isofs: mounting to regular file may succeed
Subsystems (2)
PATCH 1/1 isofs: mounting to regular file may succeed
It turned out that mounting a corrupted ISO image to a regular file may succeed, e.g. if an image was prepared as follows:
$ dd if=correct.iso of=bad.iso bs=4k count=8
We then can mount it to a regular file:
# mount -o loop -t iso9660 bad.iso /tmp/file
But mounting it to a directory fails with -ENOTDIR, simply because the root directory inode doesn't have S_IFDIR set and the condition in graft_tree() is met:
if (S_ISDIR(nd->dentry->d_inode->i_mode) != S_ISDIR(mnt->mnt_root->d_inode->i_mode)) return -ENOTDIR
This is because the root directory inode was read from an incorrect block. It's supposed to be read from sbi->s_firstdatazone, which is an absolute value and gets messed up in the case of an incorrect image.
In order to somehow circumvent this we have to check that the root directory inode is actually a directory after all.
fs/isofs/inode.c +9 -0
@@ -840,6 +840,15 @@ root_found: goto out_no_root;if (!inode->i_op) goto out_bad_root; + +/* Make sure the root inode is a directory */ +if (!S_ISDIR (inode->i_mode)) { +printk (KERN_WARNING +"isofs_fill_super: root inode is not a directory. " +"Corrupted media?\n" ); + goto out_iput; + } +/* get the root dentry */ s->s_root =d_alloc_root (inode);if (!(s->s_root))
Quoted 1 line · Kirill Kuvaldin
$ dd if=correct.iso of=bad.iso bs=4k count=8
Oops, sorry, the right command should be:
dd if=correct.iso of=bad.iso bs=4k seek=8
^^^^
Kirill
Quoted 2 lines · Kirill Kuvaldin
We then can mount it to a regular file:
Wow, this is news to me. Since when is it possible to mount files to files?
Jan
Quoted 7 lines · Jan Engelhardt
Quoted 2 lines · Kirill Kuvaldin
We then can mount it to a regular file:
Wow, this is news to me. Since when is it possible to mount files to files?
It is possible to mount a regular file to another one with --bind. The problem in question is that mounting a malformed ISO 9660 image to a directory fails, but to a regular file - succeeds.
Kirill
Quoted 37 lines · On Sat, 14 Jul 2007 03:47:21 +0400 Kirill Kuvaldin <[email protected]> wrote:
It turned out that mounting a corrupted ISO image to a regular file may succeed, e.g. if an image was prepared as follows:
$ dd if=correct.iso of=bad.iso bs=4k count=8We then can mount it to a regular file:
# mount -o loop -t iso9660 bad.iso /tmp/fileBut mounting it to a directory fails with
-ENOTDIR, simply because the root directory inode doesn't haveS_IFDIRset and the condition ingraft_tree()is met:if (S_ISDIR(nd->dentry->d_inode->i_mode) != S_ISDIR(mnt->mnt_root->d_inode->i_mode)) return -ENOTDIRThis is because the root directory inode was read from an incorrect block. It's supposed to be read from
sbi->s_firstdatazone, which is an absolute value and gets messed up in the case of an incorrect image.In order to somehow circumvent this we have to check that the root directory inode is actually a directory after all.
Signed-off-by Kirill Kuvaldin <[email protected]>fs/isofs/inode.c +2
-0@@ -840,6 +840,15 @@ root_found: goto out_no_root;if (!inode->i_op) goto out_bad_root; + +/* Make sure the root inode is a directory */
fs/isofs/inode.c root_found: · line 854
Quoted 10 lines
fs/isofs/inode.c +7
-0+if (!S_ISDIR (inode->i_mode)) { +printk (KERN_WARNING +"isofs_fill_super: root inode is not a directory. " +"Corrupted media?\n" ); + goto out_iput; + } +/* get the root dentry */ s->s_root =d_alloc_root (inode);if (!(s->s_root))
I don't think any (all?) other filesystems perform checks like this. Is this something which can/should be performed at the VFS level?
Quoted 2 lines · Andrew Morton
I don't think any (all?) other filesystems perform checks like this. Is this something which can/should be performed at the VFS level?
As far as the VFS is concerned non-directory mounts are perfectly fine. There's a lot of use cases for non-directory bind-mounts and at least some for regular filesystems with a non-directory root. E.g. the streams
folks are using something like that. Solaris even ships with non-directory
root filesystems mounted by default these days.
Quoted 2 lines · Andrew Morton
I don't think any (all?) other filesystems perform checks like this. Is this something which can/should be performed at the VFS level?
I don't see why we _should_ check that; VFS checks that we don't
turn directory into non-directory and vice versa, same as for binding. There is no reason why fs driver could not give you a single-node filesystem with root being e.g. a block device node. You could mount it on any non-directory and get a block device seen there, etc.
IOW, if filesystem driver considers fs image it's working from
as invalid, it's up to filesystem driver. _What_ is considered invalid depends on fs type; what the driver cares to report as dubious is, again, up to driver. If isofs wants to warn about an ISO image that definitely violates ISO9660, it's isofs decision. VFS will work with it as long as fs driver does...