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;
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)
42 kmem_cache_destroy(ext4_system_zone_cachep);
45 static inline int can_merge(struct ext4_system_zone *entry1,
46 struct ext4_system_zone *entry2)
48 if ((entry1->start_blk + entry1->count) == entry2->start_blk)
53 static void release_system_zone(struct ext4_system_blocks *system_blks)
55 struct ext4_system_zone *entry, *n;
57 rbtree_postorder_for_each_entry_safe(entry, n,
58 &system_blks->root, node)
59 kmem_cache_free(ext4_system_zone_cachep, entry);
63 * Mark a range of blocks as belonging to the "system zone" --- that
64 * is, filesystem metadata blocks which should never be used by
67 static int add_system_zone(struct ext4_system_blocks *system_blks,
68 ext4_fsblk_t start_blk,
71 struct ext4_system_zone *new_entry, *entry;
72 struct rb_node **n = &system_blks->root.rb_node, *node;
73 struct rb_node *parent = NULL, *new_node = NULL;
77 entry = rb_entry(parent, struct ext4_system_zone, node);
78 if (start_blk < entry->start_blk)
80 else if (start_blk >= (entry->start_blk + entry->count))
82 else /* Unexpected overlap of system zones. */
86 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
90 new_entry->start_blk = start_blk;
91 new_entry->count = count;
92 new_node = &new_entry->node;
94 rb_link_node(new_node, parent, n);
95 rb_insert_color(new_node, &system_blks->root);
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, &system_blks->root);
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, &system_blks->root);
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->root);
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);
135 node = rb_next(node);
137 printk(KERN_CONT "\n");
141 * Returns 1 if the passed-in block region (start_blk,
142 * start_blk+count) is valid; 0 if some part of the block region
143 * overlaps with filesystem metadata blocks.
145 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
146 struct ext4_system_blocks *system_blks,
147 ext4_fsblk_t start_blk,
150 struct ext4_system_zone *entry;
153 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
154 (start_blk + count < start_blk) ||
155 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
156 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
160 if (system_blks == NULL)
163 n = system_blks->root.rb_node;
165 entry = rb_entry(n, struct ext4_system_zone, node);
166 if (start_blk + count - 1 < entry->start_blk)
168 else if (start_blk >= (entry->start_blk + entry->count))
171 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
178 static int ext4_protect_reserved_inode(struct super_block *sb,
179 struct ext4_system_blocks *system_blks,
183 struct ext4_sb_info *sbi = EXT4_SB(sb);
184 struct ext4_map_blocks map;
188 if ((ino < EXT4_ROOT_INO) ||
189 (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
191 inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
193 return PTR_ERR(inode);
194 num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
199 n = ext4_map_blocks(NULL, inode, &map, 0);
207 if (!ext4_data_block_valid_rcu(sbi, system_blks,
209 ext4_error(sb, "blocks %llu-%llu from inode %u "
210 "overlap system zone", map.m_pblk,
211 map.m_pblk + map.m_len - 1, ino);
215 err = add_system_zone(system_blks, map.m_pblk, n);
225 static void ext4_destroy_system_zone(struct rcu_head *rcu)
227 struct ext4_system_blocks *system_blks;
229 system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
230 release_system_zone(system_blks);
235 * Build system zone rbtree which is used for block validity checking.
237 * The update of system_blks pointer in this function is protected by
238 * sb->s_umount semaphore. However we have to be careful as we can be
239 * racing with ext4_data_block_valid() calls reading system_blks rbtree
240 * protected only by RCU. That's why we first build the rbtree and then
243 int ext4_setup_system_zone(struct super_block *sb)
245 ext4_group_t ngroups = ext4_get_groups_count(sb);
246 struct ext4_sb_info *sbi = EXT4_SB(sb);
247 struct ext4_system_blocks *system_blks;
248 struct ext4_group_desc *gdp;
250 int flex_size = ext4_flex_bg_size(sbi);
253 system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
257 for (i=0; i < ngroups; i++) {
259 if (ext4_bg_has_super(sb, i) &&
260 ((i < 5) || ((i % flex_size) == 0)))
261 add_system_zone(system_blks,
262 ext4_group_first_block_no(sb, i),
263 ext4_bg_num_gdb(sb, i) + 1);
264 gdp = ext4_get_group_desc(sb, i, NULL);
265 ret = add_system_zone(system_blks,
266 ext4_block_bitmap(sb, gdp), 1);
269 ret = add_system_zone(system_blks,
270 ext4_inode_bitmap(sb, gdp), 1);
273 ret = add_system_zone(system_blks,
274 ext4_inode_table(sb, gdp),
275 sbi->s_itb_per_group);
279 if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
280 ret = ext4_protect_reserved_inode(sb, system_blks,
281 le32_to_cpu(sbi->s_es->s_journal_inum));
287 * System blks rbtree complete, announce it once to prevent racing
288 * with ext4_data_block_valid() accessing the rbtree at the same
291 rcu_assign_pointer(sbi->system_blks, system_blks);
293 if (test_opt(sb, DEBUG))
294 debug_print_tree(sbi);
297 release_system_zone(system_blks);
303 * Called when the filesystem is unmounted or when remounting it with
304 * noblock_validity specified.
306 * The update of system_blks pointer in this function is protected by
307 * sb->s_umount semaphore. However we have to be careful as we can be
308 * racing with ext4_data_block_valid() calls reading system_blks rbtree
309 * protected only by RCU. So we first clear the system_blks pointer and
310 * then free the rbtree only after RCU grace period expires.
312 void ext4_release_system_zone(struct super_block *sb)
314 struct ext4_system_blocks *system_blks;
316 system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
317 lockdep_is_held(&sb->s_umount));
318 rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
321 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
324 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
327 struct ext4_system_blocks *system_blks;
331 * Lock the system zone to prevent it being released concurrently
332 * when doing a remount which inverse current "[no]block_validity"
336 system_blks = rcu_dereference(sbi->system_blks);
337 ret = ext4_data_block_valid_rcu(sbi, system_blks, start_blk,
343 int ext4_check_blockref(const char *function, unsigned int line,
344 struct inode *inode, __le32 *p, unsigned int max)
346 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
350 if (ext4_has_feature_journal(inode->i_sb) &&
352 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
355 while (bref < p+max) {
356 blk = le32_to_cpu(*bref++);
358 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
360 es->s_last_error_block = cpu_to_le64(blk);
361 ext4_error_inode(inode, function, line, blk,
363 return -EFSCORRUPTED;