[PATCH] cgroups: fix a serious bug in cgroupstats

[PATCH] cgroups: fix a serious bug in cgroupstats

[Li Zefan]

Try this, and you'll get oops immediately:
 # cd Documentation/accounting/
 # gcc -o getdelays getdelays.c
 # mount -t cgroup -o debug xxx /mnt
 # ./getdelays -C /mnt/tasks

Because a normal file's dentry->d_fsdata is a pointer to struct cftype,
not struct cgroup.

After the patch, it returns EINVAL if we try to get cgroupstats
from a normal file.

CC: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
 kernel/cgroup.c |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 358e775..f6f5219 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -2039,10 +2039,13 @@ int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
 	struct cgroup *cgrp;
 	struct cgroup_iter it;
 	struct task_struct *tsk;
+
 	/*
-	 * Validate dentry by checking the superblock operations
+	 * Validate dentry by checking the superblock operations,
+	 * and make sure it's a directory.
 	 */
-	if (dentry->d_sb->s_op != &cgroup_ops)
+	if (dentry->d_sb->s_op != &cgroup_ops ||
+	    !S_ISDIR(dentry->d_inode->i_mode))
 		 goto err;
 
 	ret = 0;
-- 
1.5.4.rc3

Re: [PATCH] cgroups: fix a serious bug in cgroupstats

[Andrew Morton]

On Wed, 19 Nov 2008 16:14:41 +0800 Li Zefan <lizf@cn.fujitsu.com> wrote:

> Try this, and you'll get oops immediately:

I guess that's wrong.

>  # cd Documentation/accounting/
>  # gcc -o getdelays getdelays.c
>  # mount -t cgroup -o debug xxx /mnt
>  # ./getdelays -C /mnt/tasks
> 
> Because a normal file's dentry->d_fsdata is a pointer to struct cftype,
> not struct cgroup.
> 
> After the patch, it returns EINVAL if we try to get cgroupstats
> from a normal file.
> 
> CC: Balbir Singh <balbir@linux.vnet.ibm.com>
> Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
> ---
>  kernel/cgroup.c |    7 +++++--
>  1 files changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
> index 358e775..f6f5219 100644
> --- a/kernel/cgroup.c
> +++ b/kernel/cgroup.c
> @@ -2039,10 +2039,13 @@ int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
>  	struct cgroup *cgrp;
>  	struct cgroup_iter it;
>  	struct task_struct *tsk;
> +
>  	/*
> -	 * Validate dentry by checking the superblock operations
> +	 * Validate dentry by checking the superblock operations,
> +	 * and make sure it's a directory.
>  	 */
> -	if (dentry->d_sb->s_op != &cgroup_ops)
> +	if (dentry->d_sb->s_op != &cgroup_ops ||
> +	    !S_ISDIR(dentry->d_inode->i_mode))
>  		 goto err;
>  
>  	ret = 0;

The patch applies OK to 2.6.25, 2.6.26 and to 2.6.27.  I marked it as
needing backport to those kernel versions.  Please let me know if that
was inappropriate.

Re: [PATCH] cgroups: fix a serious bug in cgroupstats

[Li Zefan]

> The patch applies OK to 2.6.25, 2.6.26 and to 2.6.27.  I marked it as
> needing backport to those kernel versions.  Please let me know if that
> was inappropriate.
> 

Forgot to mention this:

The bug exists from 2.6.24 (when cgroupstats is introduced) to now, though
I didn't test in all those versions. I guess not many people used cgroupstats,
so it never gets triggered.

I think it can be applied to 2.6.24 too.

Re: [PATCH] cgroups: fix a serious bug in cgroupstats

[Balbir Singh]

Andrew Morton wrote:
> On Wed, 19 Nov 2008 16:14:41 +0800 Li Zefan <lizf@cn.fujitsu.com> wrote:
> 
>> Try this, and you'll get oops immediately:
> 
> I guess that's wrong.
> 
>>  # cd Documentation/accounting/
>>  # gcc -o getdelays getdelays.c
>>  # mount -t cgroup -o debug xxx /mnt
>>  # ./getdelays -C /mnt/tasks
>>
>> Because a normal file's dentry->d_fsdata is a pointer to struct cftype,
>> not struct cgroup.
>>
>> After the patch, it returns EINVAL if we try to get cgroupstats
>> from a normal file.
>>
>> CC: Balbir Singh <balbir@linux.vnet.ibm.com>
>> Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
>> ---
>>  kernel/cgroup.c |    7 +++++--
>>  1 files changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
>> index 358e775..f6f5219 100644
>> --- a/kernel/cgroup.c
>> +++ b/kernel/cgroup.c
>> @@ -2039,10 +2039,13 @@ int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
>>  	struct cgroup *cgrp;
>>  	struct cgroup_iter it;
>>  	struct task_struct *tsk;
>> +
>>  	/*
>> -	 * Validate dentry by checking the superblock operations
>> +	 * Validate dentry by checking the superblock operations,
>> +	 * and make sure it's a directory.
>>  	 */
>> -	if (dentry->d_sb->s_op != &cgroup_ops)
>> +	if (dentry->d_sb->s_op != &cgroup_ops ||
>> +	    !S_ISDIR(dentry->d_inode->i_mode))
>>  		 goto err;
>>  
>>  	ret = 0;
> 
> The patch applies OK to 2.6.25, 2.6.26 and to 2.6.27.  I marked it as
> needing backport to those kernel versions.  Please let me know if that
> was inappropriate.
> 

Thanks, it does need to be backported. Thanks Li for spotting the problem.


-- 
	Balbir

Re: [PATCH] cgroups: fix a serious bug in cgroupstats

[Paul Menage]

On Wed, Nov 19, 2008 at 12:14 AM, Li Zefan <lizf@cn.fujitsu.com> wrote:
> Try this, and you'll get oops immediately:
>  # cd Documentation/accounting/
>  # gcc -o getdelays getdelays.c
>  # mount -t cgroup -o debug xxx /mnt
>  # ./getdelays -C /mnt/tasks
>
> Because a normal file's dentry->d_fsdata is a pointer to struct cftype,
> not struct cgroup.
>
> After the patch, it returns EINVAL if we try to get cgroupstats
> from a normal file.
>
> CC: Balbir Singh <balbir@linux.vnet.ibm.com>
> Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>

Good catch, thanks.

Acked-by: Paul Menage <menage@google.com>



> ---
>  kernel/cgroup.c |    7 +++++--
>  1 files changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/cgroup.c b/kernel/cgroup.c
> index 358e775..f6f5219 100644
> --- a/kernel/cgroup.c
> +++ b/kernel/cgroup.c
> @@ -2039,10 +2039,13 @@ int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
>        struct cgroup *cgrp;
>        struct cgroup_iter it;
>        struct task_struct *tsk;
> +
>        /*
> -        * Validate dentry by checking the superblock operations
> +        * Validate dentry by checking the superblock operations,
> +        * and make sure it's a directory.
>         */
> -       if (dentry->d_sb->s_op != &cgroup_ops)
> +       if (dentry->d_sb->s_op != &cgroup_ops ||
> +           !S_ISDIR(dentry->d_inode->i_mode))
>                 goto err;
>
>        ret = 0;
> --
> 1.5.4.rc3
>