fs/ext4/block_validity.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  *  linux/fs/ext4/block_validity.c
0004  *
0005  * Copyright (C) 2009
0006  * Theodore Ts'o (tytso@mit.edu)
0007  *
0008  * Track which blocks in the filesystem are metadata blocks that
0009  * should never be used as data blocks by files or directories.
0010  */
0011
0012 #include <linux/time.h>
0013 #include <linux/fs.h>
0014 #include <linux/namei.h>
0015 #include <linux/quotaops.h>
0016 #include <linux/buffer_head.h>
0017 #include <linux/swap.h>
0018 #include <linux/pagemap.h>
0019 #include <linux/blkdev.h>
0020 #include <linux/slab.h>
0021 #include "ext4.h"
0022
0023 struct ext4_system_zone {
0024     struct rb_node  node;
0025     ext4_fsblk_t    start_blk;
0026     unsigned int    count;
0027     u32     ino;
0028 };
0029
0030 static struct kmem_cache *ext4_system_zone_cachep;
0031
0032 int __init ext4_init_system_zone(void)
0033 {
0034     ext4_system_zone_cachep = KMEM_CACHE(ext4_system_zone, 0);
0035     if (ext4_system_zone_cachep == NULL)
0036         return -ENOMEM;
0037     return 0;
0038 }
0039
0040 void ext4_exit_system_zone(void)
0041 {
0042     rcu_barrier();
0043     kmem_cache_destroy(ext4_system_zone_cachep);
0044 }
0045
0046 static inline int can_merge(struct ext4_system_zone *entry1,
0047              struct ext4_system_zone *entry2)
0048 {
0049     if ((entry1->start_blk + entry1->count) == entry2->start_blk &&
0050         entry1->ino == entry2->ino)
0051         return 1;
0052     return 0;
0053 }
0054
0055 static void release_system_zone(struct ext4_system_blocks *system_blks)
0056 {
0057     struct ext4_system_zone *entry, *n;
0058
0059     rbtree_postorder_for_each_entry_safe(entry, n,
0060                 &system_blks->root, node)
0061         kmem_cache_free(ext4_system_zone_cachep, entry);
0062 }
0063
0064 /*
0065  * Mark a range of blocks as belonging to the "system zone" --- that
0066  * is, filesystem metadata blocks which should never be used by
0067  * inodes.
0068  */
0069 static int add_system_zone(struct ext4_system_blocks *system_blks,
0070                ext4_fsblk_t start_blk,
0071                unsigned int count, u32 ino)
0072 {
0073     struct ext4_system_zone *new_entry, *entry;
0074     struct rb_node **n = &system_blks->root.rb_node, *node;
0075     struct rb_node *parent = NULL, *new_node = NULL;
0076
0077     while (*n) {
0078         parent = *n;
0079         entry = rb_entry(parent, struct ext4_system_zone, node);
0080         if (start_blk < entry->start_blk)
0081             n = &(*n)->rb_left;
0082         else if (start_blk >= (entry->start_blk + entry->count))
0083             n = &(*n)->rb_right;
0084         else    /* Unexpected overlap of system zones. */
0085             return -EFSCORRUPTED;
0086     }
0087
0088     new_entry = kmem_cache_alloc(ext4_system_zone_cachep,
0089                      GFP_KERNEL);
0090     if (!new_entry)
0091         return -ENOMEM;
0092     new_entry->start_blk = start_blk;
0093     new_entry->count = count;
0094     new_entry->ino = ino;
0095     new_node = &new_entry->node;
0096
0097     rb_link_node(new_node, parent, n);
0098     rb_insert_color(new_node, &system_blks->root);
0099
0100     /* Can we merge to the left? */
0101     node = rb_prev(new_node);
0102     if (node) {
0103         entry = rb_entry(node, struct ext4_system_zone, node);
0104         if (can_merge(entry, new_entry)) {
0105             new_entry->start_blk = entry->start_blk;
0106             new_entry->count += entry->count;
0107             rb_erase(node, &system_blks->root);
0108             kmem_cache_free(ext4_system_zone_cachep, entry);
0109         }
0110     }
0111
0112     /* Can we merge to the right? */
0113     node = rb_next(new_node);
0114     if (node) {
0115         entry = rb_entry(node, struct ext4_system_zone, node);
0116         if (can_merge(new_entry, entry)) {
0117             new_entry->count += entry->count;
0118             rb_erase(node, &system_blks->root);
0119             kmem_cache_free(ext4_system_zone_cachep, entry);
0120         }
0121     }
0122     return 0;
0123 }
0124
0125 static void debug_print_tree(struct ext4_sb_info *sbi)
0126 {
0127     struct rb_node *node;
0128     struct ext4_system_zone *entry;
0129     struct ext4_system_blocks *system_blks;
0130     int first = 1;
0131
0132     printk(KERN_INFO "System zones: ");
0133     rcu_read_lock();
0134     system_blks = rcu_dereference(sbi->s_system_blks);
0135     node = rb_first(&system_blks->root);
0136     while (node) {
