fs/ext4/block_validity.c

   1 /*
   2  *  linux/fs/ext4/block_validity.c
   3  *
   4  * Copyright (C) 2009
   5  * Theodore Ts'o (tytso@mit.edu)
   6  *
   7  * Track which blocks in the filesystem are metadata blocks that
   8  * should never be used as data blocks by files or directories.
   9  */
  10
  11 #include <linux/time.h>
  12 #include <linux/fs.h>
  13 #include <linux/namei.h>
  14 #include <linux/quotaops.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/swap.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/slab.h>
  20 #include "ext4.h"
  21
  22 struct ext4_system_zone {
  23         struct rb_node  node;
  24         ext4_fsblk_t    start_blk;
  25         unsigned int    count;
  26         u32             ino;
  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         kmem_cache_destroy(ext4_system_zone_cachep);
  42 }
  43
  44 static inline int can_merge(struct ext4_system_zone *entry1,
  45                      struct ext4_system_zone *entry2)
  46 {
  47         if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
  48             entry1->ino == entry2->ino)
  49                 return 1;
  50         return 0;
  51 }
  52
  53 /*
  54  * Mark a range of blocks as belonging to the "system zone" --- that
  55  * is, filesystem metadata blocks which should never be used by
  56  * inodes.
  57  */
  58 static int add_system_zone(struct ext4_sb_info *sbi,
  59                            ext4_fsblk_t start_blk,
  60                            unsigned int count, u32 ino)
  61 {
  62         struct ext4_system_zone *new_entry, *entry;
  63         struct rb_node **n = &sbi->system_blks.rb_node, *node;
  64         struct rb_node *parent = NULL, *new_node = NULL;
  65
  66         while (*n) {
  67                 parent = *n;
  68                 entry = rb_entry(parent, struct ext4_system_zone, node);
  69                 if (start_blk < entry->start_blk)
  70                         n = &(*n)->rb_left;
  71                 else if (start_blk >= (entry->start_blk + entry->count))
  72                         n = &(*n)->rb_right;
  73                 else    /* Unexpected overlap of system zones. */
  74                         return -EFSCORRUPTED;
  75         }
  76
  77         new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
  78                                      GFP_KERNEL);
  79         if (!new_entry)
  80                 return -ENOMEM;
  81         new_entry->start_blk = start_blk;
  82         new_entry->count = count;
  83         new_entry->ino = ino;
  84         new_node = &new_entry->node;
  85
  86         rb_link_node(new_node, parent, n);
  87         rb_insert_color(new_node, &sbi->system_blks);
  88
  89         /* Can we merge to the left? */
  90         node = rb_prev(new_node);
  91         if (node) {
  92                 entry = rb_entry(node, struct ext4_system_zone, node);
  93                 if (can_merge(entry, new_entry)) {
  94                         new_entry->start_blk = entry->start_blk;
  95                         new_entry->count += entry->count;
  96                         rb_erase(node, &sbi->system_blks);
  97                         kmem_cache_free(ext4_system_zone_cachep, entry);
  98                 }
  99         }
 100
 101         /* Can we merge to the right? */
 102         node = rb_next(new_node);
 103         if (node) {
 104                 entry = rb_entry(node, struct ext4_system_zone, node);
 105                 if (can_merge(new_entry, entry)) {
 106                         new_entry->count += entry->count;
 107                         rb_erase(node, &sbi->system_blks);
 108                         kmem_cache_free(ext4_system_zone_cachep, entry);
 109                 }
 110         }
 111         return 0;
 112 }
 113
 114 static void debug_print_tree(struct ext4_sb_info *sbi)
 115 {
 116         struct rb_node *node;
 117         struct ext4_system_zone *entry;
 118         int first = 1;
 119
 120         printk(KERN_INFO "System zones: ");
 121         node = rb_first(&sbi->system_blks);
 122         while (node) {
 123                 entry = rb_entry(node, struct ext4_system_zone, node);
 124                 printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
 125                        entry->start_blk, entry->start_blk + entry->count - 1);
 126                 first = 0;
 127                 node = rb_next(node);
 128         }
 129         printk(KERN_CONT "\n");
 130 }
 131
 132 static int ext4_protect_reserved_inode(struct super_block *sb, u32 ino)
 133 {
 134         struct inode *inode;
 135         struct ext4_sb_info *sbi = EXT4_SB(sb);
 136         struct ext4_map_blocks map;
 137         u32 i = 0, num;
 138         int err = 0, n;
 139
 140         if ((ino < EXT4_ROOT_INO) ||
 141             (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
 142                 return -EINVAL;
 143         inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
 144         if (IS_ERR(inode))
 145                 return PTR_ERR(inode);
 146         num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 147         while (i < num) {
 148                 cond_resched();
 149                 map.m_lblk = i;
 150                 map.m_len = num - i;
 151                 n = ext4_map_blocks(NULL, inode, &map, 0);
 152                 if (n < 0) {
 153                         err = n;
 154                         break;
 155                 }
 156                 if (n == 0) {
 157                         i++;
 158                 } else {
