1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/block_validity.c
6 * Theodore Ts'o (tytso@mit.edu)
8 * Track which blocks in the filesystem are metadata blocks that
9 * should never be used as data blocks by files or directories.
12 #include <linux/time.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>
23 struct ext4_system_zone {
25 ext4_fsblk_t start_blk;
30 static struct kmem_cache *ext4_system_zone_cachep;
32 int __init ext4_init_system_zone(void)
34 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 if (ext4_system_zone_cachep == NULL)
40 void ext4_exit_system_zone(void)
43 kmem_cache_destroy(ext4_system_zone_cachep);
46 static inline int can_merge(struct ext4_system_zone *entry1,
47 struct ext4_system_zone *entry2)
49 if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
50 entry1->ino == entry2->ino)
55 static void release_system_zone(struct ext4_system_blocks *system_blks)
57 struct ext4_system_zone *entry, *n;
59 rbtree_postorder_for_each_entry_safe(entry, n,
60 &system_blks->root, node)
61 kmem_cache_free(ext4_system_zone_cachep, entry);
65 * Mark a range of blocks as belonging to the "system zone" --- that
66 * is, filesystem metadata blocks which should never be used by
69 static int add_system_zone(struct ext4_system_blocks *system_blks,
70 ext4_fsblk_t start_blk,
71 unsigned int count, u32 ino)
73 struct ext4_system_zone *new_entry, *entry;
74 struct rb_node **n = &system_blks->root.rb_node, *node;
75 struct rb_node *parent = NULL, *new_node = NULL;
79 entry = rb_entry(parent, struct ext4_system_zone, node);
80 if (start_blk < entry->start_blk)
82 else if (start_blk >= (entry->start_blk + entry->count))
84 else /* Unexpected overlap of system zones. */
88 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
92 new_entry->start_blk = start_blk;
93 new_entry->count = count;
95 new_node = &new_entry->node;
97 rb_link_node(new_node, parent, n);
98 rb_insert_color(new_node, &system_blks->root);
100 /* Can we merge to the left? */
101 node = rb_prev(new_node);
103 entry = rb_entry(node, struct ext4_system_zone, node);
104 if (can_merge(entry, new_entry)) {
105 new_entry->start_blk = entry->start_blk;
106 new_entry->count += entry->count;
107 rb_erase(node, &system_blks->root);
108 kmem_cache_free(ext4_system_zone_cachep, entry);
112 /* Can we merge to the right? */
113 node = rb_next(new_node);
115 entry = rb_entry(node, struct ext4_system_zone, node);
116 if (can_merge(new_entry, entry)) {
117 new_entry->count += entry->count;
118 rb_erase(node, &system_blks->root);
119 kmem_cache_free(ext4_system_zone_cachep, entry);
125 static void debug_print_tree(struct ext4_sb_info *sbi)
127 struct rb_node *node;
128 struct ext4_system_zone *entry;
131 printk(KERN_INFO "System zones: ");
132 node = rb_first(&sbi->system_blks->root);
134 entry = rb_entry(node, struct ext4_system_zone, node);
135 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
136 entry->start_blk, entry->start_blk + entry->count - 1);
138 node = rb_next(node);
140 printk(KERN_CONT "\n");
144 * Returns 1 if the passed-in block region (start_blk,
145 * start_blk+count) is valid; 0 if some part of the block region
146 * overlaps with filesystem metadata blocks.
