[PATCH] ext4: don't remove reserved inodes in ext4_unlink()

[PATCH] ext4: don't remove reserved inodes in ext4_unlink()

[Eryu Guan]

Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
the journal inode is gone, and unmounting such a fs could trigger the
following BUG_ON() in start_this_handle().

	BUG_ON(journal->j_flags & JBD2_UNMOUNT);

	------------[ cut here ]------------
	kernel BUG at fs/jbd2/transaction.c:307!
	...
	CPU: 1 PID: 1535 Comm: umount Not tainted 3.13.0+ #14
	...
	Call Trace:
	 [<ffffffff8119f17a>] ? kmem_cache_alloc+0x1ca/0x1f0
	 [<ffffffff812850f0>] ? jbd2__journal_start+0x90/0x1e0
	 [<ffffffff81285153>] jbd2__journal_start+0xf3/0x1e0
	 [<ffffffff81242a62>] ? ext4_evict_inode+0x1b2/0x4f0
	 [<ffffffff8126d039>] __ext4_journal_start_sb+0x69/0xe0
	 [<ffffffff81242a62>] ext4_evict_inode+0x1b2/0x4f0
	 [<ffffffff811d3b8e>] evict+0x9e/0x190
	 [<ffffffff811d4373>] iput+0xf3/0x180
	 [<ffffffff8128f301>] jbd2_journal_destroy+0x191/0x220
	 [<ffffffff810b0ae0>] ? abort_exclusive_wait+0xb0/0xb0
	 [<ffffffff8125d004>] ext4_put_super+0x64/0x340
	 [<ffffffff811bbae2>] generic_shutdown_super+0x72/0xf0
	 [<ffffffff811bbd77>] kill_block_super+0x27/0x70
	 [<ffffffff811bc05d>] deactivate_locked_super+0x3d/0x60
	 [<ffffffff811bc606>] deactivate_super+0x46/0x60
	 [<ffffffff811d7f47>] mntput_no_expire+0xa7/0x140
	 [<ffffffff811d939e>] SyS_umount+0x8e/0x100
	 [<ffffffff81690c29>] system_call_fastpath+0x16/0x1b

Check inode number in ext4_unlink() and return error if the inode number
is reserved or nonexistent.

Tested by removing a reserved inode and unmounting the fs. Inodes 1-10
have been tested. Also tested by xfstests.

Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
---

If this was the right way to go, I'll send another patch for ext3 too.

 fs/ext4/namei.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 5a0408d..797ce98 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -2761,9 +2761,11 @@ end_rmdir:
 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 {
 	int retval;
+	unsigned long ino;
 	struct inode *inode;
 	struct buffer_head *bh;
 	struct ext4_dir_entry_2 *de;
+	struct super_block *sb;
 	handle_t *handle = NULL;
 
 	trace_ext4_unlink_enter(dir, dentry);
@@ -2778,13 +2780,20 @@ static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 		goto end_unlink;
 
 	inode = dentry->d_inode;
+	ino = inode->i_ino;
+	sb = dir->i_sb;
 
 	retval = -EIO;
-	if (le32_to_cpu(de->inode) != inode->i_ino)
+	if (le32_to_cpu(de->inode) != ino)
 		goto end_unlink;
+	if (ino < EXT4_FIRST_INO(sb) ||
+	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) {
+		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
+		goto end_unlink;
+	}
 
 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
-				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
+				    EXT4_DATA_TRANS_BLOCKS(sb));
 	if (IS_ERR(handle)) {
 		retval = PTR_ERR(handle);
 		handle = NULL;
-- 
1.8.5.3

Re: [PATCH] ext4: don't remove reserved inodes in ext4_unlink()

[Theodore Ts'o]

On Sat, Jan 25, 2014 at 02:58:17PM +0800, Eryu Guan wrote:
> Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
> when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
> the journal inode is gone, and unmounting such a fs could trigger the
> following BUG_ON() in start_this_handle().....

This patch is mostly good, but you need to exempt the root inode.
Otherwise, the following program, which would normally give the error
"unlink: Is a directory", will mark the file system as containing an
error, and so it could allow an unprivileged user to remount the file
system read-only, or force the system to panic and reboot.

#include <unistd.h>

int main(int argc, char **argv)
{

	if (unlink("/") < 0)
		perror("unlink");
	return 0;
}

Cheers,

					- Ted

Re: [PATCH] ext4: don't remove reserved inodes in ext4_unlink()

[Eryu Guan]

On Wed, Feb 12, 2014 at 11:38:25AM -0500, Theodore Ts'o wrote:
> On Sat, Jan 25, 2014 at 02:58:17PM +0800, Eryu Guan wrote:
> > Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
> > when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
> > the journal inode is gone, and unmounting such a fs could trigger the
> > following BUG_ON() in start_this_handle().....
> 
> This patch is mostly good, but you need to exempt the root inode.
> Otherwise, the following program, which would normally give the error
> "unlink: Is a directory", will mark the file system as containing an
> error, and so it could allow an unprivileged user to remount the file
> system read-only, or force the system to panic and reboot.

