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};
28
29static struct kmem_cache *ext4_system_zone_cachep;
30
31int __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
39void ext4_exit_system_zone(void)
40{
41 rcu_barrier();
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
53static 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 */
67static 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
132static void debug_print_tree(struct ext4_sb_info *sbi)
133{
134 struct rb_node *node;
135 struct ext4_system_zone *entry;
136 int first = 1;
137
138 printk(KERN_INFO "System zones: ");
139 node = rb_first(&sbi->system_blks->root);
140 while (node) {
141 entry = rb_entry(node, struct ext4_system_zone, node);
142 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
143 entry->start_blk, entry->start_blk + entry->count - 1);
144 first = 0;
145 node = rb_next(node);
146 }
147 printk(KERN_CONT "\n");
148}
149
150/*
151 * Returns 1 if the passed-in block region (start_blk,
152 * start_blk+count) is valid; 0 if some part of the block region
153 * overlaps with filesystem metadata blocks.
154 */
155static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
156 struct ext4_system_blocks *system_blks,
157 ext4_fsblk_t start_blk,
158 unsigned int count)
159{
160 struct ext4_system_zone *entry;
161 struct rb_node *n;
162
163 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
164 (start_blk + count < start_blk) ||
165 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
166 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
167 return 0;
168 }
169
170 if (system_blks == NULL)
171 return 1;
172
173 n = system_blks->root.rb_node;
174 while (n) {
175 entry = rb_entry(n, struct ext4_system_zone, node);
176 if (start_blk + count - 1 < entry->start_blk)
177 n = n->rb_left;
178 else if (start_blk >= (entry->start_blk + entry->count))
179 n = n->rb_right;
180 else {
181 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
182 return 0;
183 }
184 }
185 return 1;
186}
187
188static int ext4_protect_reserved_inode(struct super_block *sb,
189 struct ext4_system_blocks *system_blks,
190 u32 ino)
191{
192 struct inode *inode;
193 struct ext4_sb_info *sbi = EXT4_SB(sb);
194 struct ext4_map_blocks map;
195 u32 i = 0, num;
196 int err = 0, n;
197
198 if ((ino < EXT4_ROOT_INO) ||
199 (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
200 return -EINVAL;
201 inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
202 if (IS_ERR(inode))
203 return PTR_ERR(inode);
204 num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
205 while (i < num) {
206 map.m_lblk = i;
207 map.m_len = num - i;
208 n = ext4_map_blocks(NULL, inode, &map, 0);
209 if (n < 0) {
210 err = n;
211 break;
212 }
213 if (n == 0) {
214 i++;
215 } else {
216 if (!ext4_data_block_valid_rcu(sbi, system_blks,
217 map.m_pblk, n)) {
218 ext4_error(sb, "blocks %llu-%llu from inode %u "
219 "overlap system zone", map.m_pblk,
220 map.m_pblk + map.m_len - 1, ino);
221 err = -EFSCORRUPTED;
222 break;
223 }
224 err = add_system_zone(system_blks, map.m_pblk, n);
225 if (err < 0)
226 break;
227 i += n;
228 }
229 }
230 iput(inode);
231 return err;
232}
233
234static void ext4_destroy_system_zone(struct rcu_head *rcu)
235{
236 struct ext4_system_blocks *system_blks;
237
238 system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
239 release_system_zone(system_blks);
240 kfree(system_blks);
241}
242
243/*
244 * Build system zone rbtree which is used for block validity checking.
245 *
246 * The update of system_blks pointer in this function is protected by
247 * sb->s_umount semaphore. However we have to be careful as we can be
248 * racing with ext4_data_block_valid() calls reading system_blks rbtree
249 * protected only by RCU. That's why we first build the rbtree and then
250 * swap it in place.
