GNU Linux-libre 4.9.326-gnu1
[releases.git] / fs / ext4 / resize.c
1 /*
2  *  linux/fs/ext4/resize.c
3  *
4  * Support for resizing an ext4 filesystem while it is mounted.
5  *
6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
7  *
8  * This could probably be made into a module, because it is not often in use.
9  */
10
11
12 #define EXT4FS_DEBUG
13
14 #include <linux/errno.h>
15 #include <linux/slab.h>
16
17 #include "ext4_jbd2.h"
18
19 struct ext4_rcu_ptr {
20         struct rcu_head rcu;
21         void *ptr;
22 };
23
24 static void ext4_rcu_ptr_callback(struct rcu_head *head)
25 {
26         struct ext4_rcu_ptr *ptr;
27
28         ptr = container_of(head, struct ext4_rcu_ptr, rcu);
29         kvfree(ptr->ptr);
30         kfree(ptr);
31 }
32
33 void ext4_kvfree_array_rcu(void *to_free)
34 {
35         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
36
37         if (ptr) {
38                 ptr->ptr = to_free;
39                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
40                 return;
41         }
42         synchronize_rcu();
43         kvfree(to_free);
44 }
45
46 int ext4_resize_begin(struct super_block *sb)
47 {
48         struct ext4_sb_info *sbi = EXT4_SB(sb);
49         int ret = 0;
50
51         if (!capable(CAP_SYS_RESOURCE))
52                 return -EPERM;
53
54         /*
55          * If the reserved GDT blocks is non-zero, the resize_inode feature
56          * should always be set.
57          */
58         if (EXT4_SB(sb)->s_es->s_reserved_gdt_blocks &&
59             !ext4_has_feature_resize_inode(sb)) {
60                 ext4_error(sb, "resize_inode disabled but reserved GDT blocks non-zero");
61                 return -EFSCORRUPTED;
62         }
63
64         /*
65          * If we are not using the primary superblock/GDT copy don't resize,
66          * because the user tools have no way of handling this.  Probably a
67          * bad time to do it anyways.
68          */
69         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
70             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
71                 ext4_warning(sb, "won't resize using backup superblock at %llu",
72                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
73                 return -EPERM;
74         }
75
76         /*
77          * We are not allowed to do online-resizing on a filesystem mounted
78          * with error, because it can destroy the filesystem easily.
79          */
80         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
81                 ext4_warning(sb, "There are errors in the filesystem, "
82                              "so online resizing is not allowed");
83                 return -EPERM;
84         }
85
86         if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
87                 ret = -EBUSY;
88
89         return ret;
90 }
91
92 void ext4_resize_end(struct super_block *sb)
93 {
94         clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
95         smp_mb__after_atomic();
96 }
97
98 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
99                                              ext4_group_t group) {
100         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
101                EXT4_DESC_PER_BLOCK_BITS(sb);
102 }
103
104 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
105                                              ext4_group_t group) {
106         group = ext4_meta_bg_first_group(sb, group);
107         return ext4_group_first_block_no(sb, group);
108 }
109
110 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
111                                                 ext4_group_t group) {
112         ext4_grpblk_t overhead;
113         overhead = ext4_bg_num_gdb(sb, group);
114         if (ext4_bg_has_super(sb, group))
115                 overhead += 1 +
116                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
117         return overhead;
118 }
119
120 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
121 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
122
123 static int verify_group_input(struct super_block *sb,
124                               struct ext4_new_group_data *input)
125 {
126         struct ext4_sb_info *sbi = EXT4_SB(sb);
127         struct ext4_super_block *es = sbi->s_es;
128         ext4_fsblk_t start = ext4_blocks_count(es);
129         ext4_fsblk_t end = start + input->blocks_count;
130         ext4_group_t group = input->group;
131         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
132         unsigned overhead;
133         ext4_fsblk_t metaend;
134         struct buffer_head *bh = NULL;
135         ext4_grpblk_t free_blocks_count, offset;
136         int err = -EINVAL;
137
138         if (group != sbi->s_groups_count) {
139                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
140                              input->group, sbi->s_groups_count);
141                 return -EINVAL;
142         }
143
144         overhead = ext4_group_overhead_blocks(sb, group);
145         metaend = start + overhead;
146         input->free_blocks_count = free_blocks_count =
147                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
148
149         if (test_opt(sb, DEBUG))
150                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
151                        "(%d free, %u reserved)\n",
152                        ext4_bg_has_super(sb, input->group) ? "normal" :
153                        "no-super", input->group, input->blocks_count,
154                        free_blocks_count, input->reserved_blocks);
155
156         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
157         if (offset != 0)
158                         ext4_warning(sb, "Last group not full");
159         else if (input->reserved_blocks > input->blocks_count / 5)
160                 ext4_warning(sb, "Reserved blocks too high (%u)",
161                              input->reserved_blocks);
162         else if (free_blocks_count < 0)
163                 ext4_warning(sb, "Bad blocks count %u",
164                              input->blocks_count);
165         else if (!(bh = sb_bread(sb, end - 1)))
166                 ext4_warning(sb, "Cannot read last block (%llu)",
167                              end - 1);
168         else if (outside(input->block_bitmap, start, end))
169                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
170                              (unsigned long long)input->block_bitmap);
171         else if (outside(input->inode_bitmap, start, end))
172                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
173                              (unsigned long long)input->inode_bitmap);
174         else if (outside(input->inode_table, start, end) ||
175                  outside(itend - 1, start, end))
176                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
177                              (unsigned long long)input->inode_table, itend - 1);
178         else if (input->inode_bitmap == input->block_bitmap)
179                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
180                              (unsigned long long)input->block_bitmap);
181         else if (inside(input->block_bitmap, input->inode_table, itend))
182                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
183                              "(%llu-%llu)",
184                              (unsigned long long)input->block_bitmap,
185                              (unsigned long long)input->inode_table, itend - 1);
186         else if (inside(input->inode_bitmap, input->inode_table, itend))
187                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
188                              "(%llu-%llu)",
189                              (unsigned long long)input->inode_bitmap,
190                              (unsigned long long)input->inode_table, itend - 1);
191         else if (inside(input->block_bitmap, start, metaend))
192                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
193                              (unsigned long long)input->block_bitmap,
194                              start, metaend - 1);
195         else if (inside(input->inode_bitmap, start, metaend))
196                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
197                              (unsigned long long)input->inode_bitmap,
198                              start, metaend - 1);
199         else if (inside(input->inode_table, start, metaend) ||
200                  inside(itend - 1, start, metaend))
201                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
202                              "(%llu-%llu)",
203                              (unsigned long long)input->inode_table,
204                              itend - 1, start, metaend - 1);
205         else
206                 err = 0;
207         brelse(bh);
208
209         return err;
210 }
211
212 /*
213  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
214  * group each time.
215  */
216 struct ext4_new_flex_group_data {
217         struct ext4_new_group_data *groups;     /* new_group_data for groups
218                                                    in the flex group */
219         __u16 *bg_flags;                        /* block group flags of groups
220                                                    in @groups */
221         ext4_group_t count;                     /* number of groups in @groups
222                                                  */
223 };
224
225 /*
226  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
227  * @flexbg_size.
228  *
229  * Returns NULL on failure otherwise address of the allocated structure.
230  */
231 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
232 {
233         struct ext4_new_flex_group_data *flex_gd;
234
235         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
236         if (flex_gd == NULL)
237                 goto out3;
238
239         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
240                 goto out2;
241         flex_gd->count = flexbg_size;
242
243         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
244                                   flexbg_size, GFP_NOFS);
245         if (flex_gd->groups == NULL)
246                 goto out2;
247
248         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
249         if (flex_gd->bg_flags == NULL)
250                 goto out1;
251
252         return flex_gd;
253
254 out1:
255         kfree(flex_gd->groups);
256 out2:
257         kfree(flex_gd);
258 out3:
259         return NULL;
260 }
261
262 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
263 {
264         kfree(flex_gd->bg_flags);
265         kfree(flex_gd->groups);
266         kfree(flex_gd);
267 }
268
269 /*
270  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
271  * and inode tables for a flex group.
272  *
273  * This function is used by 64bit-resize.  Note that this function allocates
274  * group tables from the 1st group of groups contained by @flexgd, which may
275  * be a partial of a flex group.
