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/mutex.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 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 return 1; 49 return 0; 50 } 51 52 /* 53 * Mark a range of blocks as belonging to the "system zone" --- that 54 * is, filesystem metadata blocks which should never be used by 55 * inodes. 56 */ 57 static int add_system_zone(struct ext4_sb_info *sbi, 58 ext4_fsblk_t start_blk, 59 unsigned int count) 60 { 61 struct ext4_system_zone *new_entry = NULL, *entry; 62 struct rb_node **n = &sbi->system_blks.rb_node, *node; 63 struct rb_node *parent = NULL, *new_node = NULL; 64 65 while (*n) { 66 parent = *n; 67 entry = rb_entry(parent, struct ext4_system_zone, node); 68 if (start_blk < entry->start_blk) 69 n = &(*n)->rb_left; 70 else if (start_blk >= (entry->start_blk + entry->count)) 71 n = &(*n)->rb_right; 72 else { 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 } 82 } 83 84 if (!new_entry) { 85 new_entry = kmem_cache_alloc(ext4_system_zone_cachep, 86 GFP_KERNEL); 87 if (!new_entry) 88 return -ENOMEM; 89 new_entry->start_blk = start_blk; 90 new_entry->count = count; 91 new_node = &new_entry->node; 92 93 rb_link_node(new_node, parent, n); 94 rb_insert_color(new_node, &sbi->system_blks); 95 } 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, &sbi->system_blks); 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, &sbi->system_blks); 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); 130 while (node) { 131 entry = rb_entry(node, struct ext4_system_zone, node); 132 printk("%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("\n"); 138 } 139 140 int ext4_setup_system_zone(struct super_block *sb) 141 { 142 ext4_group_t ngroups = ext4_get_groups_count(sb); 143 struct ext4_sb_info *sbi = EXT4_SB(sb); 144 struct ext4_group_desc *gdp; 145 ext4_group_t i; 146 int flex_size = ext4_flex_bg_size(sbi); 147 int ret; 148 149 if (!test_opt(sb, BLOCK_VALIDITY)) { 150 if (EXT4_SB(sb)->system_blks.rb_node) 151 ext4_release_system_zone(sb); 152 return 0; 153 } 154 if (EXT4_SB(sb)->system_blks.rb_node) 155 return 0; 156 157 for (i=0; i < ngroups; i++) { 158 if (ext4_bg_has_super(sb, i) && 159 ((i < 5) || ((i % flex_size) == 0))) 160 add_system_zone(sbi, ext4_group_first_block_no(sb, i), 161 ext4_bg_num_gdb(sb, i) + 1); 162 gdp = ext4_get_group_desc(sb, i, NULL); 163 ret = add_system_zone(sbi, ext4_block_bitmap(sb, gdp), 1); 164 if (ret) 165 return ret; 166 ret = add_system_zone(sbi, ext4_inode_bitmap(sb, gdp), 1); 167 if (ret) 168 return ret; 169 ret = add_system_zone(sbi, ext4_inode_table(sb, gdp), 170 sbi->s_itb_per_group); 171 if (ret) 172 return ret; 173 } 174 175 if (test_opt(sb, DEBUG)) 176 debug_print_tree(EXT4_SB(sb)); 177 return 0; 178 } 179 180 /* Called when the filesystem is unmounted */ 181 void ext4_release_system_zone(struct super_block *sb) 182 { 183 struct rb_node *n = EXT4_SB(sb)->system_blks.rb_node; 184 struct rb_node *parent; 185 struct ext4_system_zone *entry; 186 187 while (n) { 188 /* Do the node's children first */ 189 if (n->rb_left) { 190 n = n->rb_left; 191 continue; 192 } 193 if (n->rb_right) { 194 n = n->rb_right; 195 continue; 196 } 197 /* 198 * The node has no children; free it, and then zero 199 * out parent's link to it. Finally go to the 200 * beginning of the loop and try to free the parent 201 * node. 202 */ 203 parent = rb_parent(n); 204 entry = rb_entry(n, struct ext4_system_zone, node); 205 kmem_cache_free(ext4_system_zone_cachep, entry); 206 if (!parent) 207 EXT4_SB(sb)->system_blks = RB_ROOT; 208 else if (parent->rb_left == n) 209 parent->rb_left = NULL; 210 else if (parent->rb_right == n) 211 parent->rb_right = NULL; 212 n = parent; 213 } 214 EXT4_SB(sb)->system_blks = RB_ROOT; 215 } 216 217 /* 218 * Returns 1 if the passed-in block region (start_blk, 219 * start_blk+count) is valid; 0 if some part of the block region 220 * overlaps with filesystem metadata blocks. 221 */ 222 int ext4_data_block_valid(struct ext4_sb_info *sbi, ext4_fsblk_t start_blk, 223 unsigned int count) 224 { 225 struct ext4_system_zone *entry; 226 struct rb_node *n = sbi->system_blks.rb_node; 227 228 if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 229 (start_blk + count < start_blk) || 230 (start_blk + count > ext4_blocks_count(sbi->s_es))) { 231 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk); 232 return 0; 233 } 234 while (n) { 235 entry = rb_entry(n, struct ext4_system_zone, node); 236 if (start_blk + count - 1 < entry->start_blk) 237 n = n->rb_left; 238 else if (start_blk >= (entry->start_blk + entry->count)) 239 n = n->rb_right; 240 else { 241 sbi->s_es->s_last_error_block = cpu_to_le64(start_blk); 242 return 0; 243 } 244 } 245 return 1; 246 } 247 248 int ext4_check_blockref(const char *function, unsigned int line, 249 struct inode *inode, __le32 *p, unsigned int max) 250 { 251 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es; 252 __le32 *bref = p; 253 unsigned int blk; 254 255 while (bref < p+max) { 256 blk = le32_to_cpu(*bref++); 257 if (blk && 258 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb), 259 blk, 1))) { 260 es->s_last_error_block = cpu_to_le64(blk); 261 ext4_error_inode(inode, function, line, blk, 262 "invalid block"); 263 return -EIO; 264 } 265 } 266 return 0; 267 } 268 269