Loading...
1/*
2 * linux/fs/ext4/block_validity.c
3 *
4 * Copyright (C) 2009
5 * Theodore Ts'o (tytso@mit.edu)
6 *
7 * Track which blocks in the filesystem are metadata blocks that
8 * should never be used as data blocks by files or directories.
9 */
10
11#include <linux/time.h>
12#include <linux/fs.h>
13#include <linux/namei.h>
14#include <linux/quotaops.h>
15#include <linux/buffer_head.h>
16#include <linux/module.h>
17#include <linux/swap.h>
18#include <linux/pagemap.h>
19#include <linux/blkdev.h>
20#include <linux/mutex.h>
21#include <linux/slab.h>
22#include "ext4.h"
23
24struct ext4_system_zone {
25 struct rb_node node;
26 ext4_fsblk_t start_blk;
27 unsigned int count;
28};
29
30static struct kmem_cache *ext4_system_zone_cachep;
31
32int __init ext4_init_system_zone(void)
33{
34 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 if (ext4_system_zone_cachep == NULL)
36 return -ENOMEM;
37 return 0;
38}
39
40void ext4_exit_system_zone(void)
41{
42 kmem_cache_destroy(ext4_system_zone_cachep);
43}
44
45static 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/*
54 * Mark a range of blocks as belonging to the "system zone" --- that
55 * is, filesystem metadata blocks which should never be used by
56 * inodes.
57 */
58static int add_system_zone(struct ext4_sb_info *sbi,
59 ext4_fsblk_t start_blk,
60 unsigned int count)
61{
62 struct ext4_system_zone *new_entry = NULL, *entry;
63 struct rb_node **n = &sbi->system_blks.rb_node, *node;
64 struct rb_node *parent = NULL, *new_node = NULL;
65
66 while (*n) {
67 parent = *n;
68 entry = rb_entry(parent, struct ext4_system_zone, node);
69 if (start_blk < entry->start_blk)
70 n = &(*n)->rb_left;
71 else if (start_blk >= (entry->start_blk + entry->count))
72 n = &(*n)->rb_right;
73 else {
74 if (start_blk + count > (entry->start_blk +
75 entry->count))
76 entry->count = (start_blk + count -
77 entry->start_blk);
78 new_node = *n;
79 new_entry = rb_entry(new_node, struct ext4_system_zone,
80 node);
81 break;
82 }
83 }
84
85 if (!new_entry) {
86 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
87 GFP_KERNEL);
88 if (!new_entry)
89 return -ENOMEM;
90 new_entry->start_blk = start_blk;
91 new_entry->count = count;
92 new_node = &new_entry->node;
93
94 rb_link_node(new_node, parent, n);
95 rb_insert_color(new_node, &sbi->system_blks);
96 }
97
98 /* Can we merge to the left? */
99 node = rb_prev(new_node);
100 if (node) {
101 entry = rb_entry(node, struct ext4_system_zone, node);
102 if (can_merge(entry, new_entry)) {
103 new_entry->start_blk = entry->start_blk;
104 new_entry->count += entry->count;
105 rb_erase(node, &sbi->system_blks);
106 kmem_cache_free(ext4_system_zone_cachep, entry);
107 }
108 }
109
110 /* Can we merge to the right? */
111 node = rb_next(new_node);
112 if (node) {
113 entry = rb_entry(node, struct ext4_system_zone, node);
114 if (can_merge(new_entry, entry)) {
115 new_entry->count += entry->count;
116 rb_erase(node, &sbi->system_blks);
117 kmem_cache_free(ext4_system_zone_cachep, entry);
118 }
119 }
120 return 0;
121}
122
123static void debug_print_tree(struct ext4_sb_info *sbi)
124{
125 struct rb_node *node;
126 struct ext4_system_zone *entry;
127 int first = 1;
128
129 printk(KERN_INFO "System zones: ");
130 node = rb_first(&sbi->system_blks);
131 while (node) {
132 entry = rb_entry(node, struct ext4_system_zone, node);
133 printk("%s%llu-%llu", first ? "" : ", ",
134 entry->start_blk, entry->start_blk + entry->count - 1);
135 first = 0;
136 node = rb_next(node);
137 }
138 printk("\n");
139}
140
141int ext4_setup_system_zone(struct super_block *sb)
142{
143 ext4_group_t ngroups = ext4_get_groups_count(sb);
144 struct ext4_sb_info *sbi = EXT4_SB(sb);
145 struct ext4_group_desc *gdp;
146 ext4_group_t i;
147 int flex_size = ext4_flex_bg_size(sbi);
148 int ret;
149
150 if (!test_opt(sb, BLOCK_VALIDITY)) {
151 if (EXT4_SB(sb)->system_blks.rb_node)
152 ext4_release_system_zone(sb);
153 return 0;
154 }
155 if (EXT4_SB(sb)->system_blks.rb_node)
156 return 0;
157
158 for (i=0; i < ngroups; i++) {
159 if (ext4_bg_has_super(sb, i) &&
160 ((i < 5) || ((i % flex_size) == 0)))
161 add_system_zone(sbi, ext4_group_first_block_no(sb, i),