276  *
277  * @sb: super block of fs to which the groups belongs
278  *
279  * Returns 0 on a successful allocation of the metadata blocks in the
280  * block group.
281  */
282 static int ext4_alloc_group_tables(struct super_block *sb,
283                                 struct ext4_new_flex_group_data *flex_gd,
284                                 int flexbg_size)
285 {
286         struct ext4_new_group_data *group_data = flex_gd->groups;
287         ext4_fsblk_t start_blk;
288         ext4_fsblk_t last_blk;
289         ext4_group_t src_group;
290         ext4_group_t bb_index = 0;
291         ext4_group_t ib_index = 0;
292         ext4_group_t it_index = 0;
293         ext4_group_t group;
294         ext4_group_t last_group;
295         unsigned overhead;
296         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
297
298         BUG_ON(flex_gd->count == 0 || group_data == NULL);
299
300         src_group = group_data[0].group;
301         last_group  = src_group + flex_gd->count - 1;
302
303         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
304                (last_group & ~(flexbg_size - 1))));
305 next_group:
306         group = group_data[0].group;
307         if (src_group >= group_data[0].group + flex_gd->count)
308                 return -ENOSPC;
309         start_blk = ext4_group_first_block_no(sb, src_group);
310         last_blk = start_blk + group_data[src_group - group].blocks_count;
311
312         overhead = ext4_group_overhead_blocks(sb, src_group);
313
314         start_blk += overhead;
315
316         /* We collect contiguous blocks as much as possible. */
317         src_group++;
318         for (; src_group <= last_group; src_group++) {
319                 overhead = ext4_group_overhead_blocks(sb, src_group);
320                 if (overhead == 0)
321                         last_blk += group_data[src_group - group].blocks_count;
322                 else
323                         break;
324         }
325
326         /* Allocate block bitmaps */
327         for (; bb_index < flex_gd->count; bb_index++) {
328                 if (start_blk >= last_blk)
329                         goto next_group;
330                 group_data[bb_index].block_bitmap = start_blk++;
331                 group = ext4_get_group_number(sb, start_blk - 1);
332                 group -= group_data[0].group;
333                 group_data[group].free_blocks_count--;
334                 flex_gd->bg_flags[group] &= uninit_mask;
335         }
336
337         /* Allocate inode bitmaps */
338         for (; ib_index < flex_gd->count; ib_index++) {
339                 if (start_blk >= last_blk)
340                         goto next_group;
341                 group_data[ib_index].inode_bitmap = start_blk++;
342                 group = ext4_get_group_number(sb, start_blk - 1);
343                 group -= group_data[0].group;
344                 group_data[group].free_blocks_count--;
345                 flex_gd->bg_flags[group] &= uninit_mask;
346         }
347
348         /* Allocate inode tables */
349         for (; it_index < flex_gd->count; it_index++) {
350                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
351                 ext4_fsblk_t next_group_start;
352
353                 if (start_blk + itb > last_blk)
354                         goto next_group;
355                 group_data[it_index].inode_table = start_blk;
356                 group = ext4_get_group_number(sb, start_blk);
357                 next_group_start = ext4_group_first_block_no(sb, group + 1);
358                 group -= group_data[0].group;
359
360                 if (start_blk + itb > next_group_start) {
361                         flex_gd->bg_flags[group + 1] &= uninit_mask;
362                         overhead = start_blk + itb - next_group_start;
363                         group_data[group + 1].free_blocks_count -= overhead;
364                         itb -= overhead;
365                 }
366
367                 group_data[group].free_blocks_count -= itb;
368                 flex_gd->bg_flags[group] &= uninit_mask;
369                 start_blk += EXT4_SB(sb)->s_itb_per_group;
370         }
371
372         if (test_opt(sb, DEBUG)) {
373                 int i;
374                 group = group_data[0].group;
375
376                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
377                        "%d groups, flexbg size is %d:\n", flex_gd->count,
378                        flexbg_size);
379
380                 for (i = 0; i < flex_gd->count; i++) {
381                         printk(KERN_DEBUG "adding %s group %u: %u "
382                                "blocks (%d free)\n",
383                                ext4_bg_has_super(sb, group + i) ? "normal" :
384                                "no-super", group + i,
385                                group_data[i].blocks_count,
386                                group_data[i].free_blocks_count);
387                 }
388         }
389         return 0;
390 }
391
392 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
393                                   ext4_fsblk_t blk)
394 {
395         struct buffer_head *bh;
396         int err;
397
398         bh = sb_getblk(sb, blk);
399         if (unlikely(!bh))
400                 return ERR_PTR(-ENOMEM);
401         BUFFER_TRACE(bh, "get_write_access");
402         if ((err = ext4_journal_get_write_access(handle, bh))) {
403                 brelse(bh);
404                 bh = ERR_PTR(err);
405         } else {
406                 memset(bh->b_data, 0, sb->s_blocksize);
407                 set_buffer_uptodate(bh);
408         }
409
410         return bh;
411 }
412
413 /*
414  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
415  * If that fails, restart the transaction & regain write access for the
416  * buffer head which is used for block_bitmap modifications.
417  */
418 static int extend_or_restart_transaction(handle_t *handle, int thresh)
419 {
420         int err;
421
422         if (ext4_handle_has_enough_credits(handle, thresh))
423                 return 0;
424
425         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
426         if (err < 0)
427                 return err;
428         if (err) {
429                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
430                 if (err)
431                         return err;
432         }
433
434         return 0;
435 }
436
437 /*
438  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
439  *
440  * Helper function for ext4_setup_new_group_blocks() which set .
441  *
442  * @sb: super block
443  * @handle: journal handle
444  * @flex_gd: flex group data
445  */
446 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
447                         struct ext4_new_flex_group_data *flex_gd,
448                         ext4_fsblk_t block, ext4_group_t count)
449 {
450         ext4_group_t count2;
451
452         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
453         for (count2 = count; count > 0; count -= count2, block += count2) {
454                 ext4_fsblk_t start;
455                 struct buffer_head *bh;
456                 ext4_group_t group;
457                 int err;
458
459                 group = ext4_get_group_number(sb, block);
460                 start = ext4_group_first_block_no(sb, group);
461                 group -= flex_gd->groups[0].group;
462
463                 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
464                 if (count2 > count)
465                         count2 = count;
466
467                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
468                         BUG_ON(flex_gd->count > 1);
469                         continue;
470                 }
471
472                 err = extend_or_restart_transaction(handle, 1);
473                 if (err)
474                         return err;
475
476                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
477                 if (unlikely(!bh))
478                         return -ENOMEM;
479
480                 BUFFER_TRACE(bh, "get_write_access");
481                 err = ext4_journal_get_write_access(handle, bh);
482                 if (err) {
483                         brelse(bh);
484                         return err;
485                 }
486                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
487                            block - start, count2);
488                 ext4_set_bits(bh->b_data, block - start, count2);
489
490                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
491                 brelse(bh);
492                 if (unlikely(err))
493                         return err;
494         }
495
496         return 0;
497 }
498
499 /*
500  * Set up the block and inode bitmaps, and the inode table for the new groups.
501  * This doesn't need to be part of the main transaction, since we are only
502  * changing blocks outside the actual filesystem.  We still do journaling to
503  * ensure the recovery is correct in case of a failure just after resize.
504  * If any part of this fails, we simply abort the resize.
505  *
506  * setup_new_flex_group_blocks handles a flex group as follow:
507  *  1. copy super block and GDT, and initialize group tables if necessary.
508  *     In this step, we only set bits in blocks bitmaps for blocks taken by
509  *     super block and GDT.
510  *  2. allocate group tables in block bitmaps, that is, set bits in block
511  *     bitmap for blocks taken by group tables.
