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         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 = NULL;
  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         int first = 1;
 130
 131         printk(KERN_INFO "System zones: ");
 132         node = rb_first(&sbi->system_blks->root);
 133         while (node) {
 134                 entry = rb_entry(node, struct ext4_system_zone, node);
 135                 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
 136                        entry->start_blk, entry->start_blk + entry->count - 1);
 137                 first = 0;
 138                 node = rb_next(node);
 139         }
 140         printk(KERN_CONT "\n");
 141 }
 142
 143 /*
 144  * Returns 1 if the passed-in block region (start_blk,
 145  * start_blk+count) is valid; 0 if some part of the block region
 146  * overlaps with filesystem metadata blocks.
 147  */
 148 static int ext4_data_block_valid_rcu(struct ext4_sb_info *sbi,
 149                                      struct ext4_system_blocks *system_blks,
 150                                      ext4_fsblk_t start_blk,
 151                                      unsigned int count, ino_t ino)
 152 {
 153         struct ext4_system_zone *entry;
 154         struct rb_node *n;
 155
 156         if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 157             (start_blk + count < start_blk) ||
 158             (start_blk + count > ext4_blocks_count(sbi->s_es))) {
 159                 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 160                 return 0;
 161         }
 162
 163         if (system_blks == NULL)
 164                 return 1;
 165
 166         n = system_blks->root.rb_node;
 167         while (n) {
 168                 entry = rb_entry(n, struct ext4_system_zone, node);
 169                 if (start_blk + count - 1 < entry->start_blk)
 170                         n = n->rb_left;
 171                 else if (start_blk >= (entry->start_blk + entry->count))
 172                         n = n->rb_right;
 173                 else {
 174                         if (entry->ino == ino)
 175                                 return 1;
 176                         sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 177                         return 0;
 178                 }
 179         }
 180         return 1;
 181 }
 182
 183 static int ext4_protect_reserved_inode(struct super_block *sb,
 184                                        struct ext4_system_blocks *system_blks,
 185                                        u32 ino)
 186 {
 187         struct inode *inode;
 188         struct ext4_sb_info *sbi = EXT4_SB(sb);
 189         struct ext4_map_blocks map;
 190         u32 i = 0, num;
 191         int err = 0, n;
 192
 193         if ((ino < EXT4_ROOT_INO) ||
 194             (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
 195                 return -EINVAL;
 196         inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
 197         if (IS_ERR(inode))
 198                 return PTR_ERR(inode);
 199         num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 200         while (i < num) {
 201                 cond_resched();
 202                 map.m_lblk = i;
 203                 map.m_len = num - i;
 204                 n = ext4_map_blocks(NULL, inode, &map, 0);
 205                 if (n < 0) {
 206                         err = n;
 207                         break;
 208                 }
 209                 if (n == 0) {
 210                         i++;
 211                 } else {
 212                         err = add_system_zone(system_blks, map.m_pblk, n, ino);
 213                         if (err < 0) {
 214                                 if (err == -EFSCORRUPTED) {
 215                                         ext4_error(sb,
 216                                            "blocks %llu-%llu from inode %u "
 217                                            "overlap system zone", map.m_pblk,
 218                                            map.m_pblk + map.m_len - 1, ino);
 219                                 }
 220                                 break;
 221                         }
 222                         i += n;
 223                 }
 224         }
 225         iput(inode);
 226         return err;
 227 }
 228
 229 static void ext4_destroy_system_zone(struct rcu_head *rcu)
 230 {
 231         struct ext4_system_blocks *system_blks;
 232
 233         system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
 234         release_system_zone(system_blks);
 235         kfree(system_blks);
 236 }
 237
 238 /*
 239  * Build system zone rbtree which is used for block validity checking.
 240  *
 241  * The update of system_blks pointer in this function is protected by
 242  * sb->s_umount semaphore. However we have to be careful as we can be
 243  * racing with ext4_data_block_valid() calls reading system_blks rbtree
 244  * protected only by RCU. That's why we first build the rbtree and then
 245  * swap it in place.
