1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * linux/fs/ext4/block_validity.c 3 * linux/fs/ext4/block_validity.c
4 * 4 *
5 * Copyright (C) 2009 5 * Copyright (C) 2009
6 * Theodore Ts'o (tytso@mit.edu) 6 * Theodore Ts'o (tytso@mit.edu)
7 * 7 *
8 * Track which blocks in the filesystem are me 8 * Track which blocks in the filesystem are metadata blocks that
9 * should never be used as data blocks by file 9 * should never be used as data blocks by files or directories.
10 */ 10 */
11 11
12 #include <linux/time.h> 12 #include <linux/time.h>
13 #include <linux/fs.h> 13 #include <linux/fs.h>
14 #include <linux/namei.h> 14 #include <linux/namei.h>
15 #include <linux/quotaops.h> 15 #include <linux/quotaops.h>
16 #include <linux/buffer_head.h> 16 #include <linux/buffer_head.h>
17 #include <linux/swap.h> 17 #include <linux/swap.h>
18 #include <linux/pagemap.h> 18 #include <linux/pagemap.h>
19 #include <linux/blkdev.h> 19 #include <linux/blkdev.h>
20 #include <linux/slab.h> 20 #include <linux/slab.h>
21 #include "ext4.h" 21 #include "ext4.h"
22 22
23 struct ext4_system_zone { 23 struct ext4_system_zone {
24 struct rb_node node; 24 struct rb_node node;
25 ext4_fsblk_t start_blk; 25 ext4_fsblk_t start_blk;
26 unsigned int count; 26 unsigned int count;
27 u32 ino; <<
28 }; 27 };
29 28
30 static struct kmem_cache *ext4_system_zone_cac 29 static struct kmem_cache *ext4_system_zone_cachep;
31 30
32 int __init ext4_init_system_zone(void) 31 int __init ext4_init_system_zone(void)
33 { 32 {
34 ext4_system_zone_cachep = KMEM_CACHE(e 33 ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
35 if (ext4_system_zone_cachep == NULL) 34 if (ext4_system_zone_cachep == NULL)
36 return -ENOMEM; 35 return -ENOMEM;
37 return 0; 36 return 0;
38 } 37 }
39 38
40 void ext4_exit_system_zone(void) 39 void ext4_exit_system_zone(void)
41 { 40 {
42 rcu_barrier(); <<
43 kmem_cache_destroy(ext4_system_zone_ca 41 kmem_cache_destroy(ext4_system_zone_cachep);
44 } 42 }
45 43
46 static inline int can_merge(struct ext4_system 44 static inline int can_merge(struct ext4_system_zone *entry1,
47 struct ext4_system_zone * 45 struct ext4_system_zone *entry2)
48 { 46 {
49 if ((entry1->start_blk + entry1->count !! 47 if ((entry1->start_blk + entry1->count) == entry2->start_blk)
50 entry1->ino == entry2->ino) <<
51 return 1; 48 return 1;
52 return 0; 49 return 0;
53 } 50 }
54 51
55 static void release_system_zone(struct ext4_sy <<
56 { <<
57 struct ext4_system_zone *entry, *n; <<
58 <<
59 rbtree_postorder_for_each_entry_safe(e <<
60 &system_blks-> <<
61 kmem_cache_free(ext4_system_zo <<
62 } <<
63 <<
64 /* 52 /*
65 * Mark a range of blocks as belonging to the 53 * Mark a range of blocks as belonging to the "system zone" --- that
66 * is, filesystem metadata blocks which should 54 * is, filesystem metadata blocks which should never be used by
67 * inodes. 55 * inodes.
