fs/ext4/block_validity.c

   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 };
  28
  29 static struct kmem_cache *ext4_system_zone_cachep;
  30
  31 int __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
  39 void ext4_exit_system_zone(void)
  40 {
  41         rcu_barrier();
  42         kmem_cache_destroy(ext4_system_zone_cachep);
  43 }
  44
  45 static 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 static 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  */
  67 static 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, *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    /* Unexpected overlap of system zones. */
  83                         return -EFSCORRUPTED;
  84         }
  85
  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, &system_blks->root);
  96
  97         /* Can we merge to the left? */
  98         node = rb_prev(new_node);
  99         if (node) {
 100                 entry = rb_entry(node, struct ext4_system_zone, node);
 101                 if (can_merge(entry, new_entry)) {
 102                         new_entry->start_blk = entry->start_blk;
 103                         new_entry->count += entry->count;
 104                         rb_erase(node, &system_blks->root);
 105                         kmem_cache_free(ext4_system_zone_cachep, entry);
 106                 }
 107         }
 108
 109         /* Can we merge to the right? */
 110         node = rb_next(new_node);
 111         if (node) {
 112                 entry = rb_entry(node, struct ext4_system_zone, node);
 113                 if (can_merge(new_entry, entry)) {
 114                         new_entry->count += entry->count;
 115                         rb_erase(node, &system_blks->root);
 116                         kmem_cache_free(ext4_system_zone_cachep, entry);
 117                 }
 118         }
 119         return 0;
 120 }
 121
 122 static void debug_print_tree(struct ext4_sb_info *sbi)
 123 {
 124         struct rb_node *node;
 125         struct ext4_system_zone *entry;
 126         int first = 1;
 127
 128         printk(KERN_INFO "System zones: ");
 129         node = rb_first(&sbi->system_blks->root);
 130         while (node) {
 131                 entry = rb_entry(node, struct ext4_system_zone, node);
 132                 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
 133                        entry->start_blk, entry->start_blk + entry->count - 1);
 134                 first = 0;
 135                 node = rb_next(node);
 136         }
 137         printk(KERN_CONT "\n");
 138 }
 139
 140 /*
 141  * Returns 1 if the passed-in block region (start_blk,
 142  * start_blk+count) is valid; 0 if some part of the block region
 143  * overlaps with filesystem metadata blocks.
 144  */
 145 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
 146                                      struct ext4_system_blocks *system_blks,
 147                                      ext4_fsblk_t start_blk,
 148                                      unsigned int count)
 149 {
 150         struct ext4_system_zone *entry;
 151         struct rb_node *n;
 152
 153         if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 154             (start_blk + count < start_blk) ||
 155             (start_blk + count > ext4_blocks_count(sbi->s_es))) {
 156                 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 157                 return 0;
 158         }
 159
 160         if (system_blks == NULL)
 161                 return 1;
 162
 163         n = system_blks->root.rb_node;
 164         while (n) {
 165                 entry = rb_entry(n, struct ext4_system_zone, node);
 166                 if (start_blk + count - 1 < entry->start_blk)
 167                         n = n->rb_left;
 168                 else if (start_blk >= (entry->start_blk + entry->count))
 169                         n = n->rb_right;
 170                 else {
 171                         sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 172                         return 0;
 173                 }
 174         }
 175         return 1;
 176 }
 177
 178 static int ext4_protect_reserved_inode(struct super_block *sb,
 179                                        struct ext4_system_blocks *system_blks,
 180                                        u32 ino)
 181 {
 182         struct inode *inode;
 183         struct ext4_sb_info *sbi = EXT4_SB(sb);
 184         struct ext4_map_blocks map;
 185         u32 i = 0, num;
 186         int err = 0, n;
 187
 188         if ((ino < EXT4_ROOT_INO) ||
 189             (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
 190                 return -EINVAL;
 191         inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
 192         if (IS_ERR(inode))
 193                 return PTR_ERR(inode);
 194         num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 195         while (i < num) {
 196                 cond_resched();
 197                 map.m_lblk = i;
 198                 map.m_len = num - i;
 199                 n = ext4_map_blocks(NULL, inode, &map, 0);
 200                 if (n < 0) {
 201                         err = n;
 202                         break;
 203                 }
 204                 if (n == 0) {
 205                         i++;
 206                 } else {
 207                         if (!ext4_data_block_valid_rcu(sbi, system_blks,
 208                                                 map.m_pblk, n)) {
 209                                 ext4_error(sb, "blocks %llu-%llu from inode %u "
 210                                            "overlap system zone", map.m_pblk,
 211                                            map.m_pblk + map.m_len - 1, ino);
 212                                 err = -EFSCORRUPTED;
 213                                 break;
 214                         }
 215                         err = add_system_zone(system_blks, map.m_pblk, n);
 216                         if (err < 0)
 217                                 break;
 218                         i += n;
 219                 }
 220         }
 221         iput(inode);
 222         return err;
 223 }
 224
 225 static void ext4_destroy_system_zone(struct rcu_head *rcu)
 226 {
 227         struct ext4_system_blocks *system_blks;
 228
 229         system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
 230         release_system_zone(system_blks);
 231         kfree(system_blks);
 232 }
 233
 234 /*
 235  * Build system zone rbtree which is used for block validity checking.