148 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
149 struct ext4_system_blocks *system_blks,
150 ext4_fsblk_t start_blk,
151 unsigned int count, ino_t ino)
153 struct ext4_system_zone *entry;
156 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
157 (start_blk + count < start_blk) ||
158 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
159 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
163 if (system_blks == NULL)
166 n = system_blks->root.rb_node;
168 entry = rb_entry(n, struct ext4_system_zone, node);
169 if (start_blk + count - 1 < entry->start_blk)
171 else if (start_blk >= (entry->start_blk + entry->count))
174 if (entry->ino == ino)
176 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
183 static int ext4_protect_reserved_inode(struct super_block *sb,
184 struct ext4_system_blocks *system_blks,
188 struct ext4_sb_info *sbi = EXT4_SB(sb);
189 struct ext4_map_blocks map;
193 if ((ino < EXT4_ROOT_INO) ||
194 (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
196 inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
198 return PTR_ERR(inode);
199 num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
204 n = ext4_map_blocks(NULL, inode, &map, 0);
212 err = add_system_zone(system_blks, map.m_pblk, n, ino);
214 if (err == -EFSCORRUPTED) {
216 "blocks %llu-%llu from inode %u "
217 "overlap system zone", map.m_pblk,
218 map.m_pblk + map.m_len - 1, ino);
229 static void ext4_destroy_system_zone(struct rcu_head *rcu)
231 struct ext4_system_blocks *system_blks;
233 system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
234 release_system_zone(system_blks);
239 * Build system zone rbtree which is used for block validity checking.
241 * The update of system_blks pointer in this function is protected by
242 * sb->s_umount semaphore. However we have to be careful as we can be
243 * racing with ext4_data_block_valid() calls reading system_blks rbtree
244 * protected only by RCU. That's why we first build the rbtree and then
247 int ext4_setup_system_zone(struct super_block *sb)
249 ext4_group_t ngroups = ext4_get_groups_count(sb);
250 struct ext4_sb_info *sbi = EXT4_SB(sb);
251 struct ext4_system_blocks *system_blks;
252 struct ext4_group_desc *gdp;
254 int flex_size = ext4_flex_bg_size(sbi);
257 system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
261 for (i=0; i < ngroups; i++) {
262 if (ext4_bg_has_super(sb, i) &&
263 ((i < 5) || ((i % flex_size) == 0)))
264 add_system_zone(system_blks,
265 ext4_group_first_block_no(sb, i),
266 ext4_bg_num_gdb(sb, i) + 1, 0);
267 gdp = ext4_get_group_desc(sb, i, NULL);
268 ret = add_system_zone(system_blks,
269 ext4_block_bitmap(sb, gdp), 1, 0);
272 ret = add_system_zone(system_blks,
273 ext4_inode_bitmap(sb, gdp), 1, 0);
276 ret = add_system_zone(system_blks,
277 ext4_inode_table(sb, gdp),
278 sbi->s_itb_per_group, 0);
282 if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
283 ret = ext4_protect_reserved_inode(sb, system_blks,
284 le32_to_cpu(sbi->s_es->s_journal_inum));
290 * System blks rbtree complete, announce it once to prevent racing
291 * with ext4_data_block_valid() accessing the rbtree at the same
294 rcu_assign_pointer(sbi->system_blks, system_blks);
296 if (test_opt(sb, DEBUG))
297 debug_print_tree(sbi);
300 release_system_zone(system_blks);
306 * Called when the filesystem is unmounted or when remounting it with
307 * noblock_validity specified.
309 * The update of system_blks pointer in this function is protected by
310 * sb->s_umount semaphore. However we have to be careful as we can be
311 * racing with ext4_data_block_valid() calls reading system_blks rbtree
312 * protected only by RCU. So we first clear the system_blks pointer and
313 * then free the rbtree only after RCU grace period expires.
315 void ext4_release_system_zone(struct super_block *sb)
317 struct ext4_system_blocks *system_blks;
319 system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
320 lockdep_is_held(&sb->s_umount));
321 rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
324 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
327 int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
330 struct ext4_system_blocks *system_blks;
334 * Lock the system zone to prevent it being released concurrently
335 * when doing a remount which inverse current "[no]block_validity"
339 system_blks = rcu_dereference(EXT4_SB(inode->i_sb)->system_blks);
340 ret = ext4_data_block_valid_rcu(EXT4_SB(inode->i_sb), system_blks,
341 start_blk, count, inode->i_ino);
346 int ext4_check_blockref(const char *function, unsigned int line,
347 struct inode *inode, __le32 *p, unsigned int max)
349 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
353 if (ext4_has_feature_journal(inode->i_sb) &&
355 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
358 while (bref < p+max) {
359 blk = le32_to_cpu(*bref++);
361 unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
362 es->s_last_error_block = cpu_to_le64(blk);
363 ext4_error_inode(inode, function, line, blk,
365 return -EFSCORRUPTED;