0137         entry = rb_entry(node, struct ext4_system_zone, node);
0138         printk(KERN_CONT "%s%llu-%llu", first ? "" : ", ",
0139                entry->start_blk, entry->start_blk + entry->count - 1);
0140         first = 0;
0141         node = rb_next(node);
0142     }
0143     rcu_read_unlock();
0144     printk(KERN_CONT "\n");
0145 }
0146
0147 static int ext4_protect_reserved_inode(struct super_block *sb,
0148                        struct ext4_system_blocks *system_blks,
0149                        u32 ino)
0150 {
0151     struct inode *inode;
0152     struct ext4_sb_info *sbi = EXT4_SB(sb);
0153     struct ext4_map_blocks map;
0154     u32 i = 0, num;
0155     int err = 0, n;
0156
0157     if ((ino < EXT4_ROOT_INO) ||
0158         (ino > le32_to_cpu(sbi->s_es->s_inodes_count)))
0159         return -EINVAL;
0160     inode = ext4_iget(sb, ino, EXT4_IGET_SPECIAL);
0161     if (IS_ERR(inode))
0162         return PTR_ERR(inode);
0163     num = (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
0164     while (i < num) {
0165         cond_resched();
0166         map.m_lblk = i;
0167         map.m_len = num - i;
0168         n = ext4_map_blocks(NULL, inode, &map, 0);
0169         if (n < 0) {
0170             err = n;
0171             break;
0172         }
0173         if (n == 0) {
0174             i++;
0175         } else {
0176             err = add_system_zone(system_blks, map.m_pblk, n, ino);
0177             if (err < 0) {
0178                 if (err == -EFSCORRUPTED) {
0179                     EXT4_ERROR_INODE_ERR(inode, -err,
0180                         "blocks %llu-%llu from inode overlap system zone",
0181                         map.m_pblk,
0182                         map.m_pblk + map.m_len - 1);
0183                 }
0184                 break;
0185             }
0186             i += n;
0187         }
0188     }
0189     iput(inode);
0190     return err;
0191 }
0192
0193 static void ext4_destroy_system_zone(struct rcu_head *rcu)
0194 {
0195     struct ext4_system_blocks *system_blks;
0196
0197     system_blks = container_of(rcu, struct ext4_system_blocks, rcu);
0198     release_system_zone(system_blks);
0199     kfree(system_blks);
0200 }
0201
0202 /*
0203  * Build system zone rbtree which is used for block validity checking.
0204  *
0205  * The update of system_blks pointer in this function is protected by
0206  * sb->s_umount semaphore. However we have to be careful as we can be
0207  * racing with ext4_inode_block_valid() calls reading system_blks rbtree
0208  * protected only by RCU. That's why we first build the rbtree and then
0209  * swap it in place.
0210  */
0211 int ext4_setup_system_zone(struct super_block *sb)
0212 {
0213     ext4_group_t ngroups = ext4_get_groups_count(sb);
0214     struct ext4_sb_info *sbi = EXT4_SB(sb);
0215     struct ext4_system_blocks *system_blks;
0216     struct ext4_group_desc *gdp;
0217     ext4_group_t i;
0218     int flex_size = ext4_flex_bg_size(sbi);
0219     int ret;
0220
0221     system_blks = kzalloc(sizeof(*system_blks), GFP_KERNEL);
0222     if (!system_blks)
0223         return -ENOMEM;
0224
0225     for (i=0; i < ngroups; i++) {
0226         cond_resched();
0227         if (ext4_bg_has_super(sb, i) &&
0228             ((i < 5) || ((i % flex_size) == 0))) {
0229             ret = add_system_zone(system_blks,
0230                     ext4_group_first_block_no(sb, i),
0231                     ext4_bg_num_gdb(sb, i) + 1, 0);
0232             if (ret)
0233                 goto err;
0234         }
0235         gdp = ext4_get_group_desc(sb, i, NULL);
0236         ret = add_system_zone(system_blks,
0237                 ext4_block_bitmap(sb, gdp), 1, 0);
0238         if (ret)
0239             goto err;
0240         ret = add_system_zone(system_blks,
0241                 ext4_inode_bitmap(sb, gdp), 1, 0);
0242         if (ret)
0243             goto err;
0244         ret = add_system_zone(system_blks,
0245                 ext4_inode_table(sb, gdp),
0246                 sbi->s_itb_per_group, 0);
0247         if (ret)
0248             goto err;
0249     }
0250     if (ext4_has_feature_journal(sb) && sbi->s_es->s_journal_inum) {
0251         ret = ext4_protect_reserved_inode(sb, system_blks,
0252                 le32_to_cpu(sbi->s_es->s_journal_inum));
0253         if (ret)
0254             goto err;
0255     }
0256
0257     /*
0258      * System blks rbtree complete, announce it once to prevent racing
0259      * with ext4_inode_block_valid() accessing the rbtree at the same
0260      * time.