512  */
513 static int setup_new_flex_group_blocks(struct super_block *sb,
514                                 struct ext4_new_flex_group_data *flex_gd)
515 {
516         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
517         ext4_fsblk_t start;
518         ext4_fsblk_t block;
519         struct ext4_sb_info *sbi = EXT4_SB(sb);
520         struct ext4_super_block *es = sbi->s_es;
521         struct ext4_new_group_data *group_data = flex_gd->groups;
522         __u16 *bg_flags = flex_gd->bg_flags;
523         handle_t *handle;
524         ext4_group_t group, count;
525         struct buffer_head *bh = NULL;
526         int reserved_gdb, i, j, err = 0, err2;
527         int meta_bg;
528
529         BUG_ON(!flex_gd->count || !group_data ||
530                group_data[0].group != sbi->s_groups_count);
531
532         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
533         meta_bg = ext4_has_feature_meta_bg(sb);
534
535         /* This transaction may be extended/restarted along the way */
536         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
537         if (IS_ERR(handle))
538                 return PTR_ERR(handle);
539
540         group = group_data[0].group;
541         for (i = 0; i < flex_gd->count; i++, group++) {
542                 unsigned long gdblocks;
543                 ext4_grpblk_t overhead;
544
545                 gdblocks = ext4_bg_num_gdb(sb, group);
546                 start = ext4_group_first_block_no(sb, group);
547
548                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
549                         goto handle_itb;
550
551                 if (meta_bg == 1) {
552                         ext4_group_t first_group;
553                         first_group = ext4_meta_bg_first_group(sb, group);
554                         if (first_group != group + 1 &&
555                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
556                                 goto handle_itb;
557                 }
558
559                 block = start + ext4_bg_has_super(sb, group);
560                 /* Copy all of the GDT blocks into the backup in this group */
561                 for (j = 0; j < gdblocks; j++, block++) {
562                         struct buffer_head *gdb;
563
564                         ext4_debug("update backup group %#04llx\n", block);
565                         err = extend_or_restart_transaction(handle, 1);
566                         if (err)
567                                 goto out;
568
569                         gdb = sb_getblk(sb, block);
570                         if (unlikely(!gdb)) {
571                                 err = -ENOMEM;
572                                 goto out;
573                         }
574
575                         BUFFER_TRACE(gdb, "get_write_access");
576                         err = ext4_journal_get_write_access(handle, gdb);
577                         if (err) {
578                                 brelse(gdb);
579                                 goto out;
580                         }
581                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
582                                 s_group_desc, j)->b_data, gdb->b_size);
583                         set_buffer_uptodate(gdb);
584
585                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
586                         if (unlikely(err)) {
587                                 brelse(gdb);
588                                 goto out;
589                         }
590                         brelse(gdb);
591                 }
592
593                 /* Zero out all of the reserved backup group descriptor
594                  * table blocks
595                  */
596                 if (ext4_bg_has_super(sb, group)) {
597                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
598                                         reserved_gdb, GFP_NOFS);
599                         if (err)
600                                 goto out;
601                 }
602
603 handle_itb:
604                 /* Initialize group tables of the grop @group */
605                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
606                         goto handle_bb;
607
608                 /* Zero out all of the inode table blocks */
609                 block = group_data[i].inode_table;
610                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
611                            block, sbi->s_itb_per_group);
612                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
613                                        GFP_NOFS);
614                 if (err)
615                         goto out;
616
617 handle_bb:
618                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
619                         goto handle_ib;
620
621                 /* Initialize block bitmap of the @group */
622                 block = group_data[i].block_bitmap;
623                 err = extend_or_restart_transaction(handle, 1);
624                 if (err)
625                         goto out;
626
627                 bh = bclean(handle, sb, block);
628                 if (IS_ERR(bh)) {
629                         err = PTR_ERR(bh);
630                         goto out;
631                 }
632                 overhead = ext4_group_overhead_blocks(sb, group);
633                 if (overhead != 0) {
634                         ext4_debug("mark backup superblock %#04llx (+0)\n",
635                                    start);
636                         ext4_set_bits(bh->b_data, 0, overhead);
637                 }
638                 ext4_mark_bitmap_end(group_data[i].blocks_count,
639                                      sb->s_blocksize * 8, bh->b_data);
640                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
641                 brelse(bh);
642                 if (err)
643                         goto out;
644
645 handle_ib:
646                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
647                         continue;
648
649                 /* Initialize inode bitmap of the @group */
650                 block = group_data[i].inode_bitmap;
651                 err = extend_or_restart_transaction(handle, 1);
652                 if (err)
653                         goto out;
654                 /* Mark unused entries in inode bitmap used */
655                 bh = bclean(handle, sb, block);
656                 if (IS_ERR(bh)) {
657                         err = PTR_ERR(bh);
658                         goto out;
659                 }
660
661                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
662                                      sb->s_blocksize * 8, bh->b_data);
663                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
664                 brelse(bh);
665                 if (err)
666                         goto out;
667         }
668
669         /* Mark group tables in block bitmap */
670         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
671                 count = group_table_count[j];
672                 start = (&group_data[0].block_bitmap)[j];
673                 block = start;
674                 for (i = 1; i < flex_gd->count; i++) {
675                         block += group_table_count[j];
676                         if (block == (&group_data[i].block_bitmap)[j]) {
677                                 count += group_table_count[j];
678                                 continue;
679                         }
680                         err = set_flexbg_block_bitmap(sb, handle,
681                                                 flex_gd, start, count);
682                         if (err)
683                                 goto out;
684                         count = group_table_count[j];
685                         start = (&group_data[i].block_bitmap)[j];
686                         block = start;
687                 }
688
689                 if (count) {
690                         err = set_flexbg_block_bitmap(sb, handle,
691                                                 flex_gd, start, count);
692                         if (err)
693                                 goto out;
694                 }
695         }
696
697 out:
698         err2 = ext4_journal_stop(handle);
699         if (err2 && !err)
700                 err = err2;
701
702         return err;
703 }
704
705 /*
706  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
707  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
708  * calling this for the first time.  In a sparse filesystem it will be the
709  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
710  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
711  */
712 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
713                                   unsigned *five, unsigned *seven)
714 {
715         unsigned *min = three;
716         int mult = 3;
717         unsigned ret;
718
719         if (!ext4_has_feature_sparse_super(sb)) {
720                 ret = *min;
721                 *min += 1;
722                 return ret;
723         }
724
725         if (*five < *min) {
726                 min = five;
727                 mult = 5;
728         }
729         if (*seven < *min) {
730                 min = seven;
731                 mult = 7;
732         }
733
734         ret = *min;
735         *min *= mult;
736
737         return ret;
738 }
739
740 /*
741  * Check that all of the backup GDT blocks are held in the primary GDT block.
742  * It is assumed that they are stored in group order.  Returns the number of
743  * groups in current filesystem that have BACKUPS, or -ve error code.
744  */
745 static int verify_reserved_gdb(struct super_block *sb,
746                                ext4_group_t end,
747                                struct buffer_head *primary)
748 {
749         const ext4_fsblk_t blk = primary->b_blocknr;
750         unsigned three = 1;
751         unsigned five = 5;
752         unsigned seven = 7;
753         unsigned grp;
754         __le32 *p = (__le32 *)primary->b_data;
755         int gdbackups = 0;
756
757         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
758                 if (le32_to_cpu(*p++) !=
759                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
760                         ext4_warning(sb, "reserved GDT %llu"
761                                      " missing grp %d (%llu)",
762                                      blk, grp,
763                                      grp *
764                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
765                                      blk);
766                         return -EINVAL;
767                 }
768                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
769                         return -EFBIG;
770         }
771
772         return gdbackups;
773 }
774
775 /*
776  * Called when we need to bring a reserved group descriptor table block into
777  * use from the resize inode.  The primary copy of the new GDT block currently
778  * is an indirect block (under the double indirect block in the resize inode).
779  * The new backup GDT blocks will be stored as leaf blocks in this indirect
780  * block, in group order.  Even though we know all the block numbers we need,
781  * we check to ensure that the resize inode has actually reserved these blocks.
782  *
783  * Don't need to update the block bitmaps because the blocks are still in use.
784  *
785  * We get all of the error cases out of the way, so that we are sure to not
786  * fail once we start modifying the data on disk, because JBD has no rollback.
