Line data Source code
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 : u32 ino;
28 : };
29 :
30 : static struct kmem_cache *ext4_system_zone_cachep;
31 :
32 1 : int __init ext4_init_system_zone(void)
33 : {
34 1 : ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 1 : if (ext4_system_zone_cachep == NULL)
36 0 : return -ENOMEM;
37 : return 0;
38 : }
39 :
40 0 : void ext4_exit_system_zone(void)
41 : {
42 0 : rcu_barrier();
43 0 : kmem_cache_destroy(ext4_system_zone_cachep);
44 0 : }
45 :
46 100 : static inline int can_merge(struct ext4_system_zone *entry1,
47 : struct ext4_system_zone *entry2)
48 : {
49 147 : if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
50 47 : entry1->ino == entry2->ino)
51 47 : return 1;
52 : return 0;
53 : }
54 :
55 0 : static void release_system_zone(struct ext4_system_blocks *system_blks)
56 : {
57 0 : struct ext4_system_zone *entry, *n;
58 :
59 0 : rbtree_postorder_for_each_entry_safe(entry, n,
60 : &system_blks->root, node)
61 0 : kmem_cache_free(ext4_system_zone_cachep, entry);
62 0 : }
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 : */
69 55 : static int add_system_zone(struct ext4_system_blocks *system_blks,
70 : ext4_fsblk_t start_blk,
71 : unsigned int count, u32 ino)
72 : {
73 55 : struct ext4_system_zone *new_entry, *entry;
74 55 : struct rb_node **n = &system_blks->root.rb_node, *node;
75 55 : struct rb_node *parent = NULL, *new_node = NULL;
76 :
77 215 : while (*n) {
78 160 : parent = *n;
79 160 : entry = rb_entry(parent, struct ext4_system_zone, node);
80 160 : if (start_blk < entry->start_blk)
81 71 : n = &(*n)->rb_left;
82 89 : else if (start_blk >= (entry->start_blk + entry->count))
83 89 : n = &(*n)->rb_right;
84 : else /* Unexpected overlap of system zones. */
85 : return -EFSCORRUPTED;
86 : }
87 :
88 55 : new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
89 : GFP_KERNEL);
90 55 : if (!new_entry)
91 : return -ENOMEM;
92 55 : new_entry->start_blk = start_blk;
93 55 : new_entry->count = count;
94 55 : new_entry->ino = ino;
95 55 : new_node = &new_entry->node;
96 :
97 55 : rb_link_node(new_node, parent, n);
98 55 : rb_insert_color(new_node, &system_blks->root);
99 :
100 : /* Can we merge to the left? */
101 55 : node = rb_prev(new_node);
102 55 : if (node) {
103 54 : entry = rb_entry(node, struct ext4_system_zone, node);
104 54 : if (can_merge(entry, new_entry)) {
105 45 : new_entry->start_blk = entry->start_blk;
106 45 : new_entry->count += entry->count;
107 45 : rb_erase(node, &system_blks->root);
108 45 : kmem_cache_free(ext4_system_zone_cachep, entry);
109 : }
110 : }
111 :
112 : /* Can we merge to the right? */
113 55 : node = rb_next(new_node);
114 55 : if (node) {
115 46 : entry = rb_entry(node, struct ext4_system_zone, node);
116 46 : if (can_merge(new_entry, entry)) {
117 2 : new_entry->count += entry->count;
118 2 : rb_erase(node, &system_blks->root);
119 2 : kmem_cache_free(ext4_system_zone_cachep, entry);
120 : }
121 : }
122 : return 0;
123 : }
124 :
125 0 : static void debug_print_tree(struct ext4_sb_info *sbi)
126 : {
127 0 : struct rb_node *node;
128 0 : struct ext4_system_zone *entry;
129 0 : struct ext4_system_blocks *system_blks;
130 0 : int first = 1;
131 :
132 0 : printk(KERN_INFO "System zones: ");
133 0 : rcu_read_lock();
134 0 : system_blks = rcu_dereference(sbi->s_system_blks);
135 0 : node = rb_first(&system_blks->root);
136 0 : while (node) {
137 0 : entry = rb_entry(node, struct ext4_system_zone, node);
138 0 : printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
139 0 : entry->start_blk, entry->start_blk + entry->count - 1);
140 0 : first = 0;
141 0 : node = rb_next(node);
142 : }
143 0 : rcu_read_unlock();
