary->b_blocknr;
768         unsigned three = 1;
769         unsigned five = 5;
770         unsigned seven = 7;
771         unsigned grp;
772         __le32 *p = (__le32 *)primary->b_data;
773         int gdbackups = 0;
774
775         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
776                 if (le32_to_cpu(*p++) !=
777                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
778                         ext4_warning(sb, "reserved GDT %llu"
779                                      " missing grp %d (%llu)",
780                                      blk, grp,
781                                      grp *
782                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
783                                      blk);
784                         return -EINVAL;
785                 }
786                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
787                         return -EFBIG;
788         }
789
790         return gdbackups;
791 }
792
793 /*
794  * Called when we need to bring a reserved group descriptor table block into
795  * use from the resize inode.  The primary copy of the new GDT block currently
796  * is an indirect block (under the double indirect block in the resize inode).
797  * The new backup GDT blocks will be stored as leaf blocks in this indirect
798  * block, in group order.  Even though we know all the block numbers we need,
799  * we check to ensure that the resize inode has actually reserved these blocks.
800  *
801  * Don't need to update the block bitmaps because the blocks are still in use.
802  *
803  * We get all of the error cases out of the way, so that we are sure to not
804  * fail once we start modifying the data on disk, because JBD has no rollback.
805  */
806 static int add_new_gdb(handle_t *handle, struct inode *inode,
807                        ext4_group_t group)
808 {
809         struct super_block *sb = inode->i_sb;
810         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
811         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
812         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
813         struct buffer_head **o_group_desc, **n_group_desc = NULL;
814         struct buffer_head *dind = NULL;
815         struct buffer_head *gdb_bh = NULL;
816         int gdbackups;
817         struct ext4_iloc iloc = { .bh = NULL };
818         __le32 *data;
819         int err;
820
821         if (test_opt(sb, DEBUG))
822                 printk(KERN_DEBUG
823                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
824                        gdb_num);
825
826         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
827         if (IS_ERR(gdb_bh))
828                 return PTR_ERR(gdb_bh);
829
830         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
831         if (gdbackups < 0) {
832                 err = gdbackups;
833                 goto errout;
834         }
835
836         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
837         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
838         if (IS_ERR(dind)) {
839                 err = PTR_ERR(dind);
840                 dind = NULL;
841                 goto errout;
842         }
843
844         data = (__le32 *)dind->b_data;
845         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
846                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
847                              group, gdblock);
848                 err = -EINVAL;
849                 goto errout;
850         }
851
852         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
853         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
854         if (unlikely(err))
855                 goto errout;
856
857         BUFFER_TRACE(gdb_bh, "get_write_access");
858         err = ext4_journal_get_write_access(handle, gdb_bh);
859         if (unlikely(err))
860                 goto errout;
861
862         BUFFER_TRACE(dind, "get_write_access");
863         err = ext4_journal_get_write_access(handle, dind);
864         if (unlikely(err)) {
865                 ext4_std_error(sb, err);
866                 goto errout;
867         }
868
869         /* ext4_reserve_inode_write() gets a reference on the iloc */
870         err = ext4_reserve_inode_write(handle, inode, &iloc);
871         if (unlikely(err))
872                 goto errout;
873
874         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
875                                      sizeof(struct buffer_head *),
876                                      GFP_NOFS);
877         if (!n_group_desc) {
878                 err = -ENOMEM;
879                 ext4_warning(sb, "not enough memory for %lu groups",
880                              gdb_num + 1);
881                 goto errout;
882         }
883
884         /*
885          * Finally, we have all of the possible failures behind us...
886          *
887          * Remove new GDT block from inode double-indirect block and clear out
888          * the new GDT block for use (which also "frees" the backup GDT blocks
889          * from the reserved inode).  We don't need to change the bitmaps for
890          * these blocks, because they are marked as in-use from being in the
891          * reserved inode, and will become GDT blocks (primary and backup).
892          */
893         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
894         err = ext4_handle_dirty_metadata(handle, NULL, dind);
895         if (unlikely(err)) {
896                 ext4_std_error(sb, err);
897                 goto errout;
898         }
899         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
900                            (9 - EXT4_SB(sb)->s_cluster_bits);
901         ext4_mark_iloc_dirty(handle, inode, &iloc);
902         memset(gdb_bh->b_data, 0, sb->s_blocksize);
903         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
904         if (unlikely(err)) {
905                 ext4_std_error(sb, err);
906                 iloc.bh = NULL;
907                 goto errout;
908         }
909         brelse(dind);
910
911         rcu_read_lock();
912         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
913         memcpy(n_group_desc, o_group_desc,
914                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
915         rcu_read_unlock();
916         n_group_desc[gdb_num] = gdb_bh;
917         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
918         EXT4_SB(sb)->s_gdb_count++;
919         ext4_kvfree_array_rcu(o_group_desc);
920
921         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
922         err = ext4_handle_dirty_super(handle, sb);
923         if (err)
924                 ext4_std_error(sb, err);
925         return err;
926 errout:
927         kvfree(n_group_desc);
928         brelse(iloc.bh);
929         brelse(dind);
930         brelse(gdb_bh);
931
932         ext4_debug("leaving with error %d\n", err);
933         return err;
934 }
935
936 /*
937  * add_new_gdb_meta_bg is the sister of add_new_gdb.
