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 12 : int __init ext4_init_system_zone(void)
33 : {
34 12 : ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 12 : 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 : static inline int can_merge(struct ext4_system_zone *entry1,
47 : struct ext4_system_zone *entry2)
48 : {
49 6803278 : if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
50 2499200 : entry1->ino == entry2->ino)
51 2498959 : return 1;
52 : return 0;
53 : }
54 :
55 2511 : static void release_system_zone(struct ext4_system_blocks *system_blks)
56 : {
57 2511 : struct ext4_system_zone *entry, *n;
58 :
59 91311 : rbtree_postorder_for_each_entry_safe(entry, n,
60 : &system_blks->root, node)
61 86289 : kmem_cache_free(ext4_system_zone_cachep, entry);
62 2511 : }
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 2585248 : 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 2585248 : struct ext4_system_zone *new_entry, *entry;
74 2585248 : struct rb_node **n = &system_blks->root.rb_node, *node;
75 2585248 : struct rb_node *parent = NULL, *new_node = NULL;
76 :
77 31185201 : while (*n) {
78 28599953 : parent = *n;
79 28599953 : entry = rb_entry(parent, struct ext4_system_zone, node);
80 28599953 : if (start_blk < entry->start_blk)
81 2140562 : n = &(*n)->rb_left;
82 26459391 : else if (start_blk >= (entry->start_blk + entry->count))
83 26459391 : n = &(*n)->rb_right;
84 : else /* Unexpected overlap of system zones. */
85 : return -EFSCORRUPTED;
86 : }
87 :
88 2585248 : new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
89 : GFP_KERNEL);
90 2585248 : if (!new_entry)
91 : return -ENOMEM;
92 2585248 : new_entry->start_blk = start_blk;
93 2585248 : new_entry->count = count;
94 2585248 : new_entry->ino = ino;
95 2585248 : new_node = &new_entry->node;
96 :
97 2585248 : rb_link_node(new_node, parent, n);
98 2585248 : rb_insert_color(new_node, &system_blks->root);
99 :
100 : /* Can we merge to the left? */
101 2585248 : node = rb_prev(new_node);
102 2585248 : if (node) {
103 2582737 : entry = rb_entry(node, struct ext4_system_zone, node);
104 2582737 : if (can_merge(entry, new_entry)) {
105 2391151 : new_entry->start_blk = entry->start_blk;
106 2391151 : new_entry->count += entry->count;
107 2391151 : rb_erase(node, &system_blks->root);
108 2391151 : kmem_cache_free(ext4_system_zone_cachep, entry);
109 : }
110 : }
111 :
112 : /* Can we merge to the right? */
113 2585248 : node = rb_next(new_node);
114 2585248 : if (node) {
115 1721341 : entry = rb_entry(node, struct ext4_system_zone, node);
116 1721341 : if (can_merge(new_entry, entry)) {
117 107808 : new_entry->count += entry->count;
118 107808 : rb_erase(node, &system_blks->root);
119 107808 : kmem_cache_free(ext4_system_zone_cachep, entry);
120 : }
121 : }
122 : return 0;
123 : }
124 :
125 2 : static void debug_print_tree(struct ext4_sb_info *sbi)
126 : {
127 2 : struct rb_node *node;
128 2 : struct ext4_system_zone *entry;
129 2 : struct ext4_system_blocks *system_blks;
130 2 : int first = 1;
131 :
132 2 : printk(KERN_INFO "System zones: ");
133 2 : rcu_read_lock();
134 2 : system_blks = rcu_dereference(sbi->s_system_blks);
135 2 : node = rb_first(&system_blks->root);
136 31 : while (node) {
137 29 : entry = rb_entry(node, struct ext4_system_zone, node);
138 56 : printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
139 : entry->start_blk, entry->start_blk + entry->count - 1);
140 29 : first = 0;
141 29 : node = rb_next(node);
142 : }
143 2 : rcu_read_unlock();
144 2 : printk(KERN_CONT "\n");
145 2 : }
146 :
147 2497 : static int ext4_protect_reserved_inode(struct super_block *sb,
148 : struct ext4_system_blocks *system_blks,
149 : u32 ino)
150 : {
151 2497 : struct inode *inode;
152 2497 : struct ext4_sb_info *sbi = EXT4_SB(sb);
153 2497 : struct ext4_map_blocks map;
154 2497 : u32 i = 0, num;
155 2497 : int err = 0, n;
156 :
157 2497 : if ((ino < EXT4_ROOT_INO) ||
158 2497 : (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
159 : return -EINVAL;
160 2497 : inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
161 2497 : if (IS_ERR(inode))
162 0 : return PTR_ERR(inode);
163 2497 : num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
164 7174 : while (i < num) {
165 4677 : cond_resched();
166 4677 : map.m_lblk = i;
167 4677 : map.m_len = num - i;
168 4677 : n = ext4_map_blocks(NULL, inode, &map, 0);
169 4677 : if (n < 0) {
170 : err = n;
171 : break;
172 : }
173 4677 : if (n == 0) {
174 0 : i++;
175 : } else {
176 4677 : err = add_system_zone(system_blks, map.m_pblk, n, ino);
177 4677 : 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 4677 : i += n;
187 : }
188 : }
189 2497 : iput(inode);
190 2497 : return err;
191 : }
192 :
193 2511 : static void ext4_destroy_system_zone(struct rcu_head *rcu)
194 : {
195 2511 : struct ext4_system_blocks *system_blks;
196 :
197 2511 : system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