 246  */
 247 int ext4_setup_system_zone(struct super_block *sb)
 248 {
 249         ext4_group_t ngroups = ext4_get_groups_count(sb);
 250         struct ext4_sb_info *sbi = EXT4_SB(sb);
 251         struct ext4_system_blocks *system_blks;
 252         struct ext4_group_desc *gdp;
 253         ext4_group_t i;
 254         int flex_size = ext4_flex_bg_size(sbi);
 255         int ret;
 256
 257         system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
 258         if (!system_blks)
 259                 return -ENOMEM;
 260
 261         for (i=0; i < ngroups; i++) {
 262                 if (ext4_bg_has_super(sb, i) &&
 263                     ((i < 5) || ((i % flex_size) == 0)))
 264                         add_system_zone(system_blks,
 265                                         ext4_group_first_block_no(sb, i),
 266                                         ext4_bg_num_gdb(sb, i) + 1, 0);
 267                 gdp = ext4_get_group_desc(sb, i, NULL);
 268                 ret = add_system_zone(system_blks,
 269                                 ext4_block_bitmap(sb, gdp), 1, 0);
 270                 if (ret)
 271                         goto err;
 272                 ret = add_system_zone(system_blks,
 273                                 ext4_inode_bitmap(sb, gdp), 1, 0);
 274                 if (ret)
 275                         goto err;
 276                 ret = add_system_zone(system_blks,
 277                                 ext4_inode_table(sb, gdp),
 278                                 sbi->s_itb_per_group, 0);
 279                 if (ret)
 280                         goto err;
 281         }
 282         if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
 283                 ret = ext4_protect_reserved_inode(sb, system_blks,
 284                                 le32_to_cpu(sbi->s_es->s_journal_inum));
 285                 if (ret)
 286                         goto err;
 287         }
 288
 289         /*
 290          * System blks rbtree complete, announce it once to prevent racing
 291          * with ext4_data_block_valid() accessing the rbtree at the same
 292          * time.
 293          */
 294         rcu_assign_pointer(sbi->system_blks, system_blks);
 295
 296         if (test_opt(sb, DEBUG))
 297                 debug_print_tree(sbi);
 298         return 0;
 299 err:
 300         release_system_zone(system_blks);
 301         kfree(system_blks);
 302         return ret;
 303 }
 304
 305 /*
 306  * Called when the filesystem is unmounted or when remounting it with
 307  * noblock_validity specified.
 308  *
 309  * The update of system_blks pointer in this function is protected by
 310  * sb->s_umount semaphore. However we have to be careful as we can be
 311  * racing with ext4_data_block_valid() calls reading system_blks rbtree
 312  * protected only by RCU. So we first clear the system_blks pointer and
 313  * then free the rbtree only after RCU grace period expires.
 314  */
 315 void ext4_release_system_zone(struct super_block *sb)
 316 {
 317         struct ext4_system_blocks *system_blks;
 318
 319         system_blks = rcu_dereference_protected(EXT4_SB(sb)->system_blks,
 320                                         lockdep_is_held(&sb->s_umount));
 321         rcu_assign_pointer(EXT4_SB(sb)->system_blks, NULL);
 322
 323         if (system_blks)
 324                 call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
 325 }
 326
 327 int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
 328                           unsigned int count)
 329 {
 330         struct ext4_system_blocks *system_blks;
 331         int ret;
 332
 333         /*
 334          * Lock the system zone to prevent it being released concurrently
 335          * when doing a remount which inverse current "[no]block_validity"
 336          * mount option.
 337          */
 338         rcu_read_lock();
 339         system_blks = rcu_dereference(EXT4_SB(inode->i_sb)->system_blks);
 340         ret = ext4_data_block_valid_rcu(EXT4_SB(inode->i_sb), system_blks,
 341                                         start_blk, count, inode->i_ino);
 342         rcu_read_unlock();
 343         return ret;
 344 }
 345
 346 int ext4_check_blockref(const char *function, unsigned int line,
 347                         struct inode *inode, __le32 *p, unsigned int max)
 348 {
 349         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 350         __le32 *bref = p;
 351         unsigned int blk;
 352
 353         if (ext4_has_feature_journal(inode->i_sb) &&
 354             (inode->i_ino ==
 355              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 356                 return 0;
 357
 358         while (bref < p+max) {
 359                 blk = le32_to_cpu(*bref++);
 360                 if (blk &&
 361                     unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
 362                         es->s_last_error_block = cpu_to_le64(blk);
 363                         ext4_error_inode(inode, function, line, blk,
 364                                          "invalid block");
 365                         return -EFSCORRUPTED;
 366                 }
 367         }
 368         return 0;
 369 }
 370