162 ext4_bg_num_gdb(sb, i) + 1);
163 gdp = ext4_get_group_desc(sb, i, NULL);
164 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1);
165 if (ret)
166 return ret;
167 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1);
168 if (ret)
169 return ret;
170 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
171 sbi->s_itb_per_group);
172 if (ret)
173 return ret;
174 }
175
176 if (test_opt(sb, DEBUG))
177 debug_print_tree(EXT4_SB(sb));
178 return 0;
179}
180
181/* Called when the filesystem is unmounted */
182void ext4_release_system_zone(struct super_block *sb)
183{
184 struct rb_node *n = EXT4_SB(sb)->system_blks.rb_node;
185 struct rb_node *parent;
186 struct ext4_system_zone *entry;
187
188 while (n) {
189 /* Do the node's children first */
190 if (n->rb_left) {
191 n = n->rb_left;
192 continue;
193 }
194 if (n->rb_right) {
195 n = n->rb_right;
196 continue;
197 }
198 /*
199 * The node has no children; free it, and then zero
200 * out parent's link to it. Finally go to the
201 * beginning of the loop and try to free the parent
202 * node.
203 */
204 parent = rb_parent(n);
205 entry = rb_entry(n, struct ext4_system_zone, node);
206 kmem_cache_free(ext4_system_zone_cachep, entry);
207 if (!parent)
208 EXT4_SB(sb)->system_blks = RB_ROOT;
209 else if (parent->rb_left == n)
210 parent->rb_left = NULL;
211 else if (parent->rb_right == n)
212 parent->rb_right = NULL;
213 n = parent;
214 }
215 EXT4_SB(sb)->system_blks = RB_ROOT;
216}
217
218/*
219 * Returns 1 if the passed-in block region (start_blk,
220 * start_blk+count) is valid; 0 if some part of the block region
221 * overlaps with filesystem metadata blocks.
222 */
223int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
224 unsigned int count)
225{
226 struct ext4_system_zone *entry;
227 struct rb_node *n = sbi->system_blks.rb_node;
228
229 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
230 (start_blk + count < start_blk) ||
231 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
232 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
233 return 0;
234 }
235 while (n) {
236 entry = rb_entry(n, struct ext4_system_zone, node);
237 if (start_blk + count - 1 < entry->start_blk)
238 n = n->rb_left;
239 else if (start_blk >= (entry->start_blk + entry->count))
240 n = n->rb_right;
241 else {
242 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
243 return 0;
244 }
245 }
246 return 1;
247}
248
249int ext4_check_blockref(const char *function, unsigned int line,
250 struct inode *inode, __le32 *p, unsigned int max)
251{
252 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
253 __le32 *bref = p;
254 unsigned int blk;
255
256 while (bref < p+max) {
257 blk = le32_to_cpu(*bref++);
258 if (blk &&
259 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
260 blk, 1))) {
261 es->s_last_error_block = cpu_to_le64(blk);
262 ext4_error_inode(inode, function, line, blk,
263 "invalid block");
264 return -EIO;
265 }
266 }
267 return 0;
268}
269
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
23struct ext4_system_zone {
24 struct rb_node node;
25 ext4_fsblk_t start_blk;
26 unsigned int count;
27 u32 ino;
28};
29
30static struct kmem_cache *ext4_system_zone_cachep;
31
32int __init ext4_init_system_zone(void)
33{
34 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 if (ext4_system_zone_cachep == NULL)
36 return -ENOMEM;
37 return 0;
38}
39
40void ext4_exit_system_zone(void)
41{
42 rcu_barrier();
43 kmem_cache_destroy(ext4_system_zone_cachep);
44}
45
46static inline int can_merge(struct ext4_system_zone *entry1,
47 struct ext4_system_zone *entry2)
48{
49 if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
50 entry1->ino == entry2->ino)
51 return 1;
52 return 0;
53}
54
55static void release_system_zone(struct ext4_system_blocks *system_blks)
56{
57 struct ext4_system_zone *entry, *n;
58
59 rbtree_postorder_for_each_entry_safe(entry, n,
60 &system_blks->root, node)
61 kmem_cache_free(ext4_system_zone_cachep, entry);
62}
63
64/*
65 * Mark a range of blocks as belonging to the "system zone" --- that
66 * is, filesystem metadata blocks which should never be used by
67 * inodes.