938  */
939 static int add_new_gdb_meta_bg(struct super_block *sb,
940                                handle_t *handle, ext4_group_t group) {
941         ext4_fsblk_t gdblock;
942         struct buffer_head *gdb_bh;
943         struct buffer_head **o_group_desc, **n_group_desc;
944         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
945         int err;
946
947         gdblock = ext4_meta_bg_first_block_no(sb, group) +
948                    ext4_bg_has_super(sb, group);
949         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
950         if (IS_ERR(gdb_bh))
951                 return PTR_ERR(gdb_bh);
952         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
953                                      sizeof(struct buffer_head *),
954                                      GFP_NOFS);
955         if (!n_group_desc) {
956                 brelse(gdb_bh);
957                 err = -ENOMEM;
958                 ext4_warning(sb, "not enough memory for %lu groups",
959                              gdb_num + 1);
960                 return err;
961         }
962
963         rcu_read_lock();
964         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
965         memcpy(n_group_desc, o_group_desc,
966                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
967         rcu_read_unlock();
968         n_group_desc[gdb_num] = gdb_bh;
969
970         BUFFER_TRACE(gdb_bh, "get_write_access");
971         err = ext4_journal_get_write_access(handle, gdb_bh);
972         if (err) {
973                 kvfree(n_group_desc);
974                 brelse(gdb_bh);
975                 return err;
976         }
977
978         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
979         EXT4_SB(sb)->s_gdb_count++;
980         ext4_kvfree_array_rcu(o_group_desc);
981         return err;
982 }
983
984 /*
985  * Called when we are adding a new group which has a backup copy of each of
986  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
987  * We need to add these reserved backup GDT blocks to the resize inode, so
988  * that they are kept for future resizing and not allocated to files.
989  *
990  * Each reserved backup GDT block will go into a different indirect block.
991  * The indirect blocks are actually the primary reserved GDT blocks,
992  * so we know in advance what their block numbers are.  We only get the
993  * double-indirect block to verify it is pointing to the primary reserved
994  * GDT blocks so we don't overwrite a data block by accident.  The reserved
995  * backup GDT blocks are stored in their reserved primary GDT block.
996  */
997 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
998                               ext4_group_t group)
999 {
1000         struct super_block *sb = inode->i_sb;
1001         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1002         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1003         struct buffer_head **primary;
1004         struct buffer_head *dind;
1005         struct ext4_iloc iloc;
1006         ext4_fsblk_t blk;
1007         __le32 *data, *end;
1008         int gdbackups = 0;
1009         int res, i;
1010         int err;
1011
1012         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1013         if (!primary)
1014                 return -ENOMEM;
1015
1016         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1017         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1018         if (IS_ERR(dind)) {
1019                 err = PTR_ERR(dind);
1020                 dind = NULL;
1021                 goto exit_free;
1022         }
1023
1024         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1025         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1026                                          EXT4_ADDR_PER_BLOCK(sb));
1027         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1028
1029         /* Get each reserved primary GDT block and verify it holds backups */
1030         for (res = 0; res < reserved_gdb; res++, blk++) {
1031                 if (le32_to_cpu(*data) != blk) {
1032                         ext4_warning(sb, "reserved block %llu"
1033                                      " not at offset %ld",
1034                                      blk,
1035                                      (long)(data - (__le32 *)dind->b_data));
1036                         err = -EINVAL;
1037                         goto exit_bh;
1038                 }
1039                 primary[res] = ext4_sb_bread(sb, blk, 0);
1040                 if (IS_ERR(primary[res])) {
1041                         err = PTR_ERR(primary[res]);
1042                         primary[res] = NULL;
1043                         goto exit_bh;
1044                 }
1045                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1046                 if (gdbackups < 0) {
1047                         brelse(primary[res]);
1048                         err = gdbackups;
1049                         goto exit_bh;
1050                 }
1051                 if (++data >= end)
1052                         data = (__le32 *)dind->b_data;
1053         }
1054
1055         for (i = 0; i < reserved_gdb; i++) {
1056                 BUFFER_TRACE(primary[i], "get_write_access");
1057                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1058                         goto exit_bh;
1059         }
1060
1061         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1062                 goto exit_bh;
1063
1064         /*
1065          * Finally we can add each of the reserved backup GDT blocks from
1066          * the new group to its reserved primary GDT block.
1067          */
1068         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1069         for (i = 0; i < reserved_gdb; i++) {
1070                 int err2;
1071                 data = (__le32 *)primary[i]->b_data;
1072                 /* printk("reserving backup %lu[%u] = %lu\n",
1073                        primary[i]->b_blocknr, gdbackups,
1074                        blk + primary[i]->b_blocknr); */
1075                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1076                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1077                 if (!err)
1078                         err = err2;
1079         }
1080
1081         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1082         ext4_mark_iloc_dirty(handle, inode, &iloc);
1083
1084 exit_bh:
1085         while (--res >= 0)
1086                 brelse(primary[res]);
1087         brelse(dind);
1088
1089 exit_free:
1090         kfree(primary);
1091
1092         return err;
1093 }
1094
1095 /*
1096  * Update the backup copies of the ext4 metadata.  These don't need to be part
1097  * of the main resize transaction, because e2fsck will re-write them if there
1098  * is a problem (basically only OOM will cause a problem).  However, we
1099  * _should_ update the backups if possible, in case the primary gets trashed
1100  * for some reason and we need to run e2fsck from a backup superblock.  The
1101  * important part is that the new block and inode counts are in the backup
1102  * superblocks, and the location of the new group metadata in the GDT backups.