787  */
788 static int add_new_gdb(handle_t *handle, struct inode *inode,
789                        ext4_group_t group)
790 {
791         struct super_block *sb = inode->i_sb;
792         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
793         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
794         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
795         struct buffer_head **o_group_desc, **n_group_desc;
796         struct buffer_head *dind;
797         struct buffer_head *gdb_bh;
798         int gdbackups;
799         struct ext4_iloc iloc;
800         __le32 *data;
801         int err;
802
803         if (test_opt(sb, DEBUG))
804                 printk(KERN_DEBUG
805                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
806                        gdb_num);
807
808         gdb_bh = sb_bread(sb, gdblock);
809         if (!gdb_bh)
810                 return -EIO;
811
812         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
813         if (gdbackups < 0) {
814                 err = gdbackups;
815                 goto exit_bh;
816         }
817
818         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
819         dind = sb_bread(sb, le32_to_cpu(*data));
820         if (!dind) {
821                 err = -EIO;
822                 goto exit_bh;
823         }
824
825         data = (__le32 *)dind->b_data;
826         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
827                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
828                              group, gdblock);
829                 err = -EINVAL;
830                 goto exit_dind;
831         }
832
833         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
834         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
835         if (unlikely(err))
836                 goto exit_dind;
837
838         BUFFER_TRACE(gdb_bh, "get_write_access");
839         err = ext4_journal_get_write_access(handle, gdb_bh);
840         if (unlikely(err))
841                 goto exit_dind;
842
843         BUFFER_TRACE(dind, "get_write_access");
844         err = ext4_journal_get_write_access(handle, dind);
845         if (unlikely(err))
846                 ext4_std_error(sb, err);
847
848         /* ext4_reserve_inode_write() gets a reference on the iloc */
849         err = ext4_reserve_inode_write(handle, inode, &iloc);
850         if (unlikely(err))
851                 goto exit_dind;
852
853         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
854                                      sizeof(struct buffer_head *),
855                                      GFP_NOFS);
856         if (!n_group_desc) {
857                 err = -ENOMEM;
858                 ext4_warning(sb, "not enough memory for %lu groups",
859                              gdb_num + 1);
860                 goto exit_inode;
861         }
862
863         /*
864          * Finally, we have all of the possible failures behind us...
865          *
866          * Remove new GDT block from inode double-indirect block and clear out
867          * the new GDT block for use (which also "frees" the backup GDT blocks
868          * from the reserved inode).  We don't need to change the bitmaps for
869          * these blocks, because they are marked as in-use from being in the
870          * reserved inode, and will become GDT blocks (primary and backup).
871          */
872         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
873         err = ext4_handle_dirty_metadata(handle, NULL, dind);
874         if (unlikely(err)) {
875                 ext4_std_error(sb, err);
876                 goto exit_inode;
877         }
878         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
879         ext4_mark_iloc_dirty(handle, inode, &iloc);
880         memset(gdb_bh->b_data, 0, sb->s_blocksize);
881         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
882         if (unlikely(err)) {
883                 ext4_std_error(sb, err);
884                 iloc.bh = NULL;
885                 goto exit_inode;
886         }
887         brelse(dind);
888
889         rcu_read_lock();
890         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
891         memcpy(n_group_desc, o_group_desc,
892                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
893         rcu_read_unlock();
894         n_group_desc[gdb_num] = gdb_bh;
895         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
896         EXT4_SB(sb)->s_gdb_count++;
897         ext4_kvfree_array_rcu(o_group_desc);
898
899         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
900         err = ext4_handle_dirty_super(handle, sb);
901         if (err)
902                 ext4_std_error(sb, err);
903
904         return err;
905
906 exit_inode:
907         kvfree(n_group_desc);
908         brelse(iloc.bh);
909 exit_dind:
910         brelse(dind);
911 exit_bh:
912         brelse(gdb_bh);
913
914         ext4_debug("leaving with error %d\n", err);
915         return err;
916 }
917
918 /*
919  * add_new_gdb_meta_bg is the sister of add_new_gdb.
920  */
921 static int add_new_gdb_meta_bg(struct super_block *sb,
922                                handle_t *handle, ext4_group_t group) {
923         ext4_fsblk_t gdblock;
924         struct buffer_head *gdb_bh;
925         struct buffer_head **o_group_desc, **n_group_desc;
926         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
927         int err;
928
929         gdblock = ext4_meta_bg_first_block_no(sb, group) +
930                    ext4_bg_has_super(sb, group);
931         gdb_bh = sb_bread(sb, gdblock);
932         if (!gdb_bh)
933                 return -EIO;
934         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
935                                      sizeof(struct buffer_head *),
936                                      GFP_NOFS);
937         if (!n_group_desc) {
938                 brelse(gdb_bh);
939                 err = -ENOMEM;
940                 ext4_warning(sb, "not enough memory for %lu groups",
941                              gdb_num + 1);
942                 return err;
943         }
944
945         rcu_read_lock();
946         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
947         memcpy(n_group_desc, o_group_desc,
948                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
949         rcu_read_unlock();
950         n_group_desc[gdb_num] = gdb_bh;
951
952         BUFFER_TRACE(gdb_bh, "get_write_access");
953         err = ext4_journal_get_write_access(handle, gdb_bh);
954         if (err) {
955                 kvfree(n_group_desc);
956                 brelse(gdb_bh);
957                 return err;
958         }
959
960         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
961         EXT4_SB(sb)->s_gdb_count++;
962         ext4_kvfree_array_rcu(o_group_desc);
963         return err;
964 }
965
966 /*
967  * Called when we are adding a new group which has a backup copy of each of
968  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
969  * We need to add these reserved backup GDT blocks to the resize inode, so
970  * that they are kept for future resizing and not allocated to files.
971  *
972  * Each reserved backup GDT block will go into a different indirect block.
973  * The indirect blocks are actually the primary reserved GDT blocks,
974  * so we know in advance what their block numbers are.  We only get the
975  * double-indirect block to verify it is pointing to the primary reserved
976  * GDT blocks so we don't overwrite a data block by accident.  The reserved
977  * backup GDT blocks are stored in their reserved primary GDT block.
978  */
979 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
980                               ext4_group_t group)
981 {
982         struct super_block *sb = inode->i_sb;
983         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
984         struct buffer_head **primary;
985         struct buffer_head *dind;
986         struct ext4_iloc iloc;
987         ext4_fsblk_t blk;
988         __le32 *data, *end;
989         int gdbackups = 0;
990         int res, i;
991         int err;
992
993         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
994         if (!primary)
995                 return -ENOMEM;
996
997         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
998         dind = sb_bread(sb, le32_to_cpu(*data));
999         if (!dind) {
1000                 err = -EIO;
1001                 goto exit_free;
1002         }
1003
1004         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1005         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1006                                          EXT4_ADDR_PER_BLOCK(sb));
1007         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1008
1009         /* Get each reserved primary GDT block and verify it holds backups */
1010         for (res = 0; res < reserved_gdb; res++, blk++) {
1011                 if (le32_to_cpu(*data) != blk) {
1012                         ext4_warning(sb, "reserved block %llu"
1013                                      " not at offset %ld",
1014                                      blk,
1015                                      (long)(data - (__le32 *)dind->b_data));
1016                         err = -EINVAL;
1017                         goto exit_bh;
1018                 }
1019                 primary[res] = sb_bread(sb, blk);
1020                 if (!primary[res]) {
1021                         err = -EIO;
1022                         goto exit_bh;
1023                 }
1024                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1025                 if (gdbackups < 0) {
1026                         brelse(primary[res]);
1027                         err = gdbackups;
1028                         goto exit_bh;
1029                 }
1030                 if (++data >= end)
1031                         data = (__le32 *)dind->b_data;
1032         }
1033
1034         for (i = 0; i < reserved_gdb; i++) {
1035                 BUFFER_TRACE(primary[i], "get_write_access");
1036                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1037                         goto exit_bh;
1038         }
1039
1040         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1041                 goto exit_bh;
1042
1043         /*
1044          * Finally we can add each of the reserved backup GDT blocks from
1045          * the new group to its reserved primary GDT block.