144 0 : printk(KERN_CONT "\n");
145 0 : }
146 :
147 1 : static int ext4_protect_reserved_inode(struct super_block *sb,
148 : struct ext4_system_blocks *system_blks,
149 : u32 ino)
150 : {
151 1 : struct inode *inode;
152 1 : struct ext4_sb_info *sbi = EXT4_SB(sb);
153 1 : struct ext4_map_blocks map;
154 1 : u32 i = 0, num;
155 1 : int err = 0, n;
156 :
157 1 : if ((ino < EXT4_ROOT_INO) ||
158 1 : (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
159 : return -EINVAL;
160 1 : inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
161 1 : if (IS_ERR(inode))
162 0 : return PTR_ERR(inode);
163 1 : num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
164 2 : while (i < num) {
165 1 : cond_resched();
166 1 : map.m_lblk = i;
167 1 : map.m_len = num - i;
168 1 : n = ext4_map_blocks(NULL, inode, &map, 0);
169 1 : if (n < 0) {
170 : err = n;
171 : break;
172 : }
173 1 : if (n == 0) {
174 0 : i++;
175 : } else {
176 1 : err = add_system_zone(system_blks, map.m_pblk, n, ino);
177 1 : if (err < 0) {
178 0 : if (err == -EFSCORRUPTED) {
179 0 : EXT4_ERROR_INODE_ERR(inode, -err,
180 : "blocks %llu-%llu from inode overlap system zone",
181 : map.m_pblk,
182 : map.m_pblk + map.m_len - 1);
183 : }
184 : break;
185 : }
186 1 : i += n;
187 : }
188 : }
189 1 : iput(inode);
190 1 : return err;
191 : }
192 :
193 0 : static void ext4_destroy_system_zone(struct rcu_head *rcu)
194 : {
195 0 : struct ext4_system_blocks *system_blks;
196 :
197 0 : system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
198 0 : release_system_zone(system_blks);
199 0 : kfree(system_blks);
200 0 : }
201 :
202 : /*
203 : * Build system zone rbtree which is used for block validity checking.
204 : *
205 : * The update of system_blks pointer in this function is protected by
206 : * sb->s_umount semaphore. However we have to be careful as we can be
207 : * racing with ext4_inode_block_valid() calls reading system_blks rbtree
208 : * protected only by RCU. That's why we first build the rbtree and then
209 : * swap it in place.
210 : */
211 1 : int ext4_setup_system_zone(struct super_block *sb)
212 : {
213 1 : ext4_group_t ngroups = ext4_get_groups_count(sb);
214 1 : struct ext4_sb_info *sbi = EXT4_SB(sb);
215 1 : struct ext4_system_blocks *system_blks;
216 1 : struct ext4_group_desc *gdp;
217 1 : ext4_group_t i;
218 1 : int flex_size = ext4_flex_bg_size(sbi);
219 1 : int ret;
220 :
221 1 : system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
222 1 : if (!system_blks)
223 : return -ENOMEM;
224 :
225 17 : for (i=0; i < ngroups; i++) {
226 16 : cond_resched();
227 16 : if (ext4_bg_has_super(sb, i) &&
228 3 : ((i < 5) || ((i % flex_size) == 0))) {
229 6 : ret = add_system_zone(system_blks,
230 : ext4_group_first_block_no(sb, i),
231 6 : ext4_bg_num_gdb(sb, i) + 1, 0);
232 6 : if (ret)
233 0 : goto err;
234 : }
235 16 : gdp = ext4_get_group_desc(sb, i, NULL);
236 16 : ret = add_system_zone(system_blks,
237 : ext4_block_bitmap(sb, gdp), 1, 0);
238 16 : if (ret)
239 0 : goto err;
240 16 : ret = add_system_zone(system_blks,
241 : ext4_inode_bitmap(sb, gdp), 1, 0);
242 16 : if (ret)
243 0 : goto err;
244 16 : ret = add_system_zone(system_blks,
245 : ext4_inode_table(sb, gdp),
246 16 : sbi->s_itb_per_group, 0);
247 16 : if (ret)
248 0 : goto err;
249 : }
250 1 : if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
251 1 : ret = ext4_protect_reserved_inode(sb, system_blks,
252 : le32_to_cpu(sbi->s_es->s_journal_inum));
253 1 : if (ret)
254 0 : goto err;
255 : }
256 :
257 : /*
258 : * System blks rbtree complete, announce it once to prevent racing
259 : * with ext4_inode_block_valid() accessing the rbtree at the same
260 : * time.
