xref: /openbmc/linux/fs/ext4/block_validity.c (revision f9834f18)
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 	rcu_barrier();
42 	kmem_cache_destroy(ext4_system_zone_cachep);
43 }
44 
45 static inline int can_merge(struct ext4_system_zone *entry1,
46 		     struct ext4_system_zone *entry2)
47 {
48 	if ((entry1->start_blk + entry1->count) == entry2->start_blk)
49 		return 1;
50 	return 0;
51 }
52 
53 static void release_system_zone(struct ext4_system_blocks *system_blks)
54 {
55 	struct ext4_system_zone	*entry, *n;
56 
57 	rbtree_postorder_for_each_entry_safe(entry, n,
58 				&system_blks->root, node)
59 		kmem_cache_free(ext4_system_zone_cachep, entry);
60 }
61 
62 /*
63  * Mark a range of blocks as belonging to the "system zone" --- that
64  * is, filesystem metadata blocks which should never be used by
65  * inodes.
66  */
67 static int add_system_zone(struct ext4_system_blocks *system_blks,
68 			   ext4_fsblk_t start_blk,
69 			   unsigned int count)
70 {
71 	struct ext4_system_zone *new_entry = NULL, *entry;
72 	struct rb_node **n = &system_blks->root.rb_node, *node;
73 	struct rb_node *parent = NULL, *new_node = NULL;
74 
75 	while (*n) {
76 		parent = *n;
77 		entry = rb_entry(parent, struct ext4_system_zone, node);
78 		if (start_blk < entry->start_blk)
79 			n = &(*n)->rb_left;
80 		else if (start_blk >= (entry->start_blk + entry->count))
81 			n = &(*n)->rb_right;
82 		else {
83 			if (start_blk + count > (entry->start_blk +
84 						 entry->count))
85 				entry->count = (start_blk + count -
86 						entry->start_blk);
87 			new_node = *n;
88 			new_entry = rb_entry(new_node, struct ext4_system_zone,
89 					     node);
90 			break;
91 		}
92 	}
93 
94 	if (!new_entry) {
95 		new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
96 					     GFP_KERNEL);
97 		if (!new_entry)
98 			return -ENOMEM;
99 		new_entry->start_blk = start_blk;
100 		new_entry->count = count;
101 		new_node = &new_entry->node;
102 
103 		rb_link_node(new_node, parent, n);
104 		rb_insert_color(new_node, &system_blks->root);
105 	}
106 
107 	/* Can we merge to the left? */
108 	node = rb_prev(new_node);
109 	if (node) {
110 		entry = rb_entry(node, struct ext4_system_zone, node);
111 		if (can_merge(entry, new_entry)) {
112 			new_entry->start_blk = entry->start_blk;
113 			new_entry->count += entry->count;
114 			rb_erase(node, &system_blks->root);
115 			kmem_cache_free(ext4_system_zone_cachep, entry);
116 		}
117 	}
118 
119 	/* Can we merge to the right? */
120 	node = rb_next(new_node);
121 	if (node) {
122 		entry = rb_entry(node, struct ext4_system_zone, node);
123 		if (can_merge(new_entry, entry)) {
124 			new_entry->count += entry->count;
125 			rb_erase(node, &system_blks->root);
126 			kmem_cache_free(ext4_system_zone_cachep, entry);
127 		}
128 	}
129 	return 0;
130 }
131 
132 static void debug_print_tree(struct ext4_sb_info *sbi)
133 {
134 	struct rb_node *node;
135 	struct ext4_system_zone *entry;
136 	struct ext4_system_blocks *system_blks;
137 	int first = 1;
138 
139 	printk(KERN_INFO "System zones: ");
140 	rcu_read_lock();
141 	system_blks = rcu_dereference(sbi->system_blks);
142 	node = rb_first(&system_blks->root);
143 	while (node) {
144 		entry = rb_entry(node, struct ext4_system_zone, node);
145 		printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
146 		       entry->start_blk, entry->start_blk + entry->count - 1);
147 		first = 0;
148 		node = rb_next(node);
149 	}
150 	rcu_read_unlock();
151 	printk(KERN_CONT "\n");
152 }
153 
154 /*
155  * Returns 1 if the passed-in block region (start_blk,
156  * start_blk+count) is valid; 0 if some part of the block region
157  * overlaps with filesystem metadata blocks.