Hi Ted,

I dont' see how the following program could mark the file system as
containing an error on patched kernel. I tried running it on both
patched/unpatched kernel, and it gave "unlink: Is a directory" in both
cases. And dumpe2fs -h didn't show that fs contained errors.

I traced do_unlinkat(), vfs_unlink() and ext4_unlink() in systemtap
and systemtap showed only do_unlinkat() was being called.

Am I missing something here?

Thanks,
Eryu
> 
> #include <unistd.h>
> 
> int main(int argc, char **argv)
> {
> 
> 	if (unlink("/") < 0)
> 		perror("unlink");
> 	return 0;
> }
> 
> Cheers,
> 
> 					- Ted

[PATCH v2] ext4: don't remove reserved inodes in ext4_unlink()

[Eryu Guan]

Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
the journal inode is gone, and unmounting such a fs could trigger the
following BUG_ON() in start_this_handle().

	BUG_ON(journal->j_flags & JBD2_UNMOUNT);

	------------[ cut here ]------------
	kernel BUG at fs/jbd2/transaction.c:307!
	...
	CPU: 1 PID: 1535 Comm: umount Not tainted 3.13.0+ #14
	...
	Call Trace:
	 [<ffffffff8119f17a>] ? kmem_cache_alloc+0x1ca/0x1f0
	 [<ffffffff812850f0>] ? jbd2__journal_start+0x90/0x1e0
	 [<ffffffff81285153>] jbd2__journal_start+0xf3/0x1e0
	 [<ffffffff81242a62>] ? ext4_evict_inode+0x1b2/0x4f0
	 [<ffffffff8126d039>] __ext4_journal_start_sb+0x69/0xe0
	 [<ffffffff81242a62>] ext4_evict_inode+0x1b2/0x4f0
	 [<ffffffff811d3b8e>] evict+0x9e/0x190
	 [<ffffffff811d4373>] iput+0xf3/0x180
	 [<ffffffff8128f301>] jbd2_journal_destroy+0x191/0x220
	 [<ffffffff810b0ae0>] ? abort_exclusive_wait+0xb0/0xb0
	 [<ffffffff8125d004>] ext4_put_super+0x64/0x340
	 [<ffffffff811bbae2>] generic_shutdown_super+0x72/0xf0
	 [<ffffffff811bbd77>] kill_block_super+0x27/0x70
	 [<ffffffff811bc05d>] deactivate_locked_super+0x3d/0x60
	 [<ffffffff811bc606>] deactivate_super+0x46/0x60
	 [<ffffffff811d7f47>] mntput_no_expire+0xa7/0x140
	 [<ffffffff811d939e>] SyS_umount+0x8e/0x100
	 [<ffffffff81690c29>] system_call_fastpath+0x16/0x1b

Check inode number in ext4_unlink() and return error if the inode number
is reserved or nonexistent(except EXT4_ROOT_INO, as Ted pointed out that
it's a security hole).

Tested by removing a reserved inode(modify the ondisk structure by hand)
and unmounting the fs. Inodes 1-10 have been tested. Also tested by
xfstests.

Signed-off-by: Eryu Guan <guaneryu@gmail.com>
---

(This is a v2 of an old patch, I forgot about the patch..)

v2: exempt the root inode as Ted suggested, although unlink("/") would be
catched by vfs and unlink a corrupt file with root inode number would be
catched by ext4_lookup, and won't reach ext4_unlink() in both cases

	EXT4-fs error (device loop0): ext4_lookup:1441: inode #2: comm rm: 'testfile' linked to parent dir
	Aborting journal on device loop0-8.
	EXT4-fs (loop0): Remounting filesystem read-only
	EXT4-fs error (device loop0): ext4_lookup:1441: inode #2: comm rm: 'testfile' linked to parent dir

 fs/ext4/namei.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index 603e4eb..6e6b312 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -2796,9 +2796,11 @@ end_rmdir:
 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 {
 	int retval;
+	unsigned long ino;
 	struct inode *inode;
 	struct buffer_head *bh;
 	struct ext4_dir_entry_2 *de;
+	struct super_block *sb;
 	handle_t *handle = NULL;
 
 	trace_ext4_unlink_enter(dir, dentry);
@@ -2815,13 +2817,20 @@ static int ext4_unlink(struct inode *dir, struct dentry *dentry)
 		goto end_unlink;
 
 	inode = dentry->d_inode;
+	ino = inode->i_ino;
+	sb = dir->i_sb;
 
 	retval = -EIO;
-	if (le32_to_cpu(de->inode) != inode->i_ino)
+	if (le32_to_cpu(de->inode) != ino)
 		goto end_unlink;
+	if ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
+	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) {
+		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
+		goto end_unlink;
+	}
 
