ext4: protect journal inode's blocks using block_validity
[sfrench/cifs-2.6.git] / fs / ext4 / block_validity.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/block_validity.c
4  *
5  * Copyright (C) 2009
6  * Theodore Ts'o (tytso@mit.edu)
7  *
8  * Track which blocks in the filesystem are metadata blocks that
9  * should never be used as data blocks by files or directories.
10  */
11
12 #include <linux/time.h>
13 #include <linux/fs.h>
14 #include <linux/namei.h>
15 #include <linux/quotaops.h>
16 #include <linux/buffer_head.h>
17 #include <linux/swap.h>
18 #include <linux/pagemap.h>
19 #include <linux/blkdev.h>
20 #include <linux/slab.h>
21 #include "ext4.h"
22
23 struct ext4_system_zone {
24         struct rb_node  node;
25         ext4_fsblk_t    start_blk;
26         unsigned int    count;
27 };
28
29 static struct kmem_cache *ext4_system_zone_cachep;
30
31 int __init ext4_init_system_zone(void)
32 {
33         ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
34         if (ext4_system_zone_cachep == NULL)
35                 return -ENOMEM;
36         return 0;
37 }
38
39 void ext4_exit_system_zone(void)
40 {
41         kmem_cache_destroy(ext4_system_zone_cachep);
42 }
43
44 static inline int can_merge(struct ext4_system_zone *entry1,
45                      struct ext4_system_zone *entry2)
46 {
47         if ((entry1->start_blk + entry1->count) == entry2->start_blk)
48                 return 1;
49         return 0;
50 }
51
52 /*
53  * Mark a range of blocks as belonging to the "system zone" --- that
54  * is, filesystem metadata blocks which should never be used by
55  * inodes.
56  */
57 static int add_system_zone(struct ext4_sb_info *sbi,
58                            ext4_fsblk_t start_blk,
59                            unsigned int count)
60 {
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;
64
65         while (*n) {
66                 parent = *n;
67                 entry = rb_entry(parent, struct ext4_system_zone, node);
68                 if (start_blk < entry->start_blk)
69                         n = &(*n)->rb_left;
70                 else if (start_blk >= (entry->start_blk + entry->count))
71                         n = &(*n)->rb_right;
72                 else {
73                         if (start_blk + count > (entry->start_blk +
74                                                  entry->count))
75                                 entry->count = (start_blk + count -
76                                                 entry->start_blk);
77                         new_node = *n;
78                         new_entry = rb_entry(new_node, struct ext4_system_zone,
79                                              node);
80                         break;
81                 }
82         }
83
84         if (!new_entry) {
85                 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
86                                              GFP_KERNEL);
87                 if (!new_entry)
88                         return -ENOMEM;
89                 new_entry->start_blk = start_blk;
90                 new_entry->count = count;
91                 new_node = &new_entry->node;
92
93                 rb_link_node(new_node, parent, n);
94                 rb_insert_color(new_node, &sbi->system_blks);
95         }
96
97         /* Can we merge to the left? */
98         node = rb_prev(new_node);
99         if (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);
106                 }
107         }
108
109         /* Can we merge to the right? */
110         node = rb_next(new_node);
111         if (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);
117                 }
118         }
119         return 0;
120 }
121
122 static void debug_print_tree(struct ext4_sb_info *sbi)
123 {
124         struct rb_node *node;
125         struct ext4_system_zone *entry;
126         int first = 1;
127
128         printk(KERN_INFO "System zones: ");
129         node = rb_first(&sbi->system_blks);
130         while (node) {
131                 entry = rb_entry(node, struct ext4_system_zone, node);
132                 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
133                        entry->start_blk, entry->start_blk + entry->count - 1);
134                 first = 0;
135                 node = rb_next(node);
136         }
137         printk(KERN_CONT "\n");
138 }
139
140 static int ext4_protect_reserved_inode(struct super_block *sb, u32 ino)
141 {
142         struct inode *inode;
143         struct ext4_sb_info *sbi = EXT4_SB(sb);
144         struct ext4_map_blocks map;
145         u32 i = 0, err = 0, num, n;
146
147         if ((ino < EXT4_ROOT_INO) ||
148             (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
149                 return -EINVAL;
150         inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
151         if (IS_ERR(inode))
152                 return PTR_ERR(inode);
153         num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
154         while (i < num) {
155                 map.m_lblk = i;
156                 map.m_len = num - i;
157                 n = ext4_map_blocks(NULL, inode, &map, 0);
158                 if (n < 0) {
159                         err = n;
160                         break;
161                 }