68 */
69static int add_system_zone(struct ext4_system_blocks *system_blks,
70 ext4_fsblk_t start_blk,
71 unsigned int count, u32 ino)
72{
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;
76
77 while (*n) {
78 parent = *n;
79 entry = rb_entry(parent, struct ext4_system_zone, node);
80 if (start_blk < entry->start_blk)
81 n = &(*n)->rb_left;
82 else if (start_blk >= (entry->start_blk + entry->count))
83 n = &(*n)->rb_right;
84 else /* Unexpected overlap of system zones. */
85 return -EFSCORRUPTED;
86 }
87
88 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
89 GFP_KERNEL);
90 if (!new_entry)
91 return -ENOMEM;
92 new_entry->start_blk = start_blk;
93 new_entry->count = count;
94 new_entry->ino = ino;
95 new_node = &new_entry->node;
96
97 rb_link_node(new_node, parent, n);
98 rb_insert_color(new_node, &system_blks->root);
99
100 /* Can we merge to the left? */
101 node = rb_prev(new_node);
102 if (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);
109 }
110 }
111
112 /* Can we merge to the right? */
113 node = rb_next(new_node);
114 if (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);
120 }
121 }
122 return 0;
123}
124
125static void debug_print_tree(struct ext4_sb_info *sbi)
126{
127 struct rb_node *node;
128 struct ext4_system_zone *entry;
129 struct ext4_system_blocks *system_blks;
130 int first = 1;
131
132 printk(KERN_INFO "System zones: ");
133 rcu_read_lock();
134 system_blks = rcu_dereference(sbi->system_blks);
135 node = rb_first(&system_blks->root);
136 while (node) {
137 entry = rb_entry(node, struct ext4_system_zone, node);
138 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
139 entry->start_blk, entry->start_blk + entry->count - 1);
140 first = 0;
141 node = rb_next(node);
142 }
143 rcu_read_unlock();
144 printk(KERN_CONT "\n");
145}
146
147static int ext4_protect_reserved_inode(struct super_block *sb,
148 struct ext4_system_blocks *system_blks,
149 u32 ino)
150{
151 struct inode *inode;
152 struct ext4_sb_info *sbi = EXT4_SB(sb);
153 struct ext4_map_blocks map;
154 u32 i = 0, num;
155 int err = 0, n;
156
157 if ((ino < EXT4_ROOT_INO) ||
158 (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
159 return -EINVAL;
160 inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
161 if (IS_ERR(inode))
162 return PTR_ERR(inode);
163 num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
164 while (i < num) {
165 cond_resched();
166 map.m_lblk = i;
167 map.m_len = num - i;
168 n = ext4_map_blocks(NULL, inode, &map, 0);
169 if (n < 0) {
170 err = n;
171 break;
172 }
173 if (n == 0) {
174 i++;
175 } else {
176 err = add_system_zone(system_blks, map.m_pblk, n, ino);
177 if (err < 0) {
178 if (err == -EFSCORRUPTED) {
179 __ext4_error(sb, __func__, __LINE__,
180 -err, map.m_pblk,
181 "blocks %llu-%llu from inode %u overlap system zone",
182 map.m_pblk,
183 map.m_pblk + map.m_len - 1,
184 ino);
185 }
186 break;
187 }
188 i += n;
189 }
190 }
191 iput(inode);
192 return err;
193}
194
195static void ext4_destroy_system_zone(struct rcu_head *rcu)
196{
197 struct ext4_system_blocks *system_blks;
198
199 system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
200 release_system_zone(system_blks);
201 kfree(system_blks);
202}
203
204/*
205 * Build system zone rbtree which is used for block validity checking.
206 *
207 * The update of system_blks pointer in this function is protected by
208 * sb->s_umount semaphore. However we have to be careful as we can be
209 * racing with ext4_data_block_valid() calls reading system_blks rbtree
210 * protected only by RCU. That's why we first build the rbtree and then
211 * swap it in place.
