2 * linux/fs/ext4/block_validity.c
5 * Theodore Ts'o (tytso@mit.edu)
7 * Track which blocks in the filesystem are metadata blocks that
8 * should never be used as data blocks by files or directories.
11 #include <linux/time.h>
13 #include <linux/namei.h>
14 #include <linux/quotaops.h>
15 #include <linux/buffer_head.h>
16 #include <linux/swap.h>
17 #include <linux/pagemap.h>
18 #include <linux/blkdev.h>
19 #include <linux/mutex.h>
20 #include <linux/slab.h>
23 struct ext4_system_zone
{
25 ext4_fsblk_t start_blk
;
29 static struct kmem_cache
*ext4_system_zone_cachep
;
31 int __init
ext4_init_system_zone(void)
33 ext4_system_zone_cachep
= KMEM_CACHE(ext4_system_zone
, 0);
34 if (ext4_system_zone_cachep
== NULL
)
39 void ext4_exit_system_zone(void)
41 kmem_cache_destroy(ext4_system_zone_cachep
);
44 static inline int can_merge(struct ext4_system_zone
*entry1
,
45 struct ext4_system_zone
*entry2
)
47 if ((entry1
->start_blk
+ entry1
->count
) == entry2
->start_blk
)
53 * Mark a range of blocks as belonging to the "system zone" --- that
54 * is, filesystem metadata blocks which should never be used by
57 static int add_system_zone(struct ext4_sb_info
*sbi
,
58 ext4_fsblk_t start_blk
,
61 struct ext4_system_zone
*new_entry
= NULL
, *entry
;
62 struct rb_node
**n
= &sbi
->system_blks
.rb_node
, *node
;
63 struct rb_node
*parent
= NULL
, *new_node
= NULL
;
67 entry
= rb_entry(parent
, struct ext4_system_zone
, node
);
68 if (start_blk
< entry
->start_blk
)
70 else if (start_blk
>= (entry
->start_blk
+ entry
->count
))
73 if (start_blk
+ count
> (entry
->start_blk
+
75 entry
->count
= (start_blk
+ count
-
78 new_entry
= rb_entry(new_node
, struct ext4_system_zone
,
85 new_entry
= kmem_cache_alloc(ext4_system_zone_cachep
,
89 new_entry
->start_blk
= start_blk
;
90 new_entry
->count
= count
;
91 new_node
= &new_entry
->node
;
93 rb_link_node(new_node
, parent
, n
);
94 rb_insert_color(new_node
, &sbi
->system_blks
);
97 /* Can we merge to the left? */
98 node
= rb_prev(new_node
);
100 entry
= rb_entry(node
, struct ext4_system_zone
, node
);
101 if (can_merge(entry
, new_entry
)) {
102 new_entry
->start_blk
= entry
->start_blk
;
103 new_entry
->count
+= entry
->count
;
104 rb_erase(node
, &sbi
->system_blks
);
105 kmem_cache_free(ext4_system_zone_cachep
, entry
);
109 /* Can we merge to the right? */
110 node
= rb_next(new_node
);
112 entry
= rb_entry(node
, struct ext4_system_zone
, node
);
113 if (can_merge(new_entry
, entry
)) {
114 new_entry
->count
+= entry
->count
;
115 rb_erase(node
, &sbi
->system_blks
);
116 kmem_cache_free(ext4_system_zone_cachep
, entry
);
122 static void debug_print_tree(struct ext4_sb_info
*sbi
)
124 struct rb_node
*node
;
125 struct ext4_system_zone
*entry
;
128 printk(KERN_INFO
"System zones: ");
129 node
= rb_first(&sbi
->system_blks
);
131 entry
= rb_entry(node
, struct ext4_system_zone
, node
);
132 printk("%s%llu-%llu", first
? "" : ", ",
133 entry
->start_blk
, entry
->start_blk
+ entry
->count
- 1);
135 node
= rb_next(node