162                 if (n == 0) {
163                         i++;
164                 } else {
165                         if (!ext4_data_block_valid(sbi, map.m_pblk, n)) {
166                                 ext4_error(sb, "blocks %llu-%llu from inode %u "
167                                            "overlap system zone", map.m_pblk,
168                                            map.m_pblk + map.m_len - 1, ino);
169                                 err = -EFSCORRUPTED;
170                                 break;
171                         }
172                         err = add_system_zone(sbi, map.m_pblk, n);
173                         if (err < 0)
174                                 break;
175                         i += n;
176                 }
177         }
178         iput(inode);
179         return err;
180 }
181
182 int ext4_setup_system_zone(struct super_block *sb)
183 {
184         ext4_group_t ngroups = ext4_get_groups_count(sb);
185         struct ext4_sb_info *sbi = EXT4_SB(sb);
186         struct ext4_group_desc *gdp;
187         ext4_group_t i;
188         int flex_size = ext4_flex_bg_size(sbi);
189         int ret;
190
191         if (!test_opt(sb, BLOCK_VALIDITY)) {
192                 if (sbi->system_blks.rb_node)
193                         ext4_release_system_zone(sb);
194                 return 0;
195         }
196         if (sbi->system_blks.rb_node)
197                 return 0;
198
199         for (i=0; i < ngroups; i++) {
200                 if (ext4_bg_has_super(sb, i) &&
201                     ((i < 5) || ((i % flex_size) == 0)))
202                         add_system_zone(sbi, ext4_group_first_block_no(sb, i),
203                                         ext4_bg_num_gdb(sb, i) + 1);
204                 gdp = ext4_get_group_desc(sb, i, NULL);
205                 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1);
206                 if (ret)
207                         return ret;
208                 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1);
209                 if (ret)
210                         return ret;
211                 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
212                                 sbi->s_itb_per_group);
213                 if (ret)
214                         return ret;
215         }
216         if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
217                 ret = ext4_protect_reserved_inode(sb,
218                                 le32_to_cpu(sbi->s_es->s_journal_inum));
219                 if (ret)
220                         return ret;
221         }
222
223         if (test_opt(sb, DEBUG))
224                 debug_print_tree(sbi);
225         return 0;
226 }
227
228 /* Called when the filesystem is unmounted */
229 void ext4_release_system_zone(struct super_block *sb)
230 {
231         struct ext4_system_zone *entry, *n;
232
233         rbtree_postorder_for_each_entry_safe(entry, n,
234                         &EXT4_SB(sb)->system_blks, node)
235                 kmem_cache_free(ext4_system_zone_cachep, entry);
236
237         EXT4_SB(sb)->system_blks = RB_ROOT;
238 }
239
240 /*
241  * Returns 1 if the passed-in block region (start_blk,
242  * start_blk+count) is valid; 0 if some part of the block region
243  * overlaps with filesystem metadata blocks.
244  */
245 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
246                           unsigned int count)
247 {
248         struct ext4_system_zone *entry;
249         struct rb_node *n = sbi->system_blks.rb_node;
250
251         if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
252             (start_blk + count < start_blk) ||
253             (start_blk + count > ext4_blocks_count(sbi->s_es))) {
254                 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
255                 return 0;
256         }
257         while (n) {
258                 entry = rb_entry(n, struct ext4_system_zone, node);
259                 if (start_blk + count - 1 < entry->start_blk)
260                         n = n->rb_left;
261                 else if (start_blk >= (entry->start_blk + entry->count))
262                         n = n->rb_right;
263                 else {
264                         sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
265                         return 0;
266                 }
267         }
268         return 1;
269 }
270
271 int ext4_check_blockref(const char *function, unsigned int line,
272                         struct inode *inode, __le32 *p, unsigned int max)
273 {
274         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
275         __le32 *bref = p;
276         unsigned int blk;
277
278         while (bref < p+max) {
279                 blk = le32_to_cpu(*bref++);
280                 if (blk &&
281                     unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
282                                                     blk, 1))) {
283                         es->s_last_error_block = cpu_to_le64(blk);
284                         ext4_error_inode(inode, function, line, blk,
285                                          "invalid block");
286                         return -EFSCORRUPTED;
287                 }
288         }
289         return 0;
290 }
291