 159                         err = add_system_zone(sbi, map.m_pblk, n, ino);
 160                         if (err < 0) {
 161                                 if (err == -EFSCORRUPTED) {
 162                                         ext4_error(sb,
 163                                            "blocks %llu-%llu from inode %u "
 164                                            "overlap system zone", map.m_pblk,
 165                                            map.m_pblk + map.m_len - 1, ino);
 166                                 }
 167                                 break;
 168                         }
 169                         i += n;
 170                 }
 171         }
 172         iput(inode);
 173         return err;
 174 }
 175
 176 int ext4_setup_system_zone(struct super_block *sb)
 177 {
 178         ext4_group_t ngroups = ext4_get_groups_count(sb);
 179         struct ext4_sb_info *sbi = EXT4_SB(sb);
 180         struct ext4_group_desc *gdp;
 181         ext4_group_t i;
 182         int flex_size = ext4_flex_bg_size(sbi);
 183         int ret;
 184
 185         if (!test_opt(sb, BLOCK_VALIDITY)) {
 186                 if (EXT4_SB(sb)->system_blks.rb_node)
 187                         ext4_release_system_zone(sb);
 188                 return 0;
 189         }
 190         if (EXT4_SB(sb)->system_blks.rb_node)
 191                 return 0;
 192
 193         for (i=0; i < ngroups; i++) {
 194                 if (ext4_bg_has_super(sb, i) &&
 195                     ((i < 5) || ((i % flex_size) == 0)))
 196                         add_system_zone(sbi, ext4_group_first_block_no(sb, i),
 197                                         ext4_bg_num_gdb(sb, i) + 1, 0);
 198                 gdp = ext4_get_group_desc(sb, i, NULL);
 199                 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1, 0);
 200                 if (ret)
 201                         return ret;
 202                 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1, 0);
 203                 if (ret)
 204                         return ret;
 205                 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp),
 206                                 sbi->s_itb_per_group, 0);
 207                 if (ret)
 208                         return ret;
 209         }
 210         if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
 211                 ret = ext4_protect_reserved_inode(sb,
 212                                 le32_to_cpu(sbi->s_es->s_journal_inum));
 213                 if (ret)
 214                         return ret;
 215         }
 216
 217         if (test_opt(sb, DEBUG))
 218                 debug_print_tree(EXT4_SB(sb));
 219         return 0;
 220 }
 221
 222 /* Called when the filesystem is unmounted */
 223 void ext4_release_system_zone(struct super_block *sb)
 224 {
 225         struct ext4_system_zone *entry, *n;
 226
 227         rbtree_postorder_for_each_entry_safe(entry, n,
 228                         &EXT4_SB(sb)->system_blks, node)
 229                 kmem_cache_free(ext4_system_zone_cachep, entry);
 230
 231         EXT4_SB(sb)->system_blks = RB_ROOT;
 232 }
 233
 234 /*
 235  * Returns 1 if the passed-in block region (start_blk,
 236  * start_blk+count) is valid; 0 if some part of the block region
 237  * overlaps with filesystem metadata blocks.
 238  */
 239 int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
 240                            unsigned int count)
 241 {
 242         struct ext4_system_zone *entry;
 243         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 244         struct rb_node *n = sbi->system_blks.rb_node;
 245
 246         if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 247             (start_blk + count < start_blk) ||
 248             (start_blk + count > ext4_blocks_count(sbi->s_es))) {
 249                 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 250                 return 0;
 251         }
 252         while (n) {
 253                 entry = rb_entry(n, struct ext4_system_zone, node);
 254                 if (start_blk + count - 1 < entry->start_blk)
 255                         n = n->rb_left;
 256                 else if (start_blk >= (entry->start_blk + entry->count))
 257                         n = n->rb_right;
 258                 else {
 259                         if (entry->ino == inode->i_ino)
 260                                 return 1;
 261                         sbi->s_es->s_last_error_block = cpu_to_le64(start_blk);
 262                         return 0;
 263                 }
 264         }
 265         return 1;
 266 }
 267
 268 int ext4_check_blockref(const char *function, unsigned int line,
 269                         struct inode *inode, __le32 *p, unsigned int max)
 270 {
 271         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 272         __le32 *bref = p;
 273         unsigned int blk;
 274
 275         if (ext4_has_feature_journal(inode->i_sb) &&
 276             (inode->i_ino ==
 277              le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
 278                 return 0;
 279
 280         while (bref < p+max) {
 281                 blk = le32_to_cpu(*bref++);
 282                 if (blk &&
 283                     unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
 284                         es->s_last_error_block = cpu_to_le64(blk);
 285                         ext4_error_inode(inode, function, line, blk,
 286                                          "invalid block");
 287                         return -EFSCORRUPTED;
 288                 }
 289         }
 290         return 0;
 291 }
 292