1103  *
1104  * We do not need take the s_resize_lock for this, because these
1105  * blocks are not otherwise touched by the filesystem code when it is
1106  * mounted.  We don't need to worry about last changing from
1107  * sbi->s_groups_count, because the worst that can happen is that we
1108  * do not copy the full number of backups at this time.  The resize
1109  * which changed s_groups_count will backup again.
1110  */
1111 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1112                            int size, int meta_bg)
1113 {
1114         struct ext4_sb_info *sbi = EXT4_SB(sb);
1115         ext4_group_t last;
1116         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1117         unsigned three = 1;
1118         unsigned five = 5;
1119         unsigned seven = 7;
1120         ext4_group_t group = 0;
1121         int rest = sb->s_blocksize - size;
1122         handle_t *handle;
1123         int err = 0, err2;
1124
1125         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1126         if (IS_ERR(handle)) {
1127                 group = 1;
1128                 err = PTR_ERR(handle);
1129                 goto exit_err;
1130         }
1131
1132         if (meta_bg == 0) {
1133                 group = ext4_list_backups(sb, &three, &five, &seven);
1134                 last = sbi->s_groups_count;
1135         } else {
1136                 group = ext4_get_group_number(sb, blk_off) + 1;
1137                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1138         }
1139
1140         while (group < sbi->s_groups_count) {
1141                 struct buffer_head *bh;
1142                 ext4_fsblk_t backup_block;
1143
1144                 /* Out of journal space, and can't get more - abort - so sad */
1145                 if (ext4_handle_valid(handle) &&
1146                     handle->h_buffer_credits == 0 &&
1147                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1148                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1149                         break;
1150
1151                 if (meta_bg == 0)
1152                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1153                 else
1154                         backup_block = (ext4_group_first_block_no(sb, group) +
1155                                         ext4_bg_has_super(sb, group));
1156
1157                 bh = sb_getblk(sb, backup_block);
1158                 if (unlikely(!bh)) {
1159                         err = -ENOMEM;
1160                         break;
1161                 }
1162                 ext4_debug("update metadata backup %llu(+%llu)\n",
1163                            backup_block, backup_block -
1164                            ext4_group_first_block_no(sb, group));
1165                 BUFFER_TRACE(bh, "get_write_access");
1166                 if ((err = ext4_journal_get_write_access(handle, bh))) {
1167                         brelse(bh);
1168                         break;
1169                 }
1170                 lock_buffer(bh);
1171                 memcpy(bh->b_data, data, size);
1172                 if (rest)
1173                         memset(bh->b_data + size, 0, rest);
1174                 set_buffer_uptodate(bh);
1175                 unlock_buffer(bh);
1176                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1177                 if (unlikely(err))
1178                         ext4_std_error(sb, err);
1179                 brelse(bh);
1180
1181                 if (meta_bg == 0)
1182                         group = ext4_list_backups(sb, &three, &five, &seven);
1183                 else if (group == last)
1184                         break;
1185                 else
1186                         group = last;
1187         }
1188         if ((err2 = ext4_journal_stop(handle)) && !err)
1189                 err = err2;
1190
1191         /*
1192          * Ugh! Need to have e2fsck write the backup copies.  It is too
1193          * late to revert the resize, we shouldn't fail just because of
1194          * the backup copies (they are only needed in case of corruption).
1195          *
1196          * However, if we got here we have a journal problem too, so we
1197          * can't really start a transaction to mark the superblock.
1198          * Chicken out and just set the flag on the hope it will be written
1199          * to disk, and if not - we will simply wait until next fsck.
1200          */
1201 exit_err:
1202         if (err) {
1203                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1204                              "forcing fsck on next reboot", group, err);
1205                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1206                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1207                 mark_buffer_dirty(sbi->s_sbh);
1208         }
1209 }
1210
1211 /*
1212  * ext4_add_new_descs() adds @count group descriptor of groups
1213  * starting at @group
1214  *
1215  * @handle: journal handle
1216  * @sb: super block
1217  * @group: the group no. of the first group desc to be added
1218  * @resize_inode: the resize inode
1219  * @count: number of group descriptors to be added
1220  */
1221 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1222                               ext4_group_t group, struct inode *resize_inode,
1223                               ext4_group_t count)
1224 {
1225         struct ext4_sb_info *sbi = EXT4_SB(sb);
1226         struct ext4_super_block *es = sbi->s_es;
1227         struct buffer_head *gdb_bh;
1228         int i, gdb_off, gdb_num, err = 0;
1229         int meta_bg;
1230
1231         meta_bg = ext4_has_feature_meta_bg(sb);
1232         for (i = 0; i < count; i++, group++) {
1233                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1234                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1235
1236                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1237                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1238
1239                 /*
1240                  * We will only either add reserved group blocks to a backup group
1241                  * or remove reserved blocks for the first group in a new group block.
1242                  * Doing both would be mean more complex code, and sane people don't
1243                  * use non-sparse filesystems anymore.  This is already checked above.