198 2511 : release_system_zone(system_blks);
199 2511 : kfree(system_blks);
200 2511 : }
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 2511 : int ext4_setup_system_zone(struct super_block *sb)
212 : {
213 2511 : ext4_group_t ngroups = ext4_get_groups_count(sb);
214 2511 : struct ext4_sb_info *sbi = EXT4_SB(sb);
215 2511 : struct ext4_system_blocks *system_blks;
216 2511 : struct ext4_group_desc *gdp;
217 2511 : ext4_group_t i;
218 2511 : int flex_size = ext4_flex_bg_size(sbi);
219 2511 : int ret;
220 :
221 2511 : system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
222 2511 : if (!system_blks)
223 : return -ENOMEM;
224 :
225 854499 : for (i=0; i < ngroups; i++) {
226 851988 : cond_resched();
227 851988 : if (ext4_bg_has_super(sb, i) &&
228 17407 : ((i < 5) || ((i % flex_size) == 0))) {
229 24607 : ret = add_system_zone(system_blks,
230 : ext4_group_first_block_no(sb, i),
231 24607 : ext4_bg_num_gdb(sb, i) + 1, 0);
232 24607 : if (ret)
233 0 : goto err;
234 : }
235 851988 : gdp = ext4_get_group_desc(sb, i, NULL);
236 851988 : ret = add_system_zone(system_blks,
237 : ext4_block_bitmap(sb, gdp), 1, 0);
238 851988 : if (ret)
239 0 : goto err;
240 851988 : ret = add_system_zone(system_blks,
241 : ext4_inode_bitmap(sb, gdp), 1, 0);
242 851988 : if (ret)
243 0 : goto err;
244 851988 : ret = add_system_zone(system_blks,
245 : ext4_inode_table(sb, gdp),
246 851988 : sbi->s_itb_per_group, 0);
247 851988 : if (ret)
248 0 : goto err;
249 : }
250 2511 : if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
251 2497 : ret = ext4_protect_reserved_inode(sb, system_blks,
252 : le32_to_cpu(sbi->s_es->s_journal_inum));
253 2497 : 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 2511 : rcu_assign_pointer(sbi->s_system_blks, system_blks);
263 :
264 2511 : if (test_opt(sb, DEBUG))
265 2 : 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 2513 : void ext4_release_system_zone(struct super_block *sb)
284 : {
285 2513 : struct ext4_system_blocks *system_blks;
286 :
287 2513 : system_blks = rcu_dereference_protected(EXT4_SB(sb)->s_system_blks,
288 : lockdep_is_held(&sb->s_umount));
289 2513 : rcu_assign_pointer(EXT4_SB(sb)->s_system_blks, NULL);
290 :
291 2513 : if (system_blks)
292 2511 : call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
293 2513 : }
294 :
295 34094682 : int ext4_sb_block_valid(struct super_block *sb, struct inode *inode,
296 : ext4_fsblk_t start_blk, unsigned int count)
297 : {
298 34094682 : struct ext4_sb_info *sbi = EXT4_SB(sb);
299 34094682 : struct ext4_system_blocks *system_blks;
300 34094682 : struct ext4_system_zone *entry;
301 34094682 : struct rb_node *n;
302 34094682 : int ret = 1;
303 :
304 34094682 : if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
305 68189364 : (start_blk + count < start_blk) ||
306 : (start_blk + count > ext4_blocks_count(sbi->s_es)))
307 : return 0;
308 :
309 : /*
310 : * Lock the system zone to prevent it being released concurrently
311 : * when doing a remount which inverse current "[no]block_validity"
312 : * mount option.
313 : */
314 34091692 : rcu_read_lock();
315 34097104 : system_blks = rcu_dereference(sbi->s_system_blks);
316 34097104 : if (system_blks == NULL)
317 8523 : goto out_rcu;
318 :
319 34088581 : n = system_blks->root.rb_node;
320 200606314 : while (n) {
321 166517733 : entry = rb_entry(n, struct ext4_system_zone, node);
322 166517733 : if (start_blk + count - 1 < entry->start_blk)
323 82256658 : n = n->rb_left;
324 84261075 : else if (start_blk >= (entry->start_blk + entry->count))
325 84261075 : n = n->rb_right;
326 : else {
327 0 : ret = 0;
328 0 : if (inode)
329 0 : ret = (entry->ino == inode->i_ino);
330 : break;
331 : }
332 : }
333 34088581 : out_rcu:
334 34097104 : rcu_read_unlock();
335 34097104 : return ret;
336 : }
337 :
338 : /*
339 : * Returns 1 if the passed-in block region (start_blk,
340 : * start_blk+count) is valid; 0 if some part of the block region
341 : * overlaps with some other filesystem metadata blocks.
342 : */
343 34102077 : int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
344 : unsigned int count)
345 : {
346 34102077 : return ext4_sb_block_valid(inode->i_sb, inode, start_blk, count);
347 : }
348 :
349 2503 : int ext4_check_blockref(const char *function, unsigned int line,
350 : struct inode *inode, __le32 *p, unsigned int max)
351 : {
352 2503 : __le32 *bref = p;
353 2503 : unsigned int blk;
354 :
355 2503 : if (ext4_has_feature_journal(inode->i_sb) &&
356 2502 : (inode->i_ino ==
357 2502 : le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
358 : return 0;
359 :
360 4121 : while (bref < p+max) {
361 3804 : blk = le32_to_cpu(*bref++);
362 3804 : if (blk &&
363 306 : unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
364 0 : ext4_error_inode(inode, function, line, blk,
365 : "invalid block");
366 0 : return -EFSCORRUPTED;
367 : }
368 : }
369 : return 0;
370 : }
371 :
|
