[Devel] Re: [PATCH] ecryptfs ecryptfs_read_super path_lookup errh fix

Dmitriy Monakhov dmonakhov at sw.ru
Mon Feb 19 11:11:59 PST 2007


Dmitriy Monakhov <dmonakhov at sw.ru> writes:

> if path_lookup() return non zero code we don't have to worry about 'nd' 
> parameter, but ecryptfs_read_super does path_release(&nd) after path_lookup
> has failed, and dentry counter becomes negative :) , this result in folowing BUG
>
> ecryptfs_read_super: path_lookup() failed
> BUG: Dentry f3f074c4{i=0,n=enc_dir} still in use (-1) [unmount of ext3 dm-5]
> <<<<<<<<<<<<<<<<<<<<<counter is negative here   ^^^^^^
> ------------[ cut here ]------------
> kernel BUG at fs/dcache.c:654!
>  [<c01063d6>] die+0x11e/0x233
>  [<c042ed50>] do_trap+0x8a/0xa3
>  [<c0106979>] do_invalid_op+0x97/0xa1
>  [<c042eb14>] error_code+0x7c/0x84
>  [<c0183f5f>] shrink_dcache_for_umount+0x2d/0x3a
>  [<c0176166>] generic_shutdown_super+0x18/0xbe
>  [<c017622c>] kill_block_super+0x20/0x32
>  [<c01762ce>] deactivate_super+0x3f/0x51
>  [<c018724d>] mntput_no_expire+0x42/0x6b
>  [<c017a719>] path_release+0x20/0x23
>  [<f886387d>] ecryptfs_get_sb+0x45a/0x4ad [ecryptfs]
>  [<c0176361>] vfs_kern_mount+0x81/0xf1
>  [<c0176419>] do_kern_mount+0x30/0x42
>  [<c018847b>] do_mount+0x601/0x678
>  [<c0188561>] sys_mount+0x6f/0xa9
>  [<c0104f2c>] sysenter_past_esp+0x5d/0x99
>  =======================
> Code: 30 02 00 00 89 44 24 18 8b 45 ec 89 4c 24 14 89 74 24 10 89 7c 24 0c 89 5c 24 04 89 44 24 08 c7 04 24 ee 9d 4f c0 e8 b7 74 fa ff <0f> 0b eb fe 8b 73 30 39 de 75 04 31 f6 eb 03 f0 ff 0e 8d 43 48 
> EIP: [<c01831a7>] shrink_dcache_for_umount_subtree+0x159/0x1fb SS:ESP 0068:f4269c7c
>
> This is easy to reproduce just try to mount ecryptfs to nonexisting lower path
> # mount -tecryptfs private/this_dir_not_exist root -ocipher=aes
This is fun but ecryptfs_read_super() code is more crappy whan it looks at 
first blush.
1) After path_lookup succeed we dot't have any guarantie what it is DIR.
   This must be checked explicitly.
2) path_lookup can't return negative dentry, So inode check is useless.

Following patch is updated version and have to be applied instead of first patch 
Signed-off-by: Dmitriy Monakhov <dmonakhov at openvz.org>
-------------
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: diff-mm-ecryptfs-ecryptfs_read_super-fix
URL: <http://lists.openvz.org/pipermail/devel/attachments/20070219/a20bc959/attachment-0001.ksh>


More information about the Devel mailing list