1244                  */
1245                 if (gdb_off) {
1246                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1247                                                      gdb_num);
1248                         BUFFER_TRACE(gdb_bh, "get_write_access");
1249                         err = ext4_journal_get_write_access(handle, gdb_bh);
1250
1251                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1252                                 err = reserve_backup_gdb(handle, resize_inode, group);
1253                 } else if (meta_bg != 0) {
1254                         err = add_new_gdb_meta_bg(sb, handle, group);
1255                 } else {
1256                         err = add_new_gdb(handle, resize_inode, group);
1257                 }
1258                 if (err)
1259                         break;
1260         }
1261         return err;
1262 }
1263
1264 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1265 {
1266         struct buffer_head *bh = sb_getblk(sb, block);
1267         if (unlikely(!bh))
1268                 return NULL;
1269         if (!bh_uptodate_or_lock(bh)) {
1270                 if (bh_submit_read(bh) < 0) {
1271                         brelse(bh);
1272                         return NULL;
1273                 }
1274         }
1275
1276         return bh;
1277 }
1278
1279 static int ext4_set_bitmap_checksums(struct super_block *sb,
1280                                      ext4_group_t group,
1281                                      struct ext4_group_desc *gdp,
1282                                      struct ext4_new_group_data *group_data)
1283 {
1284         struct buffer_head *bh;
1285
1286         if (!ext4_has_metadata_csum(sb))
1287                 return 0;
1288
1289         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1290         if (!bh)
1291                 return -EIO;
1292         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1293                                    EXT4_INODES_PER_GROUP(sb) / 8);
1294         brelse(bh);
1295
1296         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1297         if (!bh)
1298                 return -EIO;
1299         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1300         brelse(bh);
1301
1302         return 0;
1303 }
1304
1305 /*
1306  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1307  */
1308 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1309                                 struct ext4_new_flex_group_data *flex_gd)
1310 {
1311         struct ext4_new_group_data      *group_data = flex_gd->groups;
1312         struct ext4_group_desc          *gdp;
1313         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1314         struct buffer_head              *gdb_bh;
1315         ext4_group_t                    group;
1316         __u16                           *bg_flags = flex_gd->bg_flags;
1317         int                             i, gdb_off, gdb_num, err = 0;
1318
1319
1320         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1321                 group = group_data->group;
1322
1323                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1324                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1325
1326                 /*
1327                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1328                  */
1329                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1330                 /* Update group descriptor block for new group */
1331                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1332                                                  gdb_off * EXT4_DESC_SIZE(sb));
1333
1334                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1335                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1336                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1337                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1338                 if (err) {
1339                         ext4_std_error(sb, err);
1340                         break;
1341                 }
1342
1343                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1344                 ext4_free_group_clusters_set(sb, gdp,
1345                                              group_data->free_clusters_count);
1346                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1347                 if (ext4_has_group_desc_csum(sb))
1348                         ext4_itable_unused_set(sb, gdp,
1349                                                EXT4_INODES_PER_GROUP(sb));
1350                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1351                 ext4_group_desc_csum_set(sb, group, gdp);
1352
1353                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1354                 if (unlikely(err)) {
1355                         ext4_std_error(sb, err);
1356                         break;
1357                 }
1358
1359                 /*
1360                  * We can allocate memory for mb_alloc based on the new group
1361                  * descriptor
1362                  */
1363                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1364                 if (err)
1365                         break;
1366         }
1367         return err;
1368 }
1369
1370 /*
1371  * ext4_update_super() updates the super block so that the newly added
1372  * groups can be seen by the filesystem.
1373  *
1374  * @sb: super block
1375  * @flex_gd: new added groups
1376  */
1377 static void ext4_update_super(struct super_block *sb,
1378                              struct ext4_new_flex_group_data *flex_gd)
1379 {
1380         ext4_fsblk_t blocks_count = 0;
1381         ext4_fsblk_t free_blocks = 0;
1382         ext4_fsblk_t reserved_blocks = 0;
1383         struct ext4_new_group_data *group_data = flex_gd->groups;
1384         struct ext4_sb_info *sbi = EXT4_SB(sb);
1385         struct ext4_super_block *es = sbi->s_es;
1386         int i;
1387
1388         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1389         /*
1390          * Make the new blocks and inodes valid next.  We do this before
1391          * increasing the group count so that once the group is enabled,
1392          * all of its blocks and inodes are already valid.
1393          *
1394          * We always allocate group-by-group, then block-by-block or
1395          * inode-by-inode within a group, so enabling these
1396          * blocks/inodes before the group is live won't actually let us
1397          * allocate the new space yet.
1398          */
1399         for (i = 0; i < flex_gd->count; i++) {
1400                 blocks_count += group_data[i].blocks_count;
1401                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1402         }
1403
1404         reserved_blocks = ext4_r_blocks_count(es) * 100;
1405         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1406         reserved_blocks *= blocks_count;
1407         do_div(reserved_blocks, 100);
1408
1409         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1410         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1411         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1412                      flex_gd->count);
1413         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1414                      flex_gd->count);
1415
1416         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1417         /*
1418          * We need to protect s_groups_count against other CPUs seeing
1419          * inconsistent state in the superblock.
1420          *
1421          * The precise rules we use are:
1422          *
1423          * * Writers must perform a smp_wmb() after updating all
1424          *   dependent data and before modifying the groups count
1425          *
1426          * * Readers must perform an smp_rmb() after reading the groups
1427          *   count and before reading any dependent data.
1428          *
1429          * NB. These rules can be relaxed when checking the group count
1430          * while freeing data, as we can only allocate from a block
1431          * group after serialising against the group count, and we can
1432          * only then free after serialising in turn against that
1433          * allocation.