1046          */
1047         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1048         for (i = 0; i < reserved_gdb; i++) {
1049                 int err2;
1050                 data = (__le32 *)primary[i]->b_data;
1051                 /* printk("reserving backup %lu[%u] = %lu\n",
1052                        primary[i]->b_blocknr, gdbackups,
1053                        blk + primary[i]->b_blocknr); */
1054                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1055                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1056                 if (!err)
1057                         err = err2;
1058         }
1059         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1060         ext4_mark_iloc_dirty(handle, inode, &iloc);
1061
1062 exit_bh:
1063         while (--res >= 0)
1064                 brelse(primary[res]);
1065         brelse(dind);
1066
1067 exit_free:
1068         kfree(primary);
1069
1070         return err;
1071 }
1072
1073 /*
1074  * Update the backup copies of the ext4 metadata.  These don't need to be part
1075  * of the main resize transaction, because e2fsck will re-write them if there
1076  * is a problem (basically only OOM will cause a problem).  However, we
1077  * _should_ update the backups if possible, in case the primary gets trashed
1078  * for some reason and we need to run e2fsck from a backup superblock.  The
1079  * important part is that the new block and inode counts are in the backup
1080  * superblocks, and the location of the new group metadata in the GDT backups.
1081  *
1082  * We do not need take the s_resize_lock for this, because these
1083  * blocks are not otherwise touched by the filesystem code when it is
1084  * mounted.  We don't need to worry about last changing from
1085  * sbi->s_groups_count, because the worst that can happen is that we
1086  * do not copy the full number of backups at this time.  The resize
1087  * which changed s_groups_count will backup again.
1088  */
1089 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1090                            int size, int meta_bg)
1091 {
1092         struct ext4_sb_info *sbi = EXT4_SB(sb);
1093         ext4_group_t last;
1094         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1095         unsigned three = 1;
1096         unsigned five = 5;
1097         unsigned seven = 7;
1098         ext4_group_t group = 0;
1099         int rest = sb->s_blocksize - size;
1100         handle_t *handle;
1101         int err = 0, err2;
1102
1103         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1104         if (IS_ERR(handle)) {
1105                 group = 1;
1106                 err = PTR_ERR(handle);
1107                 goto exit_err;
1108         }
1109
1110         if (meta_bg == 0) {
1111                 group = ext4_list_backups(sb, &three, &five, &seven);
1112                 last = sbi->s_groups_count;
1113         } else {
1114                 group = ext4_get_group_number(sb, blk_off) + 1;
1115                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1116         }
1117
1118         while (group < sbi->s_groups_count) {
1119                 struct buffer_head *bh;
1120                 ext4_fsblk_t backup_block;
1121
1122                 /* Out of journal space, and can't get more - abort - so sad */
1123                 if (ext4_handle_valid(handle) &&
1124                     handle->h_buffer_credits == 0 &&
1125                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1126                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1127                         break;
1128
1129                 if (meta_bg == 0)
1130                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1131                 else
1132                         backup_block = (ext4_group_first_block_no(sb, group) +
1133                                         ext4_bg_has_super(sb, group));
1134
1135                 bh = sb_getblk(sb, backup_block);
1136                 if (unlikely(!bh)) {
1137                         err = -ENOMEM;
1138                         break;
1139                 }
1140                 ext4_debug("update metadata backup %llu(+%llu)\n",
1141                            backup_block, backup_block -
1142                            ext4_group_first_block_no(sb, group));
1143                 BUFFER_TRACE(bh, "get_write_access");
1144                 if ((err = ext4_journal_get_write_access(handle, bh))) {
1145                         brelse(bh);
1146                         break;
1147                 }
1148                 lock_buffer(bh);
1149                 memcpy(bh->b_data, data, size);
1150                 if (rest)
1151                         memset(bh->b_data + size, 0, rest);
1152                 set_buffer_uptodate(bh);
1153                 unlock_buffer(bh);
1154                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1155                 if (unlikely(err))
1156                         ext4_std_error(sb, err);
1157                 brelse(bh);
1158
1159                 if (meta_bg == 0)
1160                         group = ext4_list_backups(sb, &three, &five, &seven);
1161                 else if (group == last)
1162                         break;
1163                 else
1164                         group = last;
1165         }
1166         if ((err2 = ext4_journal_stop(handle)) && !err)
1167                 err = err2;
1168
1169         /*
1170          * Ugh! Need to have e2fsck write the backup copies.  It is too
1171          * late to revert the resize, we shouldn't fail just because of
1172          * the backup copies (they are only needed in case of corruption).
1173          *
1174          * However, if we got here we have a journal problem too, so we
1175          * can't really start a transaction to mark the superblock.
1176          * Chicken out and just set the flag on the hope it will be written
1177          * to disk, and if not - we will simply wait until next fsck.
1178          */
1179 exit_err:
1180         if (err) {
1181                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1182                              "forcing fsck on next reboot", group, err);
1183                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1184                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1185                 mark_buffer_dirty(sbi->s_sbh);
1186         }
1187 }
1188
1189 /*
1190  * ext4_add_new_descs() adds @count group descriptor of groups
1191  * starting at @group
1192  *
1193  * @handle: journal handle
1194  * @sb: super block
1195  * @group: the group no. of the first group desc to be added
1196  * @resize_inode: the resize inode
1197  * @count: number of group descriptors to be added
1198  */
1199 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1200                               ext4_group_t group, struct inode *resize_inode,
1201                               ext4_group_t count)
1202 {
1203         struct ext4_sb_info *sbi = EXT4_SB(sb);
1204         struct ext4_super_block *es = sbi->s_es;
1205         struct buffer_head *gdb_bh;
1206         int i, gdb_off, gdb_num, err = 0;
1207         int meta_bg;
1208
1209         meta_bg = ext4_has_feature_meta_bg(sb);
1210         for (i = 0; i < count; i++, group++) {
1211                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1212                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1213
1214                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1215                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1216
1217                 /*
1218                  * We will only either add reserved group blocks to a backup group
1219                  * or remove reserved blocks for the first group in a new group block.
1220                  * Doing both would be mean more complex code, and sane people don't
1221                  * use non-sparse filesystems anymore.  This is already checked above.
1222                  */
1223                 if (gdb_off) {
1224                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1225                                                      gdb_num);
1226                         BUFFER_TRACE(gdb_bh, "get_write_access");
1227                         err = ext4_journal_get_write_access(handle, gdb_bh);
1228
1229                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1230                                 err = reserve_backup_gdb(handle, resize_inode, group);
1231                 } else if (meta_bg != 0) {
1232                         err = add_new_gdb_meta_bg(sb, handle, group);
1233                 } else {
1234                         err = add_new_gdb(handle, resize_inode, group);
1235                 }
1236                 if (err)
1237                         break;
1238         }
1239         return err;
1240 }
1241
1242 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1243 {
1244         struct buffer_head *bh = sb_getblk(sb, block);
1245         if (unlikely(!bh))
1246                 return NULL;
1247         if (!bh_uptodate_or_lock(bh)) {
1248                 if (bh_submit_read(bh) < 0) {
1249                         brelse(bh);
1250                         return NULL;
1251                 }
1252         }
1253
1254         return bh;
1255 }
1256
1257 static int ext4_set_bitmap_checksums(struct super_block *sb,
1258                                      ext4_group_t group,
1259                                      struct ext4_group_desc *gdp,
1260                                      struct ext4_new_group_data *group_data)
1261 {
1262         struct buffer_head *bh;
1263
1264         if (!ext4_has_metadata_csum(sb))
1265                 return 0;
1266
1267         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1268         if (!bh)
1269                 return -EIO;
1270         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1271                                    EXT4_INODES_PER_GROUP(sb) / 8);
1272         brelse(bh);
1273
1274         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1275         if (!bh)
1276                 return -EIO;
1277         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1278         brelse(bh);
1279
1280         return 0;
1281 }
1282
1283 /*
1284  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1285  */
1286 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1287                                 struct ext4_new_flex_group_data *flex_gd)
1288 {
1289         struct ext4_new_group_data      *group_data = flex_gd->groups;
1290         struct ext4_group_desc          *gdp;
1291         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1292         struct buffer_head              *gdb_bh;
1293         ext4_group_t                    group;
1294         __u16                           *bg_flags = flex_gd->bg_flags;
1295         int                             i, gdb_off, gdb_num, err = 0;
1296         
1297
1298         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1299                 group = group_data->group;
1300
1301                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1302                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1303
1304                 /*
1305                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1306                  */
1307                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1308                 /* Update group descriptor block for new group */
1309                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1310                                                  gdb_off * EXT4_DESC_SIZE(sb));
1311
1312                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1313                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1314                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1315                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1316                 if (err) {
1317                         ext4_std_error(sb, err);
1318                         break;
1319                 }
1320
1321                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1322                 ext4_free_group_clusters_set(sb, gdp,
1323                         EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1324                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1325                 if (ext4_has_group_desc_csum(sb))
1326                         ext4_itable_unused_set(sb, gdp,
1327                                                EXT4_INODES_PER_GROUP(sb));
1328                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1329                 ext4_group_desc_csum_set(sb, group, gdp);
1330
1331                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1332                 if (unlikely(err)) {
1333                         ext4_std_error(sb, err);
1334                         break;
1335                 }
1336
1337                 /*
1338                  * We can allocate memory for mb_alloc based on the new group
1339                  * descriptor
1340                  */
1341                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1342                 if (err)
1343                         break;
1344         }
1345         return err;
1346 }
1347
1348 /*
1349  * ext4_update_super() updates the super block so that the newly added
1350  * groups can be seen by the filesystem.