68 */ 56 */
69 static int add_system_zone(struct ext4_system_ !! 57 static int add_system_zone(struct ext4_sb_info *sbi,
70 ext4_fsblk_t start_ 58 ext4_fsblk_t start_blk,
71 unsigned int count, !! 59 unsigned int count)
72 { 60 {
73 struct ext4_system_zone *new_entry, *e !! 61 struct ext4_system_zone *new_entry = NULL, *entry;
74 struct rb_node **n = &system_blks->roo !! 62 struct rb_node **n = &sbi->system_blks.rb_node, *node;
75 struct rb_node *parent = NULL, *new_no !! 63 struct rb_node *parent = NULL, *new_node = NULL;
76 64
77 while (*n) { 65 while (*n) {
78 parent = *n; 66 parent = *n;
79 entry = rb_entry(parent, struc 67 entry = rb_entry(parent, struct ext4_system_zone, node);
80 if (start_blk < entry->start_b 68 if (start_blk < entry->start_blk)
81 n = &(*n)->rb_left; 69 n = &(*n)->rb_left;
82 else if (start_blk >= (entry-> 70 else if (start_blk >= (entry->start_blk + entry->count))
83 n = &(*n)->rb_right; 71 n = &(*n)->rb_right;
84 else /* Unexpected overlap !! 72 else {
85 return -EFSCORRUPTED; !! 73 if (start_blk + count > (entry->start_blk +
>> 74 entry->count))
>> 75 entry->count = (start_blk + count -
>> 76 entry->start_blk);
>> 77 new_node = *n;
>> 78 new_entry = rb_entry(new_node, struct ext4_system_zone,
>> 79 node);
>> 80 break;
>> 81 }
86 } 82 }
87 83
88 new_entry = kmem_cache_alloc(ext4_syst !! 84 if (!new_entry) {
89 GFP_KERNE !! 85 new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
90 if (!new_entry) !! 86 GFP_KERNEL);
91 return -ENOMEM; !! 87 if (!new_entry)
92 new_entry->start_blk = start_blk; !! 88 return -ENOMEM;
93 new_entry->count = count; !! 89 new_entry->start_blk = start_blk;
94 new_entry->ino = ino; !! 90 new_entry->count = count;
95 new_node = &new_entry->node; !! 91 new_node = &new_entry->node;
96 92
97 rb_link_node(new_node, parent, n); !! 93 rb_link_node(new_node, parent, n);
98 rb_insert_color(new_node, &system_blks !! 94 rb_insert_color(new_node, &sbi->system_blks);
>> 95 }
99 96
100 /* Can we merge to the left? */ 97 /* Can we merge to the left? */
101 node = rb_prev(new_node); 98 node = rb_prev(new_node);
102 if (node) { 99 if (node) {
103 entry = rb_entry(node, struct 100 entry = rb_entry(node, struct ext4_system_zone, node);
104 if (can_merge(entry, new_entry 101 if (can_merge(entry, new_entry)) {
105 new_entry->start_blk = 102 new_entry->start_blk = entry->start_blk;
106 new_entry->count += en 103 new_entry->count += entry->count;
107 rb_erase(node, &system !! 104 rb_erase(node, &sbi->system_blks);
108 kmem_cache_free(ext4_s 105 kmem_cache_free(ext4_system_zone_cachep, entry);
109 } 106 }
110 } 107 }
111 108
112 /* Can we merge to the right? */ 109 /* Can we merge to the right? */
113 node = rb_next(new_node); 110 node = rb_next(new_node);
114 if (node) { 111 if (node) {
115 entry = rb_entry(node, struct 112 entry = rb_entry(node, struct ext4_system_zone, node);
116 if (can_merge(new_entry, entry 113 if (can_merge(new_entry, entry)) {
117 new_entry->count += en 114 new_entry->count += entry->count;
118 rb_erase(node, &system !! 115 rb_erase(node, &sbi->system_blks);
119 kmem_cache_free(ext4_s 116 kmem_cache_free(ext4_system_zone_cachep, entry);