1434          */
1435         smp_wmb();
1436
1437         /* Update the global fs size fields */
1438         sbi->s_groups_count += flex_gd->count;
1439         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1440                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1441
1442         /* Update the reserved block counts only once the new group is
1443          * active. */
1444         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1445                                 reserved_blocks);
1446
1447         /* Update the free space counts */
1448         percpu_counter_add(&sbi->s_freeclusters_counter,
1449                            EXT4_NUM_B2C(sbi, free_blocks));
1450         percpu_counter_add(&sbi->s_freeinodes_counter,
1451                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1452
1453         ext4_debug("free blocks count %llu",
1454                    percpu_counter_read(&sbi->s_freeclusters_counter));
1455         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1456                 ext4_group_t flex_group;
1457                 struct flex_groups *fg;
1458
1459                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1460                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1461                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1462                              &fg->free_clusters);
1463                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1464                            &fg->free_inodes);
1465         }
1466
1467         /*
1468          * Update the fs overhead information
1469          */
1470         ext4_calculate_overhead(sb);
1471
1472         if (test_opt(sb, DEBUG))
1473                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1474                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1475                        blocks_count, free_blocks, reserved_blocks);
1476 }
1477
1478 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1479  * _before_ we start modifying the filesystem, because we cannot abort the
1480  * transaction and not have it write the data to disk.
1481  */
1482 static int ext4_flex_group_add(struct super_block *sb,
1483                                struct inode *resize_inode,
1484                                struct ext4_new_flex_group_data *flex_gd)
1485 {
1486         struct ext4_sb_info *sbi = EXT4_SB(sb);
1487         struct ext4_super_block *es = sbi->s_es;
1488         ext4_fsblk_t o_blocks_count;
1489         ext4_grpblk_t last;
1490         ext4_group_t group;
1491         handle_t *handle;
1492         unsigned reserved_gdb;
1493         int err = 0, err2 = 0, credit;
1494
1495         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1496
1497         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1498         o_blocks_count = ext4_blocks_count(es);
1499         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1500         BUG_ON(last);
1501
1502         err = setup_new_flex_group_blocks(sb, flex_gd);
1503         if (err)
1504                 goto exit;
1505         /*
1506          * We will always be modifying at least the superblock and  GDT
1507          * blocks.  If we are adding a group past the last current GDT block,
1508          * we will also modify the inode and the dindirect block.  If we
1509          * are adding a group with superblock/GDT backups  we will also
1510          * modify each of the reserved GDT dindirect blocks.
1511          */
1512         credit = 3;     /* sb, resize inode, resize inode dindirect */
1513         /* GDT blocks */
1514         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1515         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1516         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1517         if (IS_ERR(handle)) {
1518                 err = PTR_ERR(handle);
1519                 goto exit;
1520         }
1521
1522         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1523         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1524         if (err)
1525                 goto exit_journal;
1526
1527         group = flex_gd->groups[0].group;
1528         BUG_ON(group != sbi->s_groups_count);
1529         err = ext4_add_new_descs(handle, sb, group,
1530                                 resize_inode, flex_gd->count);
1531         if (err)
1532                 goto exit_journal;
1533
1534         err = ext4_setup_new_descs(handle, sb, flex_gd);
1535         if (err)
1536                 goto exit_journal;
1537
1538         ext4_update_super(sb, flex_gd);
1539
1540         err = ext4_handle_dirty_super(handle, sb);
1541
1542 exit_journal:
1543         err2 = ext4_journal_stop(handle);
1544         if (!err)
1545                 err = err2;
1546
1547         if (!err) {
1548                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1549                 int gdb_num_end = ((group + flex_gd->count - 1) /
1550                                    EXT4_DESC_PER_BLOCK(sb));
1551                 int meta_bg = ext4_has_feature_meta_bg(sb);
1552                 sector_t old_gdb = 0;
1553
1554                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1555                                sizeof(struct ext4_super_block), 0);
1556                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1557                         struct buffer_head *gdb_bh;
1558
1559                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1560                                                      gdb_num);
1561                         if (old_gdb == gdb_bh->b_blocknr)
1562                                 continue;
1563                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1564                                        gdb_bh->b_size, meta_bg);
1565                         old_gdb = gdb_bh->b_blocknr;
1566                 }
1567         }
1568 exit:
1569         return err;
1570 }
1571
1572 static int ext4_setup_next_flex_gd(struct super_block *sb,
1573                                     struct ext4_new_flex_group_data *flex_gd,
1574                                     ext4_fsblk_t n_blocks_count,
1575                                     unsigned long flexbg_size)
1576 {
1577         struct ext4_sb_info *sbi = EXT4_SB(sb);
1578         struct ext4_super_block *es = sbi->s_es;
1579         struct ext4_new_group_data *group_data = flex_gd->groups;
1580         ext4_fsblk_t o_blocks_count;
1581         ext4_group_t n_group;
1582         ext4_group_t group;
1583         ext4_group_t last_group;
1584         ext4_grpblk_t last;
1585         ext4_grpblk_t clusters_per_group;
1586         unsigned long i;
1587
1588         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1589
1590         o_blocks_count = ext4_blocks_count(es);
1591
1592         if (o_blocks_count == n_blocks_count)
1593                 return 0;
1594
1595         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1596         BUG_ON(last);
1597         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1598
1599         last_group = group | (flexbg_size - 1);
1600         if (last_group > n_group)
1601                 last_group = n_group;
1602
1603         flex_gd->count = last_group - group + 1;
1604
1605         for (i = 0; i < flex_gd->count; i++) {
1606                 int overhead;
1607
1608                 group_data[i].group = group + i;
1609                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1610                 overhead = ext4_group_overhead_blocks(sb, group + i);
1611                 group_data[i].mdata_blocks = overhead;
1612                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1613                 if (ext4_has_group_desc_csum(sb)) {
1614                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1615                                                EXT4_BG_INODE_UNINIT;
1616                         if (!test_opt(sb, INIT_INODE_TABLE))
1617                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1618                 } else
1619                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1620         }
1621
1622         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1623                 /* We need to initialize block bitmap of last group. */
1624                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1625
1626         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1627                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1628                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1629                                                        last - 1;
1630         }
1631
1632         return 1;
1633 }
1634
1635 /* Add group descriptor data to an existing or new group descriptor block.