158  */
159 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
160 				     struct ext4_system_blocks *system_blks,
161 				     ext4_fsblk_t start_blk,
162 				     unsigned int count)
163 {
164 	struct ext4_system_zone *entry;
165 	struct rb_node *n;
166 
167 	if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
168 	    (start_blk + count < start_blk) ||
169 	    (start_blk + count > ext4_blocks_count(sbi->s_es))) {
170 		sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
171 		return 0;
172 	}
173 
174 	if (system_blks == NULL)
175 		return 1;
176 
177 	n = system_blks->root.rb_node;
178 	while (n) {
179 		entry = rb_entry(n, struct ext4_system_zone, node);
180 		if (start_blk + count - 1 < entry->start_blk)
181 			n = n->rb_left;
182 		else if (start_blk >= (entry->start_blk + entry->count))
183 			n = n->rb_right;
184 		else {
185 			sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
186 			return 0;
187 		}
188 	}
189 	return 1;
190 }
191 
192 static int ext4_protect_reserved_inode(struct super_block *sb,
193 				       struct ext4_system_blocks *system_blks,
194 				       u32 ino)
195 {
196 	struct inode *inode;
197 	struct ext4_sb_info *sbi = EXT4_SB(sb);
198 	struct ext4_map_blocks map;
199 	u32 i = 0, num;
200 	int err = 0, n;
201 
202 	if ((ino < EXT4_ROOT_INO) ||
203 	    (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
204 		return -EINVAL;
205 	inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
206 	if (IS_ERR(inode))
207 		return PTR_ERR(inode);
208 	num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
209 	while (i < num) {
210 		cond_resched();
211 		map.m_lblk = i;
212 		map.m_len = num - i;
213 		n = ext4_map_blocks(NULL, inode, &map, 0);
214 		if (n < 0) {
215 			err = n;
216 			break;
217 		}
218 		if (n == 0) {
219 			i++;
220 		} else {
221 			if (!ext4_data_block_valid_rcu(sbi, system_blks,
222 						map.m_pblk, n)) {
223 				ext4_error(sb, "blocks %llu-%llu from inode %u "
224 					   "overlap system zone", map.m_pblk,
225 					   map.m_pblk + map.m_len - 1, ino);
226 				err = -EFSCORRUPTED;
227 				break;
228 			}
229 			err = add_system_zone(system_blks, map.m_pblk, n);
230 			if (err < 0)
231 				break;
232 			i += n;
233 		}
234 	}
235 	iput(inode);
236 	return err;
237 }
238 
239 static void ext4_destroy_system_zone(struct rcu_head *rcu)
240 {
241 	struct ext4_system_blocks *system_blks;
242 
243 	system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
244 	release_system_zone(system_blks);
245 	kfree(system_blks);
246 }
247 
248 /*
249  * Build system zone rbtree which is used for block validity checking.
250  *
251  * The update of system_blks pointer in this function is protected by
252  * sb->s_umount semaphore. However we have to be careful as we can be
253  * racing with ext4_data_block_valid() calls reading system_blks rbtree
254  * protected only by RCU. That's why we first build the rbtree and then
255  * swap it in place.
