| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
/* |
| 3 |
* linux/fs/ext4/block_validity.c |
| 4 |
* |
| 5 |
* Copyright (C) 2009 |
| 6 |
* Theodore Ts'o ([email protected]) |
| 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 |
int __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 |
|
| 40 |
void ext4_exit_system_zone(void) |
| 41 |
{ |
| 42 |
rcu_barrier(); |
| 43 |
kmem_cache_destroy(ext4_system_zone_cachep); |
| 44 |
} |
| 45 |
|
| 46 |
static 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 |
|
| 55 |
static 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 |
*/ |
| 69 |
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 |
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; |
| 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 |
|
| 125 |
static 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->s_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 |
|
| 147 |
static 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_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 |
i += n; |
| 187 |
} |
| 188 |
} |
| 189 |
iput(inode); |
| 190 |
return err; |
| 191 |
} |
| 192 |
|
| 193 |
static void ext4_destroy_system_zone(struct rcu_head *rcu) |
| 194 |
{ |
| 195 |
struct ext4_system_blocks *system_blks; |
| 196 |
|
| 197 |
system_blks = container_of(rcu, struct ext4_system_blocks, rcu); |
| 198 |
release_system_zone(system_blks); |
| 199 |
kfree(system_blks); |
| 200 |
} |
| 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 |
int ext4_setup_system_zone(struct super_block *sb) |
| 212 |
{ |
| 213 |
ext4_group_t ngroups = ext4_get_groups_count(sb); |
| 214 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 215 |
struct ext4_system_blocks *system_blks; |
| 216 |
struct ext4_group_desc *gdp; |
| 217 |
ext4_group_t i; |
| 218 |
int ret; |
| 219 |
|
| 220 |
system_blks = kzalloc_obj(*system_blks); |
| 221 |
if (!system_blks) |
| 222 |
return -ENOMEM; |
| 223 |
|
| 224 |
for (i=0; i < ngroups; i++) { |
| 225 |
unsigned int meta_blks = ext4_num_base_meta_blocks(sb, i); |
| 226 |
|
| 227 |
cond_resched(); |
| 228 |
if (meta_blks != 0) { |
| 229 |
ret = add_system_zone(system_blks, |
| 230 |
ext4_group_first_block_no(sb, i), |
| 231 |
meta_blks, 0); |
| 232 |
if (ret) |
| 233 |
goto err; |
| 234 |
} |
| 235 |
gdp = ext4_get_group_desc(sb, i, NULL); |
| 236 |
ret = add_system_zone(system_blks, |
| 237 |
ext4_block_bitmap(sb, gdp), 1, 0); |
| 238 |
if (ret) |
| 239 |
goto err; |
| 240 |
ret = add_system_zone(system_blks, |
| 241 |
ext4_inode_bitmap(sb, gdp), 1, 0); |
| 242 |
if (ret) |
| 243 |
goto err; |
| 244 |
ret = add_system_zone(system_blks, |
| 245 |
ext4_inode_table(sb, gdp), |
| 246 |
sbi->s_itb_per_group, 0); |
| 247 |
if (ret) |
| 248 |
goto err; |
| 249 |
} |
| 250 |
if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) { |
| 251 |
ret = ext4_protect_reserved_inode(sb, system_blks, |
| 252 |
le32_to_cpu(sbi->s_es->s_journal_inum)); |
| 253 |
if (ret) |
| 254 |
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 |
rcu_assign_pointer(sbi->s_system_blks, system_blks); |
| 263 |
|
| 264 |
if (test_opt(sb, DEBUG)) |
| 265 |
debug_print_tree(sbi); |
| 266 |
return 0; |
| 267 |
err: |
| 268 |
release_system_zone(system_blks); |
| 269 |
kfree(system_blks); |
| 270 |
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 |
void ext4_release_system_zone(struct super_block *sb) |
| 284 |
{ |
| 285 |
struct ext4_system_blocks *system_blks; |
| 286 |
|
| 287 |
system_blks = rcu_dereference_protected(EXT4_SB(sb)->s_system_blks, |
| 288 |
lockdep_is_held(&sb->s_umount)); |
| 289 |
rcu_assign_pointer(EXT4_SB(sb)->s_system_blks, NULL); |
| 290 |
|
| 291 |
if (system_blks) |
| 292 |
call_rcu(&system_blks->rcu, ext4_destroy_system_zone); |
| 293 |
} |
| 294 |
|
| 295 |
int ext4_sb_block_valid(struct super_block *sb, struct inode *inode, |
| 296 |
ext4_fsblk_t start_blk, unsigned int count) |
| 297 |
{ |
| 298 |
struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 299 |
struct ext4_system_blocks *system_blks; |
| 300 |
struct ext4_system_zone *entry; |
| 301 |
struct rb_node *n; |
| 302 |
int ret = 1; |
| 303 |
|
| 304 |
if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || |
| 305 |
(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 |
rcu_read_lock(); |
| 315 |
system_blks = rcu_dereference(sbi->s_system_blks); |
| 316 |
if (system_blks == NULL) |
| 317 |
goto out_rcu; |
| 318 |
|
| 319 |
n = system_blks->root.rb_node; |
| 320 |
while (n) { |
| 321 |
entry = rb_entry(n, struct ext4_system_zone, node); |
| 322 |
if (start_blk + count - 1 < entry->start_blk) |
| 323 |
n = n->rb_left; |
| 324 |
else if (start_blk >= (entry->start_blk + entry->count)) |
| 325 |
n = n->rb_right; |
| 326 |
else { |
| 327 |
ret = 0; |
| 328 |
if (inode) |
| 329 |
ret = (entry->ino == inode->i_ino); |
| 330 |
break; |
| 331 |
} |
| 332 |
} |
| 333 |
out_rcu: |
| 334 |
rcu_read_unlock(); |
| 335 |
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 |
int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk, |
| 344 |
unsigned int count) |
| 345 |
{ |
| 346 |
return ext4_sb_block_valid(inode->i_sb, inode, start_blk, count); |
| 347 |
} |
| 348 |
|
| 349 |
int ext4_check_blockref(const char *function, unsigned int line, |
| 350 |
struct inode *inode, __le32 *p, unsigned int max) |
| 351 |
{ |
| 352 |
__le32 *bref = p; |
| 353 |
unsigned int blk; |
| 354 |
journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 355 |
|
| 356 |
if (journal && inode == journal->j_inode) |
| 357 |
return 0; |
| 358 |
|
| 359 |
while (bref < p+max) { |
| 360 |
blk = le32_to_cpu(*bref++); |
| 361 |
if (blk && |
| 362 |
unlikely(!ext4_inode_block_valid(inode, blk, 1))) { |
| 363 |
ext4_error_inode(inode, function, line, blk, |
| 364 |
"invalid block"); |
| 365 |
return -EFSCORRUPTED; |
| 366 |
} |
| 367 |
} |
| 368 |
return 0; |
| 369 |
} |
| 370 |
|