1636  * Ensure we handle all possible error conditions _before_ we start modifying
1637  * the filesystem, because we cannot abort the transaction and not have it
1638  * write the data to disk.
1639  *
1640  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1641  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1642  *
1643  * We only need to hold the superblock lock while we are actually adding
1644  * in the new group's counts to the superblock.  Prior to that we have
1645  * not really "added" the group at all.  We re-check that we are still
1646  * adding in the last group in case things have changed since verifying.
1647  */
1648 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1649 {
1650         struct ext4_new_flex_group_data flex_gd;
1651         struct ext4_sb_info *sbi = EXT4_SB(sb);
1652         struct ext4_super_block *es = sbi->s_es;
1653         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1654                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1655         struct inode *inode = NULL;
1656         int gdb_off;
1657         int err;
1658         __u16 bg_flags = 0;
1659
1660         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1661
1662         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1663                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1664                 return -EPERM;
1665         }
1666
1667         if (ext4_blocks_count(es) + input->blocks_count <
1668             ext4_blocks_count(es)) {
1669                 ext4_warning(sb, "blocks_count overflow");
1670                 return -EINVAL;
1671         }
1672
1673         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1674             le32_to_cpu(es->s_inodes_count)) {
1675                 ext4_warning(sb, "inodes_count overflow");
1676                 return -EINVAL;
1677         }
1678
1679         if (reserved_gdb || gdb_off == 0) {
1680                 if (!ext4_has_feature_resize_inode(sb) ||
1681                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1682                         ext4_warning(sb,
1683                                      "No reserved GDT blocks, can't resize");
1684                         return -EPERM;
1685                 }
1686                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1687                 if (IS_ERR(inode)) {
1688                         ext4_warning(sb, "Error opening resize inode");
1689                         return PTR_ERR(inode);
1690                 }
1691         }
1692
1693
1694         err = verify_group_input(sb, input);
1695         if (err)
1696                 goto out;
1697
1698         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1699         if (err)
1700                 goto out;
1701
1702         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1703         if (err)
1704                 goto out;
1705
1706         flex_gd.count = 1;
1707         flex_gd.groups = input;
1708         flex_gd.bg_flags = &bg_flags;
1709         err = ext4_flex_group_add(sb, inode, &flex_gd);
1710 out:
1711         iput(inode);
1712         return err;
1713 } /* ext4_group_add */
1714
1715 /*
1716  * extend a group without checking assuming that checking has been done.
1717  */
1718 static int ext4_group_extend_no_check(struct super_block *sb,
1719                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1720 {
1721         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1722         handle_t *handle;
1723         int err = 0, err2;
1724
1725         /* We will update the superblock, one block bitmap, and
1726          * one group descriptor via ext4_group_add_blocks().
1727          */
1728         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1729         if (IS_ERR(handle)) {
1730                 err = PTR_ERR(handle);
1731                 ext4_warning(sb, "error %d on journal start", err);
1732                 return err;
1733         }
1734
1735         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1736         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1737         if (err) {
1738                 ext4_warning(sb, "error %d on journal write access", err);
1739                 goto errout;
1740         }
1741
1742         ext4_blocks_count_set(es, o_blocks_count + add);
1743         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1744         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1745                    o_blocks_count + add);
1746         /* We add the blocks to the bitmap and set the group need init bit */
1747         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1748         if (err)
1749                 goto errout;
1750         ext4_handle_dirty_super(handle, sb);
1751         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1752                    o_blocks_count + add);
1753 errout:
1754         err2 = ext4_journal_stop(handle);
1755         if (err2 && !err)
1756                 err = err2;
1757
1758         if (!err) {
1759                 if (test_opt(sb, DEBUG))
1760                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1761                                "blocks\n", ext4_blocks_count(es));
1762                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1763                                (char *)es, sizeof(struct ext4_super_block), 0);
1764         }
1765         return err;
1766 }
1767
1768 /*
1769  * Extend the filesystem to the new number of blocks specified.  This entry
1770  * point is only used to extend the current filesystem to the end of the last
1771  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1772  * for emergencies (because it has no dependencies on reserved blocks).
1773  *
1774  * If we _really_ wanted, we could use default values to call ext4_group_add()
1775  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1776  * GDT blocks are reserved to grow to the desired size.