1351  *
1352  * @sb: super block
1353  * @flex_gd: new added groups
1354  */
1355 static void ext4_update_super(struct super_block *sb,
1356                              struct ext4_new_flex_group_data *flex_gd)
1357 {
1358         ext4_fsblk_t blocks_count = 0;
1359         ext4_fsblk_t free_blocks = 0;
1360         ext4_fsblk_t reserved_blocks = 0;
1361         struct ext4_new_group_data *group_data = flex_gd->groups;
1362         struct ext4_sb_info *sbi = EXT4_SB(sb);
1363         struct ext4_super_block *es = sbi->s_es;
1364         int i;
1365
1366         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1367         /*
1368          * Make the new blocks and inodes valid next.  We do this before
1369          * increasing the group count so that once the group is enabled,
1370          * all of its blocks and inodes are already valid.
1371          *
1372          * We always allocate group-by-group, then block-by-block or
1373          * inode-by-inode within a group, so enabling these
1374          * blocks/inodes before the group is live won't actually let us
1375          * allocate the new space yet.
1376          */
1377         for (i = 0; i < flex_gd->count; i++) {
1378                 blocks_count += group_data[i].blocks_count;
1379                 free_blocks += group_data[i].free_blocks_count;
1380         }
1381
1382         reserved_blocks = ext4_r_blocks_count(es) * 100;
1383         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1384         reserved_blocks *= blocks_count;
1385         do_div(reserved_blocks, 100);
1386
1387         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1388         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1389         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1390                      flex_gd->count);
1391         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1392                      flex_gd->count);
1393
1394         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1395         /*
1396          * We need to protect s_groups_count against other CPUs seeing
1397          * inconsistent state in the superblock.
1398          *
1399          * The precise rules we use are:
1400          *
1401          * * Writers must perform a smp_wmb() after updating all
1402          *   dependent data and before modifying the groups count
1403          *
1404          * * Readers must perform an smp_rmb() after reading the groups
1405          *   count and before reading any dependent data.
1406          *
1407          * NB. These rules can be relaxed when checking the group count
1408          * while freeing data, as we can only allocate from a block
1409          * group after serialising against the group count, and we can
1410          * only then free after serialising in turn against that
1411          * allocation.
1412          */
1413         smp_wmb();
1414
1415         /* Update the global fs size fields */
1416         sbi->s_groups_count += flex_gd->count;
1417         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1418                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1419
1420         /* Update the reserved block counts only once the new group is
1421          * active. */
1422         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1423                                 reserved_blocks);
1424
1425         /* Update the free space counts */
1426         percpu_counter_add(&sbi->s_freeclusters_counter,
1427                            EXT4_NUM_B2C(sbi, free_blocks));
1428         percpu_counter_add(&sbi->s_freeinodes_counter,
1429                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1430
1431         ext4_debug("free blocks count %llu",
1432                    percpu_counter_read(&sbi->s_freeclusters_counter));
1433         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1434                 ext4_group_t flex_group;
1435                 struct flex_groups *fg;
1436
1437                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1438                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1439                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1440                              &fg->free_clusters);
1441                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1442                            &fg->free_inodes);
1443         }
1444
1445         /*
1446          * Update the fs overhead information
1447          */
1448         ext4_calculate_overhead(sb);
1449         es->s_overhead_clusters = cpu_to_le32(sbi->s_overhead);
1450
1451         if (test_opt(sb, DEBUG))
1452                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1453                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1454                        blocks_count, free_blocks, reserved_blocks);
1455 }
1456
1457 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1458  * _before_ we start modifying the filesystem, because we cannot abort the
1459  * transaction and not have it write the data to disk.
1460  */
1461 static int ext4_flex_group_add(struct super_block *sb,
1462                                struct inode *resize_inode,
1463                                struct ext4_new_flex_group_data *flex_gd)
1464 {
1465         struct ext4_sb_info *sbi = EXT4_SB(sb);
1466         struct ext4_super_block *es = sbi->s_es;
1467         ext4_fsblk_t o_blocks_count;
1468         ext4_grpblk_t last;
1469         ext4_group_t group;
1470         handle_t *handle;
1471         unsigned reserved_gdb;
1472         int err = 0, err2 = 0, credit;
1473
1474         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1475
1476         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1477         o_blocks_count = ext4_blocks_count(es);
1478         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1479         BUG_ON(last);
1480
1481         err = setup_new_flex_group_blocks(sb, flex_gd);
1482         if (err)
1483                 goto exit;
1484         /*
1485          * We will always be modifying at least the superblock and  GDT
1486          * blocks.  If we are adding a group past the last current GDT block,
1487          * we will also modify the inode and the dindirect block.  If we
1488          * are adding a group with superblock/GDT backups  we will also
1489          * modify each of the reserved GDT dindirect blocks.
1490          */
1491         credit = 3;     /* sb, resize inode, resize inode dindirect */
1492         /* GDT blocks */
1493         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1494         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1495         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1496         if (IS_ERR(handle)) {
1497                 err = PTR_ERR(handle);
1498                 goto exit;
1499         }
1500
1501         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1502         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1503         if (err)
1504                 goto exit_journal;
1505
1506         group = flex_gd->groups[0].group;
1507         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1508         err = ext4_add_new_descs(handle, sb, group,
1509                                 resize_inode, flex_gd->count);
1510         if (err)
1511                 goto exit_journal;
1512
1513         err = ext4_setup_new_descs(handle, sb, flex_gd);
1514         if (err)
1515                 goto exit_journal;
1516
1517         ext4_update_super(sb, flex_gd);
1518
1519         err = ext4_handle_dirty_super(handle, sb);
1520
1521 exit_journal:
1522         err2 = ext4_journal_stop(handle);
1523         if (!err)
1524                 err = err2;
1525
1526         if (!err) {
1527                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1528                 int gdb_num_end = ((group + flex_gd->count - 1) /
1529                                    EXT4_DESC_PER_BLOCK(sb));
1530                 int meta_bg = ext4_has_feature_meta_bg(sb);
1531                 sector_t old_gdb = 0;
1532
1533                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1534                                sizeof(struct ext4_super_block), 0);
1535                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1536                         struct buffer_head *gdb_bh;
1537
1538                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1539                                                      gdb_num);
1540                         if (old_gdb == gdb_bh->b_blocknr)
1541                                 continue;
1542                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1543                                        gdb_bh->b_size, meta_bg);
1544                         old_gdb = gdb_bh->b_blocknr;
1545                 }
1546         }
1547 exit:
1548         return err;
1549 }
1550
1551 static int ext4_setup_next_flex_gd(struct super_block *sb,
1552                                     struct ext4_new_flex_group_data *flex_gd,
1553                                     ext4_fsblk_t n_blocks_count,
1554                                     unsigned long flexbg_size)
1555 {
1556         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1557         struct ext4_new_group_data *group_data = flex_gd->groups;
1558         ext4_fsblk_t o_blocks_count;
1559         ext4_group_t n_group;
1560         ext4_group_t group;
1561         ext4_group_t last_group;
1562         ext4_grpblk_t last;
1563         ext4_grpblk_t blocks_per_group;
1564         unsigned long i;
1565
1566         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1567
1568         o_blocks_count = ext4_blocks_count(es);
1569
1570         if (o_blocks_count == n_blocks_count)
1571                 return 0;
1572
1573         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1574         BUG_ON(last);
1575         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1576
1577         last_group = group | (flexbg_size - 1);
1578         if (last_group > n_group)
1579                 last_group = n_group;
1580
1581         flex_gd->count = last_group - group + 1;
1582
1583         for (i = 0; i < flex_gd->count; i++) {
1584                 int overhead;
1585
1586                 group_data[i].group = group + i;
1587                 group_data[i].blocks_count = blocks_per_group;
1588                 overhead = ext4_group_overhead_blocks(sb, group + i);
1589                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1590                 if (ext4_has_group_desc_csum(sb)) {
1591                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1592                                                EXT4_BG_INODE_UNINIT;
1593                         if (!test_opt(sb, INIT_INODE_TABLE))
1594                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1595                 } else
1596                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1597         }
1598
1599         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1600                 /* We need to initialize block bitmap of last group. */
1601                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1602
1603         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1604                 group_data[i - 1].blocks_count = last + 1;
1605                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1606                                         last - 1;
1607         }
1608
1609         return 1;
1610 }
1611
1612 /* Add group descriptor data to an existing or new group descriptor block.