);
140 int ext4_setup_system_zone(struct super_block
*sb
)
142 ext4_group_t ngroups
= ext4_get_groups_count(sb
);
143 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
144 struct ext4_group_desc
*gdp
;
146 int flex_size
= ext4_flex_bg_size(sbi
);
149 if (!test_opt(sb
, BLOCK_VALIDITY
)) {
150 if (EXT4_SB(sb
)->system_blks
.rb_node
)
151 ext4_release_system_zone(sb
);
154 if (EXT4_SB(sb
)->system_blks
.rb_node
)
157 for (i
=0; i
< ngroups
; i
++) {
158 if (ext4_bg_has_super(sb
, i
) &&
159 ((i
< 5) || ((i
% flex_size
) == 0)))
160 add_system_zone(sbi
, ext4_group_first_block_no(sb
, i
),
161 ext4_bg_num_gdb(sb
, i
) + 1);
162 gdp
= ext4_get_group_desc(sb
, i
, NULL
);
163 ret
= add_system_zone(sbi
, ext4_block_bitmap(sb
, gdp
), 1);
166 ret
= add_system_zone(sbi
, ext4_inode_bitmap(sb
, gdp
), 1);
169 ret
= add_system_zone(sbi
, ext4_inode_table(sb
, gdp
),
170 sbi
->s_itb_per_group
);
175 if (test_opt(sb
, DEBUG
))
176 debug_print_tree(EXT4_SB(sb
));
180 /* Called when the filesystem is unmounted */
181 void ext4_release_system_zone(struct super_block
*sb
)
183 struct rb_node
*n
= EXT4_SB(sb
)->system_blks
.rb_node
;
184 struct rb_node
*parent
;
185 struct ext4_system_zone
*entry
;
188 /* Do the node's children first */
198 * The node has no children; free it, and then zero
199 * out parent's link to it. Finally go to the
200 * beginning of the loop and try to free the parent
203 parent
= rb_parent(n
);
204 entry
= rb_entry(n
, struct ext4_system_zone
, node
);
205 kmem_cache_free(ext4_system_zone_cachep
, entry
);
207 EXT4_SB(sb
)->system_blks
= RB_ROOT
;
208 else if (parent
->rb_left
== n
)
209 parent
->rb_left
= NULL
;
210 else if (parent
->rb_right
== n
)
211 parent
->rb_right
= NULL
;
214 EXT4_SB(sb
)->system_blks
= RB_ROOT
;
218 * Returns 1 if the passed-in block region (start_blk,
219 * start_blk+count) is valid; 0 if some part of the block region
220 * overlaps with filesystem metadata blocks.
222 int ext4_data_block_valid(struct ext4_sb_info
*sbi
, ext4_fsblk_t start_blk
,
225 struct ext4_system_zone
*entry
;
226 struct rb_node
*n
= sbi
->system_blks
.rb_node
;
228 if ((start_blk
<= le32_to_cpu(sbi
->s_es
->s_first_data_block
)) ||
229 (start_blk
+ count
< start_blk
) ||
230 (start_blk
+ count
> ext4_blocks_count(sbi
->s_es
))) {
231 sbi
->s_es
->s_last_error_block
= cpu_to_le64(start_blk
);
235 entry
= rb_entry(n
, struct ext4_system_zone
, node
);
236 if (start_blk
+ count
- 1 < entry
->start_blk
)
238 else if (start_blk
>= (entry
->start_blk
+ entry
->count
))
241 sbi
->s_es
->s_last_error_block
= cpu_to_le64(start_blk
);
248 int ext4_check_blockref(const char *function
, unsigned int line
,
249 struct inode
*inode
, __le32
*p
, unsigned int max
)
251 struct ext4_super_block
*es
= EXT4_SB(inode
->i_sb
)->s_es
;
255 while (bref
< p
+max
) {
256 blk
= le32_to_cpu(*bref
++);
258 unlikely(!ext4_data_block_valid(EXT4_SB(inode
->i_sb
),
260 es
->s_last_error_block
= cpu_to_le64(blk
);
261 ext4_error_inode(inode
, function
, line
, blk
,