120 } 117 }
121 } 118 }
122 return 0; 119 return 0;
123 } 120 }
124 121
125 static void debug_print_tree(struct ext4_sb_in 122 static void debug_print_tree(struct ext4_sb_info *sbi)
126 { 123 {
127 struct rb_node *node; 124 struct rb_node *node;
128 struct ext4_system_zone *entry; 125 struct ext4_system_zone *entry;
129 struct ext4_system_blocks *system_blks <<
130 int first = 1; 126 int first = 1;
131 127
132 printk(KERN_INFO "System zones: "); 128 printk(KERN_INFO "System zones: ");
133 rcu_read_lock(); !! 129 node = rb_first(&sbi->system_blks);
134 system_blks = rcu_dereference(sbi->s_s <<
135 node = rb_first(&system_blks->root); <<
136 while (node) { 130 while (node) {
137 entry = rb_entry(node, struct 131 entry = rb_entry(node, struct ext4_system_zone, node);
138 printk(KERN_CONT "%s%llu-%llu" 132 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
139 entry->start_blk, entry 133 entry->start_blk, entry->start_blk + entry->count - 1);
140 first = 0; 134 first = 0;
141 node = rb_next(node); 135 node = rb_next(node);
142 } 136 }
143 rcu_read_unlock(); <<
144 printk(KERN_CONT "\n"); 137 printk(KERN_CONT "\n");
145 } 138 }
146 139
147 static int ext4_protect_reserved_inode(struct <<
148 struct <<
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_in <<
159 return -EINVAL; <<
160 inode = ext4_iget(sb, ino, EXT4_IGET_S <<
161 if (IS_ERR(inode)) <<
162 return PTR_ERR(inode); <<
163 num = (inode->i_size + sb->s_blocksize <<
164 while (i < num) { <<
165 cond_resched(); <<
166 map.m_lblk = i; <<
167 map.m_len = num - i; <<
168 n = ext4_map_blocks(NULL, inod <<
169 if (n < 0) { <<
170 err = n; <<
171 break; <<
172 } <<
173 if (n == 0) { <<
174 i++; <<
175 } else { <<
176 err = add_system_zone( <<
177 if (err < 0) { <<
178 if (err == -EF <<
179 EXT4_E <<
180 <<
181 <<
182 <<
183 } <<
184 break; <<
185 } <<
186 i += n; <<
187 } <<
188 } <<
189 iput(inode); <<
190 return err; <<
191 } <<
192 <<
193 static void ext4_destroy_system_zone(struct rc <<
194 { <<
195 struct ext4_system_blocks *system_blks <<
196 <<
197 system_blks = container_of(rcu, struct <<
198 release_system_zone(system_blks); <<
199 kfree(system_blks); <<
200 } <<
201 <<
202 /* <<
203 * Build system zone rbtree which is used for <<
204 * <<
205 * The update of system_blks pointer in this f <<
206 * sb->s_umount semaphore. However we have to <<
207 * racing with ext4_inode_block_valid() calls <<
208 * protected only by RCU. That's why we first <<
209 * swap it in place. <<
210 */ <<
211 int ext4_setup_system_zone(struct super_block 140 int ext4_setup_system_zone(struct super_block *sb)
212 { 141 {
213 ext4_group_t ngroups = ext4_get_groups 142 ext4_group_t ngroups = ext4_get_groups_count(sb);
214 struct ext4_sb_info *sbi = EXT4_SB(sb) 143 struct ext4_sb_info *sbi = EXT4_SB(sb);
215 struct ext4_system_blocks *system_blks <<
216 struct ext4_group_desc *gdp; 144 struct ext4_group_desc *gdp;
217 ext4_group_t i; 145 ext4_group_t i;
>> 146 int flex_size = ext4_flex_bg_size(sbi);
218 int ret; 147 int ret;
219 148
220 system_blks = kzalloc(sizeof(*system_b !! 149 if (!test_opt(sb, BLOCK_VALIDITY)) {
221 if (!system_blks) !! 150 if (sbi->system_blks.rb_node)