 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
-				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
+				    EXT4_DATA_TRANS_BLOCKS(sb));
 	if (IS_ERR(handle)) {
 		retval = PTR_ERR(handle);
 		handle = NULL;
-- 
1.8.3.1

Re: [PATCH v2] ext4: don't remove reserved inodes in ext4_unlink()

[Darrick J. Wong]

On Sun, Oct 12, 2014 at 04:50:58PM +0800, Eryu Guan wrote:
> Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
> when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
> the journal inode is gone, and unmounting such a fs could trigger the
> following BUG_ON() in start_this_handle().
> 
> 	BUG_ON(journal->j_flags & JBD2_UNMOUNT);
> 
> 	------------[ cut here ]------------
> 	kernel BUG at fs/jbd2/transaction.c:307!
> 	...
> 	CPU: 1 PID: 1535 Comm: umount Not tainted 3.13.0+ #14
> 	...
> 	Call Trace:
> 	 [<ffffffff8119f17a>] ? kmem_cache_alloc+0x1ca/0x1f0
> 	 [<ffffffff812850f0>] ? jbd2__journal_start+0x90/0x1e0
> 	 [<ffffffff81285153>] jbd2__journal_start+0xf3/0x1e0
> 	 [<ffffffff81242a62>] ? ext4_evict_inode+0x1b2/0x4f0
> 	 [<ffffffff8126d039>] __ext4_journal_start_sb+0x69/0xe0
> 	 [<ffffffff81242a62>] ext4_evict_inode+0x1b2/0x4f0
> 	 [<ffffffff811d3b8e>] evict+0x9e/0x190
> 	 [<ffffffff811d4373>] iput+0xf3/0x180
> 	 [<ffffffff8128f301>] jbd2_journal_destroy+0x191/0x220
> 	 [<ffffffff810b0ae0>] ? abort_exclusive_wait+0xb0/0xb0
> 	 [<ffffffff8125d004>] ext4_put_super+0x64/0x340
> 	 [<ffffffff811bbae2>] generic_shutdown_super+0x72/0xf0
> 	 [<ffffffff811bbd77>] kill_block_super+0x27/0x70
> 	 [<ffffffff811bc05d>] deactivate_locked_super+0x3d/0x60
> 	 [<ffffffff811bc606>] deactivate_super+0x46/0x60
> 	 [<ffffffff811d7f47>] mntput_no_expire+0xa7/0x140
> 	 [<ffffffff811d939e>] SyS_umount+0x8e/0x100
> 	 [<ffffffff81690c29>] system_call_fastpath+0x16/0x1b
> 
> Check inode number in ext4_unlink() and return error if the inode number
> is reserved or nonexistent(except EXT4_ROOT_INO, as Ted pointed out that
> it's a security hole).
> 
> Tested by removing a reserved inode(modify the ondisk structure by hand)
> and unmounting the fs. Inodes 1-10 have been tested. Also tested by
> xfstests.
> 
> Signed-off-by: Eryu Guan <guaneryu@gmail.com>

Looks reasonable to me, you can add Reviewed-by if you like.