256  */
257 int ext4_setup_system_zone(struct super_block *sb)
258 {
259 	ext4_group_t ngroups = ext4_get_groups_count(sb);
260 	struct ext4_sb_info *sbi = EXT4_SB(sb);
261 	struct ext4_system_blocks *system_blks;
262 	struct ext4_group_desc *gdp;
263 	ext4_group_t i;
264 	int flex_size = ext4_flex_bg_size(sbi);
265 	int ret;
266 
267 	if (!test_opt(sb, BLOCK_VALIDITY)) {
268 		if (sbi->system_blks)
269 			ext4_release_system_zone(sb);
270 		return 0;
271 	}
272 	if (sbi->system_blks)
273 		return 0;
274 
275 	system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
276 	if (!system_blks)
277 		return -ENOMEM;
278 
279 	for (i=0; i < ngroups; i++) {
280 		cond_resched();
281 		if (ext4_bg_has_super(sb, i) &&
282 		    ((i < 5) || ((i % flex_size) == 0)))
283 			add_system_zone(system_blks,
284 					ext4_group_first_block_no(sb, i),
285 					ext4_bg_num_gdb(sb, i) + 1);
286 		gdp = ext4_get_group_desc(sb, i, NULL);
287 		ret = add_system_zone(system_blks,
288 				ext4_block_bitmap(sb, gdp), 1);
289 		if (ret)
290 			goto err;
291 		ret = add_system_zone(system_blks,
292 				ext4_inode_bitmap(sb, gdp), 1);
293 		if (ret)
294 			goto err;
295 		ret = add_system_zone(system_blks,
296 				ext4_inode_table(sb, gdp),
297 				sbi->s_itb_per_group);
298 		if (ret)
299 			goto err;
300 	}
301 	if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
302 		ret = ext4_protect_reserved_inode(sb, system_blks,
303 				le32_to_cpu(sbi->s_es->s_journal_inum));
304 		if (ret)
305 			goto err;
306 	}
307 
308 	/*
309 	 * System blks rbtree complete, announce it once to prevent racing
310 	 * with ext4_data_block_valid() accessing the rbtree at the same
311 	 * time.
312 	 */
313 	rcu_assign_pointer(sbi->system_blks, system_blks);
314 
315 	if (test_opt(sb, DEBUG))
316 		debug_print_tree(sbi);
317 	return 0;
318 err:
319 	release_system_zone(system_blks);
320 	kfree(system_blks);
321 	return ret;
322 }
323 
324 /*
325  * Called when the filesystem is unmounted or when remounting it with
326  * noblock_validity specified.
327  *
328  * The update of system_blks pointer in this function is protected by
329  * sb->s_umount semaphore. However we have to be careful as we can be
330  * racing with ext4_data_block_valid() calls reading system_blks rbtree
331  * protected only by RCU. So we first clear the system_blks pointer and
332  * then free the rbtree only after RCU grace period expires.
333  */
334 void ext4_release_system_zone(struct super_block *sb)
335 {
336 	struct ext4_system_blocks *system_blks;
337 
338 	system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
339 					lockdep_is_held(&sb->s_umount));
340 	rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
341 
342 	if (system_blks)
343 		call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
344 }
345 
346 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
347 			  unsigned int count)
348 {
349 	struct ext4_system_blocks *system_blks;
350 	int ret;
351 
352 	/*
353 	 * Lock the system zone to prevent it being released concurrently
354 	 * when doing a remount which inverse current "[no]block_validity"
355 	 * mount option.
356 	 */
357 	rcu_read_lock();
358 	system_blks = rcu_dereference(sbi->system_blks);
359 	ret = ext4_data_block_valid_rcu(sbi, system_blks, start_blk,
360 					count);
361 	rcu_read_unlock();
362 	return ret;
363 }
364 
365 int ext4_check_blockref(const char *function, unsigned int line,
366 			struct inode *inode, __le32 *p, unsigned int max)
367 {
368 	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
369 	__le32 *bref = p;
370 	unsigned int blk;
371 
372 	if (ext4_has_feature_journal(inode->i_sb) &&
373 	    (inode->i_ino ==
374 	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
375 		return 0;
376 
377 	while (bref < p+max) {
378 		blk = le32_to_cpu(*bref++);
379 		if (blk &&
380 		    unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
381 						    blk, 1))) {
382 			es->s_last_error_block = cpu_to_le64(blk);
383 			ext4_error_inode(inode, function, line, blk,
384 					 "invalid block");
385 			return -EFSCORRUPTED;
386 		}
387 	}
388 	return 0;
389 }
390 
391