 236  *
 237  * The update of system_blks pointer in this function is protected by
 238  * sb->s_umount semaphore. However we have to be careful as we can be
 239  * racing with ext4_data_block_valid() calls reading system_blks rbtree
 240  * protected only by RCU. That's why we first build the rbtree and then
 241  * swap it in place.
 242  */
 243 int ext4_setup_system_zone(struct super_block *sb)
 244 {
 245         ext4_group_t ngroups = ext4_get_groups_count(sb);
 246         struct ext4_sb_info *sbi = EXT4_SB(sb);
 247         struct ext4_system_blocks *system_blks;
 248         struct ext4_group_desc *gdp;
 249         ext4_group_t i;
 250         int flex_size = ext4_flex_bg_size(sbi);
 251         int ret;
 252
 253         system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
 254         if (!system_blks)
 255                 return -ENOMEM;
 256
 257         for (i=0; i < ngroups; i++) {
 258                 cond_resched();
 259                 if (ext4_bg_has_super(sb, i) &&
 260                     ((i < 5) || ((i % flex_size) == 0)))
 261                         add_system_zone(system_blks,
 262                                         ext4_group_first_block_no(sb, i),
 263                                         ext4_bg_num_gdb(sb, i) + 1);
 264                 gdp = ext4_get_group_desc(sb, i, NULL);
 265                 ret = add_system_zone(system_blks,
 266                                 ext4_block_bitmap(sb, gdp), 1);
 267                 if (ret)
 268                         goto err;
 269                 ret = add_system_zone(system_blks,
 270                                 ext4_inode_bitmap(sb, gdp), 1);
 271                 if (ret)
 272                         goto err;
 273                 ret = add_system_zone(system_blks,
 274                                 ext4_inode_table(sb, gdp),
 275                                 sbi->s_itb_per_group);
 276                 if (ret)
 277                         goto err;
 278         }
 279         if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
 280                 ret = ext4_protect_reserved_inode(sb, system_blks,
 281                                 le32_to_cpu(sbi->s_es->s_journal_inum));
 282                 if (ret)
 283                         goto err;
 284         }
 285
 286         /*
 287          * System blks rbtree complete, announce it once to prevent racing
 288          * with ext4_data_block_valid() accessing the rbtree at the same
 289          * time.
 290          */
 291         rcu_assign_pointer(sbi->system_blks, system_blks);
 292
 293         if (test_opt(sb, DEBUG))
 294                 debug_print_tree(sbi);
 295         return 0;
 296 err:
 297         release_system_zone(system_blks);
 298         kfree(system_blks);
 299         return ret;
 300 }
 301
 302 /*
 303  * Called when the filesystem is unmounted or when remounting it with
 304  * noblock_validity specified.
 305  *
 306  * The update of system_blks pointer in this function is protected by
 307  * sb->s_umount semaphore. However we have to be careful as we can be
 308  * racing with ext4_data_block_valid() calls reading system_blks rbtree
 309  * protected only by RCU. So we first clear the system_blks pointer and
 310  * then free the rbtree only after RCU grace period expires.
 311  */
 312 void ext4_release_system_zone(struct super_block *sb)
 313 {
 314         struct ext4_system_blocks *system_blks;
 315
 316         system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
 317                                         lockdep_is_held(&sb->s_umount));
 318         rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
 319
 320         if (system_blks)
 321                 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
 322 }
 323
 324 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk,
 325                           unsigned int count)
 326 {
 327         struct ext4_system_blocks *system_blks;
 328         int ret;
 329
 330         /*
 331          * Lock the system zone to prevent it being released concurrently
 332          * when doing a remount which inverse current "[no]block_validity"
 333          * mount option.
 334          */
 335         rcu_read_lock();
 336         system_blks = rcu_dereference(sbi->system_blks);
 337         ret = ext4_data_block_valid_rcu(sbi, system_blks, start_blk,
 338                                         count);
 339         rcu_read_unlock();
 340         return ret;
 341 }
 342
 343 int ext4_check_blockref(const char *function, unsigned int line,
 344                         struct inode *inode, __le32 *p, unsigned int max)
 345 {
 346         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 347         __le32 *bref = p;
 348         unsigned int blk;
 349
 350         if (ext4_has_feature_journal(inode->i_sb) &&
 351             (inode->i_ino ==
 352              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 353                 return 0;
 354
 355         while (bref < p+max) {
 356                 blk = le32_to_cpu(*bref++);
 357                 if (blk &&
 358                     unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
 359                                                     blk, 1))) {
 360                         es->s_last_error_block = cpu_to_le64(blk);
 361                         ext4_error_inode(inode, function, line, blk,
 362                                          "invalid block");
 363                         return -EFSCORRUPTED;
 364                 }
 365         }
 366         return 0;
 367 }
 368