212 */
213int ext4_setup_system_zone(struct super_block *sb)
214{
215 ext4_group_t ngroups = ext4_get_groups_count(sb);
216 struct ext4_sb_info *sbi = EXT4_SB(sb);
217 struct ext4_system_blocks *system_blks;
218 struct ext4_group_desc *gdp;
219 ext4_group_t i;
220 int flex_size = ext4_flex_bg_size(sbi);
221 int ret;
222
223 system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
224 if (!system_blks)
225 return -ENOMEM;
226
227 for (i=0; i < ngroups; i++) {
228 cond_resched();
229 if (ext4_bg_has_super(sb, i) &&
230 ((i < 5) || ((i % flex_size) == 0))) {
231 ret = add_system_zone(system_blks,
232 ext4_group_first_block_no(sb, i),
233 ext4_bg_num_gdb(sb, i) + 1, 0);
234 if (ret)
235 goto err;
236 }
237 gdp = ext4_get_group_desc(sb, i, NULL);
238 ret = add_system_zone(system_blks,
239 ext4_block_bitmap(sb, gdp), 1, 0);
240 if (ret)
241 goto err;
242 ret = add_system_zone(system_blks,
243 ext4_inode_bitmap(sb, gdp), 1, 0);
244 if (ret)
245 goto err;
246 ret = add_system_zone(system_blks,
247 ext4_inode_table(sb, gdp),
248 sbi->s_itb_per_group, 0);
249 if (ret)
250 goto err;
251 }
252 if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
253 ret = ext4_protect_reserved_inode(sb, system_blks,
254 le32_to_cpu(sbi->s_es->s_journal_inum));
255 if (ret)
256 goto err;
257 }
258
259 /*
260 * System blks rbtree complete, announce it once to prevent racing
261 * with ext4_data_block_valid() accessing the rbtree at the same
262 * time.
263 */
264 rcu_assign_pointer(sbi->system_blks, system_blks);
265
266 if (test_opt(sb, DEBUG))
267 debug_print_tree(sbi);
268 return 0;
269err:
270 release_system_zone(system_blks);
271 kfree(system_blks);
272 return ret;
273}
274
275/*
276 * Called when the filesystem is unmounted or when remounting it with
277 * noblock_validity specified.
278 *
279 * The update of system_blks pointer in this function is protected by
280 * sb->s_umount semaphore. However we have to be careful as we can be
281 * racing with ext4_data_block_valid() calls reading system_blks rbtree
282 * protected only by RCU. So we first clear the system_blks pointer and
283 * then free the rbtree only after RCU grace period expires.
284 */
285void ext4_release_system_zone(struct super_block *sb)
286{
287 struct ext4_system_blocks *system_blks;
288
289 system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
290 lockdep_is_held(&sb->s_umount));
291 rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
292
293 if (system_blks)
294 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
295}
296
297/*
298 * Returns 1 if the passed-in block region (start_blk,
299 * start_blk+count) is valid; 0 if some part of the block region
300 * overlaps with some other filesystem metadata blocks.
301 */
302int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
303 unsigned int count)
304{
305 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
306 struct ext4_system_blocks *system_blks;
307 struct ext4_system_zone *entry;
308 struct rb_node *n;
309 int ret = 1;
310
311 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
312 (start_blk + count < start_blk) ||
313 (start_blk + count > ext4_blocks_count(sbi->s_es)))
314 return 0;
315
316 /*
317 * Lock the system zone to prevent it being released concurrently
318 * when doing a remount which inverse current "[no]block_validity"
319 * mount option.
320 */
321 rcu_read_lock();
322 system_blks = rcu_dereference(sbi->system_blks);
323 if (system_blks == NULL)
324 goto out_rcu;
325
326 n = system_blks->root.rb_node;
327 while (n) {
328 entry = rb_entry(n, struct ext4_system_zone, node);
329 if (start_blk + count - 1 < entry->start_blk)
330 n = n->rb_left;
331 else if (start_blk >= (entry->start_blk + entry->count))
332 n = n->rb_right;
333 else {
334 ret = (entry->ino == inode->i_ino);
335 break;
336 }
337 }
338out_rcu:
339 rcu_read_unlock();
340 return ret;
341}
342
343int ext4_check_blockref(const char *function, unsigned int line,
344 struct inode *inode, __le32 *p, unsigned int max)
345{
346 __le32 *bref = p;
347 unsigned int blk;
348
349 if (ext4_has_feature_journal(inode->i_sb) &&
350 (inode->i_ino ==
351 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
352 return 0;
353
354 while (bref < p+max) {
355 blk = le32_to_cpu(*bref++);
356 if (blk &&
357 unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
358 ext4_error_inode(inode, function, line, blk,
359 "invalid block");
360 return -EFSCORRUPTED;
361 }
362 }
363 return 0;
364}
365