222 return -ENOMEM; !! 151 ext4_release_system_zone(sb);
>> 152 return 0;
>> 153 }
>> 154 if (sbi->system_blks.rb_node)
>> 155 return 0;
223 156
224 for (i=0; i < ngroups; i++) { 157 for (i=0; i < ngroups; i++) {
225 unsigned int meta_blks = ext4_ !! 158 if (ext4_bg_has_super(sb, i) &&
226 !! 159 ((i < 5) || ((i % flex_size) == 0)))
227 cond_resched(); !! 160 add_system_zone(sbi, ext4_group_first_block_no(sb, i),
228 if (meta_blks != 0) { !! 161 ext4_bg_num_gdb(sb, i) + 1);
229 ret = add_system_zone( <<
230 ext4_g <<
231 meta_b <<
232 if (ret) <<
233 goto err; <<
234 } <<
235 gdp = ext4_get_group_desc(sb, 162 gdp = ext4_get_group_desc(sb, i, NULL);
236 ret = add_system_zone(system_b !! 163 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1);
237 ext4_block_bit <<
238 if (ret) 164 if (ret)
239 goto err; !! 165 return ret;
240 ret = add_system_zone(system_b !! 166 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1);
241 ext4_inode_bit <<
242 if (ret) 167 if (ret)
243 goto err; !! 168 return ret;
244 ret = add_system_zone(system_b !! 169 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
245 ext4_inode_tab !! 170 sbi->s_itb_per_group);
246 sbi->s_itb_per <<
247 if (ret) <<
248 goto err; <<
249 } <<
250 if (ext4_has_feature_journal(sb) && sb <<
251 ret = ext4_protect_reserved_in <<
252 le32_to_cpu(sb <<
253 if (ret) 171 if (ret)
254 goto err; !! 172 return ret;
255 } 173 }
256 174
257 /* <<
258 * System blks rbtree complete, announ <<
259 * with ext4_inode_block_valid() acces <<
260 * time. <<
261 */ <<
262 rcu_assign_pointer(sbi->s_system_blks, <<
263 <<
264 if (test_opt(sb, DEBUG)) 175 if (test_opt(sb, DEBUG))
265 debug_print_tree(sbi); 176 debug_print_tree(sbi);
266 return 0; 177 return 0;
267 err: <<
268 release_system_zone(system_blks); <<
269 kfree(system_blks); <<
270 return ret; <<
271 } 178 }
272 179
273 /* !! 180 /* Called when the filesystem is unmounted */
274 * Called when the filesystem is unmounted or <<
275 * noblock_validity specified. <<
276 * <<
277 * The update of system_blks pointer in this f <<
278 * sb->s_umount semaphore. However we have to <<
279 * racing with ext4_inode_block_valid() calls <<
280 * protected only by RCU. So we first clear th <<
281 * then free the rbtree only after RCU grace p <<
282 */ <<
283 void ext4_release_system_zone(struct super_blo 181 void ext4_release_system_zone(struct super_block *sb)
284 { 182 {
285 struct ext4_system_blocks *system_blks !! 183 struct ext4_system_zone *entry, *n;
286 184
287 system_blks = rcu_dereference_protecte !! 185 rbtree_postorder_for_each_entry_safe(entry, n,
288 lockde !! 186 &EXT4_SB(sb)->system_blks, node)
289 rcu_assign_pointer(EXT4_SB(sb)->s_syst !! 187 kmem_cache_free(ext4_system_zone_cachep, entry);
290 188
291 if (system_blks) !! 189 EXT4_SB(sb)->system_blks = RB_ROOT;
292 call_rcu(&system_blks->rcu, ex <<
293 } 190 }
294 191
295 int ext4_sb_block_valid(struct super_block *sb !! 192 /*
296 ext4_fsblk_t s !! 193 * Returns 1 if the passed-in block region (start_blk,
>> 194 * start_blk+count) is valid; 0 if some part of the block region
>> 195 * overlaps with filesystem metadata blocks.