0261      */
0262     rcu_assign_pointer(sbi->s_system_blks, system_blks);
0263
0264     if (test_opt(sb, DEBUG))
0265         debug_print_tree(sbi);
0266     return 0;
0267 err:
0268     release_system_zone(system_blks);
0269     kfree(system_blks);
0270     return ret;
0271 }
0272
0273 /*
0274  * Called when the filesystem is unmounted or when remounting it with
0275  * noblock_validity specified.
0276  *
0277  * The update of system_blks pointer in this function is protected by
0278  * sb->s_umount semaphore. However we have to be careful as we can be
0279  * racing with ext4_inode_block_valid() calls reading system_blks rbtree
0280  * protected only by RCU. So we first clear the system_blks pointer and
0281  * then free the rbtree only after RCU grace period expires.
0282  */
0283 void ext4_release_system_zone(struct super_block *sb)
0284 {
0285     struct ext4_system_blocks *system_blks;
0286
0287     system_blks = rcu_dereference_protected(EXT4_SB(sb)->s_system_blks,
0288                     lockdep_is_held(&sb->s_umount));
0289     rcu_assign_pointer(EXT4_SB(sb)->s_system_blks, NULL);
0290
0291     if (system_blks)
0292         call_rcu(&system_blks->rcu, ext4_destroy_system_zone);
0293 }
0294
0295 int ext4_sb_block_valid(struct super_block *sb, struct inode *inode,
0296                 ext4_fsblk_t start_blk, unsigned int count)
0297 {
0298     struct ext4_sb_info *sbi = EXT4_SB(sb);
0299     struct ext4_system_blocks *system_blks;
0300     struct ext4_system_zone *entry;
0301     struct rb_node *n;
0302     int ret = 1;
0303
0304     if ((start_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
0305         (start_blk + count < start_blk) ||
0306         (start_blk + count > ext4_blocks_count(sbi->s_es)))
0307         return 0;
0308
0309     /*
0310      * Lock the system zone to prevent it being released concurrently
0311      * when doing a remount which inverse current "[no]block_validity"
0312      * mount option.
0313      */
0314     rcu_read_lock();
0315     system_blks = rcu_dereference(sbi->s_system_blks);
0316     if (system_blks == NULL)
0317         goto out_rcu;
0318
0319     n = system_blks->root.rb_node;
0320     while (n) {
0321         entry = rb_entry(n, struct ext4_system_zone, node);
0322         if (start_blk + count - 1 < entry->start_blk)
0323             n = n->rb_left;
0324         else if (start_blk >= (entry->start_blk + entry->count))
0325             n = n->rb_right;
0326         else {
0327             ret = 0;
0328             if (inode)
0329                 ret = (entry->ino == inode->i_ino);
0330             break;
0331         }
0332     }
0333 out_rcu:
0334     rcu_read_unlock();
0335     return ret;
0336 }
0337
0338 /*
0339  * Returns 1 if the passed-in block region (start_blk,
0340  * start_blk+count) is valid; 0 if some part of the block region
0341  * overlaps with some other filesystem metadata blocks.
0342  */
0343 int ext4_inode_block_valid(struct inode *inode, ext4_fsblk_t start_blk,
0344               unsigned int count)
0345 {
0346     return ext4_sb_block_valid(inode->i_sb, inode, start_blk, count);
0347 }
0348
0349 int ext4_check_blockref(const char *function, unsigned int line,
0350             struct inode *inode, __le32 *p, unsigned int max)
0351 {
0352     __le32 *bref = p;
0353     unsigned int blk;
0354
0355     if (ext4_has_feature_journal(inode->i_sb) &&
0356         (inode->i_ino ==
0357          le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
0358         return 0;
0359
0360     while (bref < p+max) {
0361         blk = le32_to_cpu(*bref++);
0362         if (blk &&
0363             unlikely(!ext4_inode_block_valid(inode, blk, 1))) {
0364             ext4_error_inode(inode, function, line, blk,
0365                      "invalid block");
0366             return -EFSCORRUPTED;
0367         }
0368     }
0369     return 0;
0370 }
0371