--D

> ---
> 
> (This is a v2 of an old patch, I forgot about the patch..)
> 
> v2: exempt the root inode as Ted suggested, although unlink("/") would be
> catched by vfs and unlink a corrupt file with root inode number would be
> catched by ext4_lookup, and won't reach ext4_unlink() in both cases
> 
> 	EXT4-fs error (device loop0): ext4_lookup:1441: inode #2: comm rm: 'testfile' linked to parent dir
> 	Aborting journal on device loop0-8.
> 	EXT4-fs (loop0): Remounting filesystem read-only
> 	EXT4-fs error (device loop0): ext4_lookup:1441: inode #2: comm rm: 'testfile' linked to parent dir
> 
>  fs/ext4/namei.c | 13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> index 603e4eb..6e6b312 100644
> --- a/fs/ext4/namei.c
> +++ b/fs/ext4/namei.c
> @@ -2796,9 +2796,11 @@ end_rmdir:
>  static int ext4_unlink(struct inode *dir, struct dentry *dentry)
>  {
>  	int retval;
> +	unsigned long ino;
>  	struct inode *inode;
>  	struct buffer_head *bh;
>  	struct ext4_dir_entry_2 *de;
> +	struct super_block *sb;
>  	handle_t *handle = NULL;
>  
>  	trace_ext4_unlink_enter(dir, dentry);
> @@ -2815,13 +2817,20 @@ static int ext4_unlink(struct inode *dir, struct dentry *dentry)
>  		goto end_unlink;
>  
>  	inode = dentry->d_inode;
> +	ino = inode->i_ino;
> +	sb = dir->i_sb;
>  
>  	retval = -EIO;
> -	if (le32_to_cpu(de->inode) != inode->i_ino)
> +	if (le32_to_cpu(de->inode) != ino)
>  		goto end_unlink;
> +	if ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
> +	    ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) {
> +		ext4_error(sb, "reserved or nonexistent inode %lu", ino);
> +		goto end_unlink;
> +	}
>  
>  	handle = ext4_journal_start(dir, EXT4_HT_DIR,
> -				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
> +				    EXT4_DATA_TRANS_BLOCKS(sb));
>  	if (IS_ERR(handle)) {
>  		retval = PTR_ERR(handle);
>  		handle = NULL;
> -- 
> 1.8.3.1
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v2] ext4: don't remove reserved inodes in ext4_unlink()

[Theodore Ts'o]

On Sun, Oct 12, 2014 at 04:50:58PM +0800, Eryu Guan wrote:
> Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
> when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
> the journal inode is gone, and unmounting such a fs could trigger the
> following BUG_ON() in start_this_handle()....
> 

I believe the bug that this patch is trying to fix has been addressed
by this commit:

http://git.kernel.org/cgit/linux/kernel/git/tytso/ext4.git/commit/?h=dev&id=bf8ad98e1bffa5ce178ef5e4ea803a86ac30f9e5

ext4: add ext4_iget_normal() which is to be used for dir tree lookups
If there is a corrupted file system which has directory entries that
point at reserved, metadata inodes, prohibit them from being used by
treating them the same way we treat Boot Loader inodes --- that is,
mark them to be bad inodes.  This prohibits them from being opened,
deleted, or modified via chmod, chown, utimes, etc.

In particular, this prevents a corrupted file system which has a
directory entry which points at the journal inode from being deleted
and its blocks released, after which point Much Hilarity Ensues.

Reported-by: Sami Liedes <sami.liedes@iki.fi>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

					- Ted

Re: [PATCH v2] ext4: don't remove reserved inodes in ext4_unlink()

[Eryu Guan]

On Mon, Oct 13, 2014 at 12:21:00PM -0400, Theodore Ts'o wrote:
> On Sun, Oct 12, 2014 at 04:50:58PM +0800, Eryu Guan wrote:
> > Corrupted ext4_dir_entry_2 struct on disk may have wrong inode number,
> > when the inode number is 8 (EXT4_JOURNAL_INO) and the file is deleted,
> > the journal inode is gone, and unmounting such a fs could trigger the
> > following BUG_ON() in start_this_handle()....
> > 
> 
> I believe the bug that this patch is trying to fix has been addressed
> by this commit:
> 
> http://git.kernel.org/cgit/linux/kernel/git/tytso/ext4.git/commit/?h=dev&id=bf8ad98e1bffa5ce178ef5e4ea803a86ac30f9e5

Yes, this patch fixes the issue I'm seeing, thanks for pointing it out!

I have one concern thouth, removing a reserved inode (I tested
EXT4_JOURNAL_INO) on corrupted ext4 returns EIO as expect but the fs
is not marked as containing error(as other EIOs in ext4_iget()) and no
error logs in dmesg. User may have no idea what happened and the
corruped fs is still being used as normal. I think EXT4_ERROR_INODE
should be called too somewhere in such case.

Thanks,
Eryu
> 
> ext4: add ext4_iget_normal() which is to be used for dir tree lookups
> If there is a corrupted file system which has directory entries that
> point at reserved, metadata inodes, prohibit them from being used by
> treating them the same way we treat Boot Loader inodes --- that is,
> mark them to be bad inodes.  This prohibits them from being opened,
> deleted, or modified via chmod, chown, utimes, etc.
> 
> In particular, this prevents a corrupted file system which has a
> directory entry which points at the journal inode from being deleted
> and its blocks released, after which point Much Hilarity Ensues.
> 
> Reported-by: Sami Liedes <sami.liedes@iki.fi>
> Signed-off-by: Theodore Ts'o <tytso@mit.edu>
> Cc: stable@vger.kernel.org
> 
> 					- Ted
> 
>