>> 196 */
>> 197 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
>> 198 unsigned int count)
297 { 199 {
298 struct ext4_sb_info *sbi = EXT4_SB(sb) <<
299 struct ext4_system_blocks *system_blks <<
300 struct ext4_system_zone *entry; 200 struct ext4_system_zone *entry;
301 struct rb_node *n; !! 201 struct rb_node *n = sbi->system_blks.rb_node;
302 int ret = 1; <<
303 202
304 if ((start_blk <= le32_to_cpu(sbi->s_e 203 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
305 (start_blk + count < start_blk) || 204 (start_blk + count < start_blk) ||
306 (start_blk + count > ext4_blocks_c !! 205 (start_blk + count > ext4_blocks_count(sbi->s_es))) {
>> 206 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
307 return 0; 207 return 0;
308 !! 208 }
309 /* <<
310 * Lock the system zone to prevent it <<
311 * when doing a remount which inverse <<
312 * mount option. <<
313 */ <<
314 rcu_read_lock(); <<
315 system_blks = rcu_dereference(sbi->s_s <<
316 if (system_blks == NULL) <<
317 goto out_rcu; <<
318 <<
319 n = system_blks->root.rb_node; <<
320 while (n) { 209 while (n) {
321 entry = rb_entry(n, struct ext 210 entry = rb_entry(n, struct ext4_system_zone, node);
322 if (start_blk + count - 1 < en 211 if (start_blk + count - 1 < entry->start_blk)
323 n = n->rb_left; 212 n = n->rb_left;
324 else if (start_blk >= (entry-> 213 else if (start_blk >= (entry->start_blk + entry->count))
325 n = n->rb_right; 214 n = n->rb_right;
326 else { 215 else {
327 ret = 0; !! 216 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
328 if (inode) !! 217 return 0;
329 ret = (entry-> <<
330 break; <<
331 } 218 }
332 } 219 }
333 out_rcu: !! 220 return 1;
334 rcu_read_unlock(); <<
335 return ret; <<
336 } <<
337 <<
338 /* <<
339 * Returns 1 if the passed-in block region (st <<
340 * start_blk+count) is valid; 0 if some part o <<
341 * overlaps with some other filesystem metadat <<
342 */ <<
343 int ext4_inode_block_valid(struct inode *inode <<
344 unsigned int count) <<
345 { <<
346 return ext4_sb_block_valid(inode->i_sb <<
347 } 221 }
348 222
349 int ext4_check_blockref(const char *function, 223 int ext4_check_blockref(const char *function, unsigned int line,
350 struct inode *inode, _ 224 struct inode *inode, __le32 *p, unsigned int max)
351 { 225 {
>> 226 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
352 __le32 *bref = p; 227 __le32 *bref = p;
353 unsigned int blk; 228 unsigned int blk;
354 229
355 if (ext4_has_feature_journal(inode->i_ <<
356 (inode->i_ino == <<
357 le32_to_cpu(EXT4_SB(inode->i_sb)- <<
358 return 0; <<
359 <<
360 while (bref < p+max) { 230 while (bref < p+max) {
361 blk = le32_to_cpu(*bref++); 231 blk = le32_to_cpu(*bref++);
362 if (blk && 232 if (blk &&
363 unlikely(!ext4_inode_block !! 233 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
>> 234 blk, 1))) {
>> 235 es->s_last_error_block = cpu_to_le64(blk);
364 ext4_error_inode(inode 236 ext4_error_inode(inode, function, line, blk,
365 "inva 237 "invalid block");
366 return -EFSCORRUPTED; 238 return -EFSCORRUPTED;
367 } 239 }
368 } 240 }
369 return 0; 241 return 0;
370 } 242 }
371 243
372 244
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.