ext4: fix oops on corrupted ext4 mount
authorEric Sandeen <sandeen@redhat.com>
Tue, 29 Jan 2008 04:58:27 +0000 (23:58 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Tue, 29 Jan 2008 04:58:27 +0000 (23:58 -0500)
When mounting an ext4 filesystem with corrupted s_first_data_block, things
can go very wrong and oops.

Because blocks_count in ext4_fill_super is a u64, and we must use do_div,
the calculation of db_count is done differently than on ext4.  If
first_data_block is corrupted such that it is larger than ext4_blocks_count,
for example, then the intermediate blocks_count value may go negative,
but sign-extend to a very large value:

        blocks_count = (ext4_blocks_count(es) -
                        le32_to_cpu(es->s_first_data_block) +
                        EXT4_BLOCKS_PER_GROUP(sb) - 1);

This is then assigned to s_groups_count which is an unsigned long:

        sbi->s_groups_count = blocks_count;

This may result in a value of 0xFFFFFFFF which is then used to compute
db_count:

        db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
                   EXT4_DESC_PER_BLOCK(sb);

and in this case db_count will wind up as 0 because the addition overflows
32 bits.  This in turn causes the kmalloc for group_desc to be of 0 size:

        sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
                                    GFP_KERNEL);

and eventually in ext4_check_descriptors, dereferencing
sbi->s_group_desc[desc_block] will result in a NULL pointer dereference.

The simplest test seems to be to sanity check s_first_data_block,
EXT4_BLOCKS_PER_GROUP, and ext4_blocks_count values to be sure
their combination won't result in a bad intermediate value for
blocks_count.  We could just check for db_count == 0, but
catching it at the root cause seems like it provides more info.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Mingming Cao <cmm@us.ibm.com>
fs/ext4/super.c

index 1484a087bba0ca4784075687804d622c5c8c6a54..32e3ecb35cd7ac9c654959fda61c95dd37ebbf66 100644 (file)
@@ -1997,6 +1997,17 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
 
        if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
                goto cantfind_ext4;
+
+       /* ensure blocks_count calculation below doesn't sign-extend */
+       if (ext4_blocks_count(es) + EXT4_BLOCKS_PER_GROUP(sb) <
+           le32_to_cpu(es->s_first_data_block) + 1) {
+               printk(KERN_WARNING "EXT4-fs: bad geometry: block count %llu, "
+                      "first data block %u, blocks per group %lu\n",
+                       ext4_blocks_count(es),
+                       le32_to_cpu(es->s_first_data_block),
+                       EXT4_BLOCKS_PER_GROUP(sb));
+               goto failed_mount;
+       }
        blocks_count = (ext4_blocks_count(es) -
                        le32_to_cpu(es->s_first_data_block) +
                        EXT4_BLOCKS_PER_GROUP(sb) - 1);