1613  * Ensure we handle all possible error conditions _before_ we start modifying
1614  * the filesystem, because we cannot abort the transaction and not have it
1615  * write the data to disk.
1616  *
1617  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1618  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1619  *
1620  * We only need to hold the superblock lock while we are actually adding
1621  * in the new group's counts to the superblock.  Prior to that we have
1622  * not really "added" the group at all.  We re-check that we are still
1623  * adding in the last group in case things have changed since verifying.
1624  */
1625 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1626 {
1627         struct ext4_new_flex_group_data flex_gd;
1628         struct ext4_sb_info *sbi = EXT4_SB(sb);
1629         struct ext4_super_block *es = sbi->s_es;
1630         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1631                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1632         struct inode *inode = NULL;
1633         int gdb_off;
1634         int err;
1635         __u16 bg_flags = 0;
1636
1637         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1638
1639         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1640                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1641                 return -EPERM;
1642         }
1643
1644         if (ext4_blocks_count(es) + input->blocks_count <
1645             ext4_blocks_count(es)) {
1646                 ext4_warning(sb, "blocks_count overflow");
1647                 return -EINVAL;
1648         }
1649
1650         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1651             le32_to_cpu(es->s_inodes_count)) {
1652                 ext4_warning(sb, "inodes_count overflow");
1653                 return -EINVAL;
1654         }
1655
1656         if (reserved_gdb || gdb_off == 0) {
1657                 if (!ext4_has_feature_resize_inode(sb) ||
1658                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1659                         ext4_warning(sb,
1660                                      "No reserved GDT blocks, can't resize");
1661                         return -EPERM;
1662                 }
1663                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1664                 if (IS_ERR(inode)) {
1665                         ext4_warning(sb, "Error opening resize inode");
1666                         return PTR_ERR(inode);
1667                 }
1668         }
1669
1670
1671         err = verify_group_input(sb, input);
1672         if (err)
1673                 goto out;
1674
1675         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1676         if (err)
1677                 goto out;
1678
1679         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1680         if (err)
1681                 goto out;
1682
1683         flex_gd.count = 1;
1684         flex_gd.groups = input;
1685         flex_gd.bg_flags = &bg_flags;
1686         err = ext4_flex_group_add(sb, inode, &flex_gd);
1687 out:
1688         iput(inode);
1689         return err;
1690 } /* ext4_group_add */
1691
1692 /*
1693  * extend a group without checking assuming that checking has been done.
1694  */
1695 static int ext4_group_extend_no_check(struct super_block *sb,
1696                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1697 {
1698         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1699         handle_t *handle;
1700         int err = 0, err2;
1701
1702         /* We will update the superblock, one block bitmap, and
1703          * one group descriptor via ext4_group_add_blocks().
1704          */
1705         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1706         if (IS_ERR(handle)) {
1707                 err = PTR_ERR(handle);
1708                 ext4_warning(sb, "error %d on journal start", err);
1709                 return err;
1710         }
1711
1712         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1713         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1714         if (err) {
1715                 ext4_warning(sb, "error %d on journal write access", err);
1716                 goto errout;
1717         }
1718
1719         ext4_blocks_count_set(es, o_blocks_count + add);
1720         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1721         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1722                    o_blocks_count + add);
1723         /* We add the blocks to the bitmap and set the group need init bit */
1724         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1725         if (err)
1726                 goto errout;
1727         ext4_handle_dirty_super(handle, sb);
1728         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1729                    o_blocks_count + add);
1730 errout:
1731         err2 = ext4_journal_stop(handle);
1732         if (err2 && !err)
1733                 err = err2;
1734
1735         if (!err) {
1736                 if (test_opt(sb, DEBUG))
1737                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1738                                "blocks\n", ext4_blocks_count(es));
1739                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1740                                (char *)es, sizeof(struct ext4_super_block), 0);
1741         }
1742         return err;
1743 }
1744
1745 /*
1746  * Extend the filesystem to the new number of blocks specified.  This entry
1747  * point is only used to extend the current filesystem to the end of the last
1748  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1749  * for emergencies (because it has no dependencies on reserved blocks).
1750  *
1751  * If we _really_ wanted, we could use default values to call ext4_group_add()
1752  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1753  * GDT blocks are reserved to grow to the desired size.
1754  */
1755 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1756                       ext4_fsblk_t n_blocks_count)
1757 {
1758         ext4_fsblk_t o_blocks_count;
1759         ext4_grpblk_t last;
1760         ext4_grpblk_t add;
1761         struct buffer_head *bh;
1762         int err;
1763         ext4_group_t group;
1764
1765         o_blocks_count = ext4_blocks_count(es);
1766
1767         if (test_opt(sb, DEBUG))
1768                 ext4_msg(sb, KERN_DEBUG,
1769                          "extending last group from %llu to %llu blocks",
1770                          o_blocks_count, n_blocks_count);
1771
1772         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1773                 return 0;
1774
1775         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1776                 ext4_msg(sb, KERN_ERR,
1777                          "filesystem too large to resize to %llu blocks safely",
1778                          n_blocks_count);
1779                 if (sizeof(sector_t) < 8)
1780                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1781                 return -EINVAL;
1782         }
1783
1784         if (n_blocks_count < o_blocks_count) {
1785                 ext4_warning(sb, "can't shrink FS - resize aborted");
1786                 return -EINVAL;
1787         }
1788
1789         /* Handle the remaining blocks in the last group only. */
1790         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1791
1792         if (last == 0) {
1793                 ext4_warning(sb, "need to use ext2online to resize further");
1794                 return -EPERM;
1795         }
1796
1797         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1798
1799         if (o_blocks_count + add < o_blocks_count) {
1800                 ext4_warning(sb, "blocks_count overflow");
1801                 return -EINVAL;
1802         }
1803
1804         if (o_blocks_count + add > n_blocks_count)
1805                 add = n_blocks_count - o_blocks_count;
1806
1807         if (o_blocks_count + add < n_blocks_count)
1808                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1809                              o_blocks_count + add, add);
1810
1811         /* See if the device is actually as big as what was requested */
1812         bh = sb_bread(sb, o_blocks_count + add - 1);
1813         if (!bh) {
1814                 ext4_warning(sb, "can't read last block, resize aborted");
1815                 return -ENOSPC;
1816         }
1817         brelse(bh);
1818
1819         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1820         return err;
1821 } /* ext4_group_extend */
1822
1823
1824 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1825 {
1826         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1827 }
1828
1829 /*
1830  * Release the resize inode and drop the resize_inode feature if there
1831  * are no more reserved gdt blocks, and then convert the file system
1832  * to enable meta_bg
1833  */
1834 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1835 {
1836         handle_t *handle;
1837         struct ext4_sb_info *sbi = EXT4_SB(sb);
1838         struct ext4_super_block *es = sbi->s_es;
1839         struct ext4_inode_info *ei = EXT4_I(inode);
1840         ext4_fsblk_t nr;
1841         int i, ret, err = 0;
1842         int credits = 1;
1843
1844         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1845         if (inode) {
1846                 if (es->s_reserved_gdt_blocks) {
1847                         ext4_error(sb, "Unexpected non-zero "
1848                                    "s_reserved_gdt_blocks");
1849                         return -EPERM;
1850                 }
1851
1852                 /* Do a quick sanity check of the resize inode */
1853                 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1854                         goto invalid_resize_inode;
1855                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1856                         if (i == EXT4_DIND_BLOCK) {
1857                                 if (ei->i_data[i])
1858                                         continue;
1859                                 else
1860                                         goto invalid_resize_inode;
1861                         }
1862                         if (ei->i_data[i])
1863                                 goto invalid_resize_inode;
1864                 }