1777  */
1778 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1779                       ext4_fsblk_t n_blocks_count)
1780 {
1781         ext4_fsblk_t o_blocks_count;
1782         ext4_grpblk_t last;
1783         ext4_grpblk_t add;
1784         struct buffer_head *bh;
1785         int err;
1786         ext4_group_t group;
1787
1788         o_blocks_count = ext4_blocks_count(es);
1789
1790         if (test_opt(sb, DEBUG))
1791                 ext4_msg(sb, KERN_DEBUG,
1792                          "extending last group from %llu to %llu blocks",
1793                          o_blocks_count, n_blocks_count);
1794
1795         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1796                 return 0;
1797
1798         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1799                 ext4_msg(sb, KERN_ERR,
1800                          "filesystem too large to resize to %llu blocks safely",
1801                          n_blocks_count);
1802                 if (sizeof(sector_t) < 8)
1803                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1804                 return -EINVAL;
1805         }
1806
1807         if (n_blocks_count < o_blocks_count) {
1808                 ext4_warning(sb, "can't shrink FS - resize aborted");
1809                 return -EINVAL;
1810         }
1811
1812         /* Handle the remaining blocks in the last group only. */
1813         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1814
1815         if (last == 0) {
1816                 ext4_warning(sb, "need to use ext2online to resize further");
1817                 return -EPERM;
1818         }
1819
1820         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1821
1822         if (o_blocks_count + add < o_blocks_count) {
1823                 ext4_warning(sb, "blocks_count overflow");
1824                 return -EINVAL;
1825         }
1826
1827         if (o_blocks_count + add > n_blocks_count)
1828                 add = n_blocks_count - o_blocks_count;
1829
1830         if (o_blocks_count + add < n_blocks_count)
1831                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1832                              o_blocks_count + add, add);
1833
1834         /* See if the device is actually as big as what was requested */
1835         bh = sb_bread(sb, o_blocks_count + add - 1);
1836         if (!bh) {
1837                 ext4_warning(sb, "can't read last block, resize aborted");
1838                 return -ENOSPC;
1839         }
1840         brelse(bh);
1841
1842         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1843         return err;
1844 } /* ext4_group_extend */
1845
1846
1847 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1848 {
1849         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1850 }
1851
1852 /*
1853  * Release the resize inode and drop the resize_inode feature if there
1854  * are no more reserved gdt blocks, and then convert the file system
1855  * to enable meta_bg
1856  */
1857 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1858 {
1859         handle_t *handle;
1860         struct ext4_sb_info *sbi = EXT4_SB(sb);
1861         struct ext4_super_block *es = sbi->s_es;
1862         struct ext4_inode_info *ei = EXT4_I(inode);
1863         ext4_fsblk_t nr;
1864         int i, ret, err = 0;
1865         int credits = 1;
1866
1867         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1868         if (inode) {
1869                 if (es->s_reserved_gdt_blocks) {
1870                         ext4_error(sb, "Unexpected non-zero "
1871                                    "s_reserved_gdt_blocks");
1872                         return -EPERM;
1873                 }
1874
1875                 /* Do a quick sanity check of the resize inode */
1876                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1877                                              (9 - sbi->s_cluster_bits)))
1878                         goto invalid_resize_inode;
1879                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1880                         if (i == EXT4_DIND_BLOCK) {
1881                                 if (ei->i_data[i])
1882                                         continue;
1883                                 else
1884                                         goto invalid_resize_inode;
1885                         }
1886                         if (ei->i_data[i])
1887                                 goto invalid_resize_inode;
1888                 }
1889                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1890         }
1891
1892         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1893         if (IS_ERR(handle))
1894                 return PTR_ERR(handle);
1895
1896         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1897         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1898         if (err)
1899                 goto errout;
1900
1901         ext4_clear_feature_resize_inode(sb);
1902         ext4_set_feature_meta_bg(sb);
1903         sbi->s_es->s_first_meta_bg =
1904                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1905
1906         err = ext4_handle_dirty_super(handle, sb);
1907         if (err) {
1908                 ext4_std_error(sb, err);
1909                 goto errout;
1910         }
1911
1912         if (inode) {
1913                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1914                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1915                                  EXT4_FREE_BLOCKS_METADATA |
1916                                  EXT4_FREE_BLOCKS_FORGET);
1917                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1918                 inode->i_blocks = 0;
1919
1920                 err = ext4_mark_inode_dirty(handle, inode);
1921                 if (err)
1922                         ext4_std_error(sb, err);
1923         }
1924
1925 errout:
1926         ret = ext4_journal_stop(handle);
1927         if (!err)
1928                 err = ret;
1929         return ret;
1930
1931 invalid_resize_inode:
1932         ext4_error(sb, "corrupted/inconsistent resize inode");
1933         return -EINVAL;
1934 }
1935
1936 /*
1937  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1938  *
1939  * @sb: super block of the fs to be resized
1940  * @n_blocks_count: the number of blocks resides in the resized fs
1941  */
1942 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1943 {
1944         struct ext4_new_flex_group_data *flex_gd = NULL;
1945         struct ext4_sb_info *sbi = EXT4_SB(sb);
1946         struct ext4_super_block *es = sbi->s_es;
1947         struct buffer_head *bh;
1948         struct inode *resize_inode = NULL;
1949         ext4_grpblk_t add, offset;
1950         unsigned long n_desc_blocks;
1951         unsigned long o_desc_blocks;
1952         ext4_group_t o_group;
1953         ext4_group_t n_group;
1954         ext4_fsblk_t o_blocks_count;
1955         ext4_fsblk_t n_blocks_count_retry = 0;
1956         unsigned long last_update_time = 0;
1957         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1958         int meta_bg;
1959