251 */
252int ext4_setup_system_zone(struct super_block *sb)
253{
254 ext4_group_t ngroups = ext4_get_groups_count(sb);
255 struct ext4_sb_info *sbi = EXT4_SB(sb);
256 struct ext4_system_blocks *system_blks;
257 struct ext4_group_desc *gdp;
258 ext4_group_t i;
259 int flex_size = ext4_flex_bg_size(sbi);
260 int ret;
261
262 if (!test_opt(sb, BLOCK_VALIDITY)) {
263 if (sbi->system_blks)
264 ext4_release_system_zone(sb);
265 return 0;
266 }
267 if (sbi->system_blks)
268 return 0;
269
270 system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
271 if (!system_blks)
272 return -ENOMEM;
273
274 for (i=0; i < ngroups; i++) {
275 cond_resched();
276 if (ext4_bg_has_super(sb, i) &&
277 ((i < 5) || ((i % flex_size) == 0)))
278 add_system_zone(system_blks,
279 ext4_group_first_block_no(sb, i),
280 ext4_bg_num_gdb(sb, i) + 1);
281 gdp = ext4_get_group_desc(sb, i, NULL);
282 ret = add_system_zone(system_blks,
283 ext4_block_bitmap(sb, gdp), 1);
284 if (ret)
285 goto err;
286 ret = add_system_zone(system_blks,
287 ext4_inode_bitmap(sb, gdp), 1);
288 if (ret)
289 goto err;
290 ret = add_system_zone(system_blks,
291 ext4_inode_table(sb, gdp),
292 sbi->s_itb_per_group);
293 if (ret)
294 goto err;
295 }
296 if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
297 ret = ext4_protect_reserved_inode(sb, system_blks,
298 le32_to_cpu(sbi->s_es->s_journal_inum));
299 if (ret)
300 goto err;
301 }
302
303 /*
304 * System blks rbtree complete, announce it once to prevent racing
305 * with ext4_data_block_valid() accessing the rbtree at the same
306 * time.
307 */
308 rcu_assign_pointer(sbi->system_blks, system_blks);
309
310 if (test_opt(sb, DEBUG))
311 debug_print_tree(sbi);
312 return 0;
313err:
314 release_system_zone(system_blks);
315 kfree(system_blks);
316 return ret;
317}
318
319/*
320 * Called when the filesystem is unmounted or when remounting it with
321 * noblock_validity specified.
322 *
323 * The update of system_blks pointer in this function is protected by
324 * sb->s_umount semaphore. However we have to be careful as we can be
325 * racing with ext4_data_block_valid() calls reading system_blks rbtree
326 * protected only by RCU. So we first clear the system_blks pointer and
327 * then free the rbtree only after RCU grace period expires.
328 */
329void ext4_release_system_zone(struct super_block *sb)
330{
331 struct ext4_system_blocks *system_blks;
332
333 system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
334 lockdep_is_held(&sb->s_umount));
335 rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
336
337 if (system_blks)
338 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
339}
340
341int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
342 unsigned int count)
343{
344 struct ext4_system_blocks *system_blks;
345 int ret;
346
347 /*
348 * Lock the system zone to prevent it being released concurrently
349 * when doing a remount which inverse current "[no]block_validity"
350 * mount option.
351 */
352 rcu_read_lock();
353 system_blks = rcu_dereference(sbi->system_blks);
354 ret = ext4_data_block_valid_rcu(sbi, system_blks, start_blk,
355 count);
356 rcu_read_unlock();
357 return ret;
358}
359
360int ext4_check_blockref(const char *function, unsigned int line,
361 struct inode *inode, __le32 *p, unsigned int max)
362{
363 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
364 __le32 *bref = p;
365 unsigned int blk;
366
367 if (ext4_has_feature_journal(inode->i_sb) &&
368 (inode->i_ino ==
369 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
370 return 0;
371
372 while (bref < p+max) {
373 blk = le32_to_cpu(*bref++);
374 if (blk &&
375 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
376 blk, 1))) {
377 es->s_last_error_block = cpu_to_le64(blk);
378 ext4_error_inode(inode, function, line, blk,
379 "invalid block");
380 return -EFSCORRUPTED;
381 }
382 }
383 return 0;
384}
385