1865                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1866         }
1867
1868         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1869         if (IS_ERR(handle))
1870                 return PTR_ERR(handle);
1871
1872         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1873         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1874         if (err)
1875                 goto errout;
1876
1877         ext4_clear_feature_resize_inode(sb);
1878         ext4_set_feature_meta_bg(sb);
1879         sbi->s_es->s_first_meta_bg =
1880                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1881
1882         err = ext4_handle_dirty_super(handle, sb);
1883         if (err) {
1884                 ext4_std_error(sb, err);
1885                 goto errout;
1886         }
1887
1888         if (inode) {
1889                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1890                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1891                                  EXT4_FREE_BLOCKS_METADATA |
1892                                  EXT4_FREE_BLOCKS_FORGET);
1893                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1894                 inode->i_blocks = 0;
1895
1896                 err = ext4_mark_inode_dirty(handle, inode);
1897                 if (err)
1898                         ext4_std_error(sb, err);
1899         }
1900
1901 errout:
1902         ret = ext4_journal_stop(handle);
1903         if (!err)
1904                 err = ret;
1905         return ret;
1906
1907 invalid_resize_inode:
1908         ext4_error(sb, "corrupted/inconsistent resize inode");
1909         return -EINVAL;
1910 }
1911
1912 /*
1913  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1914  *
1915  * @sb: super block of the fs to be resized
1916  * @n_blocks_count: the number of blocks resides in the resized fs
1917  */
1918 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1919 {
1920         struct ext4_new_flex_group_data *flex_gd = NULL;
1921         struct ext4_sb_info *sbi = EXT4_SB(sb);
1922         struct ext4_super_block *es = sbi->s_es;
1923         struct buffer_head *bh;
1924         struct inode *resize_inode = NULL;
1925         ext4_grpblk_t add, offset;
1926         unsigned long n_desc_blocks;
1927         unsigned long o_desc_blocks;
1928         ext4_group_t o_group;
1929         ext4_group_t n_group;
1930         ext4_fsblk_t o_blocks_count;
1931         ext4_fsblk_t n_blocks_count_retry = 0;
1932         unsigned long last_update_time = 0;
1933         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1934         int meta_bg;
1935
1936         /* See if the device is actually as big as what was requested */
1937         bh = sb_bread(sb, n_blocks_count - 1);
1938         if (!bh) {
1939                 ext4_warning(sb, "can't read last block, resize aborted");
1940                 return -ENOSPC;
1941         }
1942         brelse(bh);
1943
1944         /*
1945          * For bigalloc, trim the requested size to the nearest cluster
1946          * boundary to avoid creating an unusable filesystem. We do this
1947          * silently, instead of returning an error, to avoid breaking
1948          * callers that blindly resize the filesystem to the full size of
1949          * the underlying block device.
1950          */
1951         if (ext4_has_feature_bigalloc(sb))
1952                 n_blocks_count &= ~((1 << EXT4_CLUSTER_BITS(sb)) - 1);
1953
1954 retry:
1955         o_blocks_count = ext4_blocks_count(es);
1956
1957         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1958                  "to %llu blocks", o_blocks_count, n_blocks_count);
1959
1960         if (n_blocks_count < o_blocks_count) {
1961                 /* On-line shrinking not supported */
1962                 ext4_warning(sb, "can't shrink FS - resize aborted");
1963                 return -EINVAL;
1964         }
1965
1966         if (n_blocks_count == o_blocks_count)
1967                 /* Nothing need to do */
1968                 return 0;
1969
1970         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1971         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1972                 ext4_warning(sb, "resize would cause inodes_count overflow");
1973                 return -EINVAL;
1974         }
1975         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1976
1977         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1978         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1979
1980         meta_bg = ext4_has_feature_meta_bg(sb);
1981
1982         if (ext4_has_feature_resize_inode(sb)) {
1983                 if (meta_bg) {
1984                         ext4_error(sb, "resize_inode and meta_bg enabled "
1985                                    "simultaneously");
1986                         return -EINVAL;
1987                 }
1988                 if (n_desc_blocks > o_desc_blocks +
1989                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1990                         n_blocks_count_retry = n_blocks_count;
1991                         n_desc_blocks = o_desc_blocks +
1992                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1993                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1994                         n_blocks_count = (ext4_fsblk_t)n_group *
1995                                 EXT4_BLOCKS_PER_GROUP(sb) +
1996                                 le32_to_cpu(es->s_first_data_block);
1997                         n_group--; /* set to last group number */
1998                 }
1999
2000                 if (!resize_inode)
2001                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2002                                                  EXT4_IGET_SPECIAL);
2003                 if (IS_ERR(resize_inode)) {
2004                         ext4_warning(sb, "Error opening resize inode");
2005                         return PTR_ERR(resize_inode);
2006                 }
2007         }
2008
2009         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2010                 err = ext4_convert_meta_bg(sb, resize_inode);
2011                 if (err)
2012                         goto out;
2013                 if (resize_inode) {
2014                         iput(resize_inode);
2015                         resize_inode = NULL;
2016                 }
2017                 if (n_blocks_count_retry) {
2018                         n_blocks_count = n_blocks_count_retry;
2019                         n_blocks_count_retry = 0;
2020                         goto retry;
2021                 }
2022         }
2023
2024         /*
2025          * Make sure the last group has enough space so that it's
2026          * guaranteed to have enough space for all metadata blocks
2027          * that it might need to hold.  (We might not need to store
2028          * the inode table blocks in the last block group, but there
2029          * will be cases where this might be needed.)
2030          */
2031         if ((ext4_group_first_block_no(sb, n_group) +
2032              ext4_group_overhead_blocks(sb, n_group) + 2 +
2033              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2034                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2035                 n_group--;
2036                 n_blocks_count_retry = 0;
2037                 if (resize_inode) {
2038                         iput(resize_inode);
2039                         resize_inode = NULL;
2040                 }
2041                 goto retry;
2042         }
2043
2044         /* extend the last group */
2045         if (n_group == o_group)
2046                 add = n_blocks_count - o_blocks_count;
2047         else
2048                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
2049         if (add > 0) {
2050                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2051                 if (err)
2052                         goto out;
2053         }
2054
2055         if (ext4_blocks_count(es) == n_blocks_count)
2056                 goto out;
2057
2058         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2059         if (err)
2060                 goto out;
2061
2062         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2063         if (err)
2064                 goto out;
2065
2066         flex_gd = alloc_flex_gd(flexbg_size);
2067         if (flex_gd == NULL) {
2068                 err = -ENOMEM;
2069                 goto out;
2070         }
2071
2072         /* Add flex groups. Note that a regular group is a
2073          * flex group with 1 group.
2074          */
2075         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2076                                               flexbg_size)) {
2077                 if (jiffies - last_update_time > HZ * 10) {
2078                         if (last_update_time)
2079                                 ext4_msg(sb, KERN_INFO,
2080                                          "resized to %llu blocks",
2081                                          ext4_blocks_count(es));
2082                         last_update_time = jiffies;
2083                 }
2084                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2085                         break;
2086                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2087                 if (unlikely(err))
2088                         break;
2089         }
2090
2091         if (!err && n_blocks_count_retry) {
2092                 n_blocks_count = n_blocks_count_retry;
2093                 n_blocks_count_retry = 0;
2094                 free_flex_gd(flex_gd);
2095                 flex_gd = NULL;
2096                 if (resize_inode) {
2097                         iput(resize_inode);
2098                         resize_inode = NULL;
2099                 }
2100                 goto retry;
2101         }
2102
2103 out:
2104         if (flex_gd)
2105                 free_flex_gd(flex_gd);
2106         if (resize_inode != NULL)
2107                 iput(resize_inode);
2108         if (err)
2109                 ext4_warning(sb, "error (%d) occurred during "
2110                              "file system resize", err);
2111         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2112                  ext4_blocks_count(es));
2113         return err;
2114 }