1960         /* See if the device is actually as big as what was requested */
1961         bh = sb_bread(sb, n_blocks_count - 1);
1962         if (!bh) {
1963                 ext4_warning(sb, "can't read last block, resize aborted");
1964                 return -ENOSPC;
1965         }
1966         brelse(bh);
1967
1968 retry:
1969         o_blocks_count = ext4_blocks_count(es);
1970
1971         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1972                  "to %llu blocks", o_blocks_count, n_blocks_count);
1973
1974         if (n_blocks_count < o_blocks_count) {
1975                 /* On-line shrinking not supported */
1976                 ext4_warning(sb, "can't shrink FS - resize aborted");
1977                 return -EINVAL;
1978         }
1979
1980         if (n_blocks_count == o_blocks_count)
1981                 /* Nothing need to do */
1982                 return 0;
1983
1984         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1985         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1986                 ext4_warning(sb, "resize would cause inodes_count overflow");
1987                 return -EINVAL;
1988         }
1989         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1990
1991         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1992         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1993
1994         meta_bg = ext4_has_feature_meta_bg(sb);
1995
1996         if (ext4_has_feature_resize_inode(sb)) {
1997                 if (meta_bg) {
1998                         ext4_error(sb, "resize_inode and meta_bg enabled "
1999                                    "simultaneously");
2000                         return -EINVAL;
2001                 }
2002                 if (n_desc_blocks > o_desc_blocks +
2003                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
2004                         n_blocks_count_retry = n_blocks_count;
2005                         n_desc_blocks = o_desc_blocks +
2006                                 le16_to_cpu(es->s_reserved_gdt_blocks);
2007                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2008                         n_blocks_count = (ext4_fsblk_t)n_group *
2009                                 EXT4_BLOCKS_PER_GROUP(sb) +
2010                                 le32_to_cpu(es->s_first_data_block);
2011                         n_group--; /* set to last group number */
2012                 }
2013
2014                 if (!resize_inode)
2015                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2016                                                  EXT4_IGET_SPECIAL);
2017                 if (IS_ERR(resize_inode)) {
2018                         ext4_warning(sb, "Error opening resize inode");
2019                         return PTR_ERR(resize_inode);
2020                 }
2021         }
2022
2023         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2024                 err = ext4_convert_meta_bg(sb, resize_inode);
2025                 if (err)
2026                         goto out;
2027                 if (resize_inode) {
2028                         iput(resize_inode);
2029                         resize_inode = NULL;
2030                 }
2031                 if (n_blocks_count_retry) {
2032                         n_blocks_count = n_blocks_count_retry;
2033                         n_blocks_count_retry = 0;
2034                         goto retry;
2035                 }
2036         }
2037
2038         /*
2039          * Make sure the last group has enough space so that it's
2040          * guaranteed to have enough space for all metadata blocks
2041          * that it might need to hold.  (We might not need to store
2042          * the inode table blocks in the last block group, but there
2043          * will be cases where this might be needed.)
2044          */
2045         if ((ext4_group_first_block_no(sb, n_group) +
2046              ext4_group_overhead_blocks(sb, n_group) + 2 +
2047              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2048                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2049                 n_group--;
2050                 n_blocks_count_retry = 0;
2051                 if (resize_inode) {
2052                         iput(resize_inode);
2053                         resize_inode = NULL;
2054                 }
2055                 goto retry;
2056         }
2057
2058         /* extend the last group */
2059         if (n_group == o_group)
2060                 add = n_blocks_count - o_blocks_count;
2061         else
2062                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2063         if (add > 0) {
2064                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2065                 if (err)
2066                         goto out;
2067         }
2068
2069         if (ext4_blocks_count(es) == n_blocks_count)
2070                 goto out;
2071
2072         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2073         if (err)
2074                 goto out;
2075
2076         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2077         if (err)
2078                 goto out;
2079
2080         flex_gd = alloc_flex_gd(flexbg_size);
2081         if (flex_gd == NULL) {
2082                 err = -ENOMEM;
2083                 goto out;
2084         }
2085
2086         /* Add flex groups. Note that a regular group is a
2087          * flex group with 1 group.
2088          */
2089         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2090                                               flexbg_size)) {
2091                 if (jiffies - last_update_time > HZ * 10) {
2092                         if (last_update_time)
2093                                 ext4_msg(sb, KERN_INFO,
2094                                          "resized to %llu blocks",
2095                                          ext4_blocks_count(es));
2096                         last_update_time = jiffies;
2097                 }
2098                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2099                         break;
2100                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2101                 if (unlikely(err))
2102                         break;
2103         }
2104
2105         if (!err && n_blocks_count_retry) {
2106                 n_blocks_count = n_blocks_count_retry;
2107                 n_blocks_count_retry = 0;
2108                 free_flex_gd(flex_gd);
2109                 flex_gd = NULL;
2110                 if (resize_inode) {
2111                         iput(resize_inode);
2112                         resize_inode = NULL;
2113                 }
2114                 goto retry;
2115         }
2116
2117 out:
2118         if (flex_gd)
2119                 free_flex_gd(flex_gd);
2120         if (resize_inode != NULL)
2121                 iput(resize_inode);
2122         if (err)
2123                 ext4_warning(sb, "error (%d) occurred during "
2124                              "file system resize", err);
2125         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2126                  ext4_blocks_count(es));
2127         return err;
2128 }