261 : */
262 1 : rcu_assign_pointer(sbi->s_system_blks, system_blks);
263 :
264 1 : if (test_opt(sb, DEBUG))
265 0 : debug_print_tree(sbi);
266 : return 0;
267 0 : err:
268 0 : release_system_zone(system_blks);
269 0 : kfree(system_blks);
270 0 : return ret;
271 : }
272 :
273 : /*
274 : * Called when the filesystem is unmounted or when remounting it with
275 : * noblock_validity specified.
276 : *
277 : * The update of system_blks pointer in this function is protected by
278 : * sb->s_umount semaphore. However we have to be careful as we can be
279 : * racing with ext4_inode_block_valid() calls reading system_blks rbtree
280 : * protected only by RCU. So we first clear the system_blks pointer and
281 : * then free the rbtree only after RCU grace period expires.
282 : */
283 0 : void ext4_release_system_zone(struct super_block *sb)
284 : {
285 0 : struct ext4_system_blocks *system_blks;
286 :
287 0 : system_blks = rcu_dereference_protected(EXT4_SB(sb)->s_system_blks,
288 : lockdep_is_held(&sb->s_umount));
289 0 : rcu_assign_pointer(EXT4_SB(sb)->s_system_blks, NULL);
290 :
291 0 : if (system_blks)
292 0 : call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
293 0 : }
294 :
295 : /*
296 : * Returns 1 if the passed-in block region (start_blk,
297 : * start_blk+count) is valid; 0 if some part of the block region
298 : * overlaps with some other filesystem metadata blocks.
299 : */
300 19305 : int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
301 : unsigned int count)
302 : {
303 19305 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
304 19305 : struct ext4_system_blocks *system_blks;
305 19305 : struct ext4_system_zone *entry;
306 19305 : struct rb_node *n;
307 19305 : int ret = 1;
308 :
309 19305 : if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
310 19305 : (start_blk + count < start_blk) ||
311 19305 : (start_blk + count > ext4_blocks_count(sbi->s_es)))
312 : return 0;
313 :
314 : /*
315 : * Lock the system zone to prevent it being released concurrently
316 : * when doing a remount which inverse current "[no]block_validity"
317 : * mount option.
318 : */
319 19306 : rcu_read_lock();
320 19305 : system_blks = rcu_dereference(sbi->s_system_blks);
321 19306 : if (system_blks == NULL)
322 2 : goto out_rcu;
323 :
324 19304 : n = system_blks->root.rb_node;
325 68513 : while (n) {
326 49209 : entry = rb_entry(n, struct ext4_system_zone, node);
327 49209 : if (start_blk + count - 1 < entry->start_blk)
328 23779 : n = n->rb_left;
329 25430 : else if (start_blk >= (entry->start_blk + entry->count))
330 25430 : n = n->rb_right;
331 : else {
332 0 : ret = (entry->ino == inode->i_ino);
333 0 : break;
334 : }
335 : }
336 19304 : out_rcu:
337 19306 : rcu_read_unlock();
338 19306 : return ret;
339 : }
340 :
341 0 : int ext4_check_blockref(const char *function, unsigned int line,
342 : struct inode *inode, __le32 *p, unsigned int max)
343 : {
344 0 : __le32 *bref = p;
345 0 : unsigned int blk;
346 :
347 0 : if (ext4_has_feature_journal(inode->i_sb) &&
348 0 : (inode->i_ino ==
349 0 : le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
350 : return 0;
351 :
352 0 : while (bref < p+max) {
353 0 : blk = le32_to_cpu(*bref++);
354 0 : if (blk &&
355 0 : unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
356 0 : ext4_error_inode(inode, function, line, blk,
357 : "invalid block");
358 0 : return -EFSCORRUPTED;
359 : }
360 : }
361 : return 0;
362 : }
363 :
|
