[CRIU] [PATCH 1/2] files: restore cwd after switching mntns

[Pavel Emelyanov]

On 11/19/2015 03:37 PM, Andrew Vagin wrote:
> On Thu, Nov 19, 2015 at 03:23:30PM +0300, Pavel Emelyanov wrote:
>> On 11/18/2015 02:04 PM, Andrey Vagin wrote:
>>> From: Andrew Vagin <avagin at virtuozzo.com>
>>
>> Comment required.
> 
> Here is a problem that restore_ns() restores a mount namespace, but
> doesn't restore cwd, so cwd will be an root of the restired mntns.
> 
> CRIU doesn't expect this, so we will see errors. Fro example, Stas
> reports an error that criu can't umount an roots yard.

Nice. Does the same problem exists for the root dir? Do we want to
fix it too?

>>
>>> Cc: Stanislav Kinsbursky <skinsbursky at openvz.org>
>>> Reported-by: Stanislav Kinsbursky <skinsbursky at openvz.org>
>>> Signed-off-by: Andrew Vagin <avagin at virtuozzo.com>
>>> ---
>>>  files-reg.c | 16 ++++++++++++++++
>>>  1 file changed, 16 insertions(+)
>>>
>>> diff --git a/files-reg.c b/files-reg.c
>>> index 1e2daa9..ac334e7 100644
>>> --- a/files-reg.c
>>> +++ b/files-reg.c
>>> @@ -423,6 +423,7 @@ void try_clean_remaps(int ns_fd)
>>>  {
>>>  	struct remap_info *ri;
>>>  	int old_ns = -1;
>>> +	int cwd_fd = -1;
>>>  
>>>  	if (list_empty(&remaps))
>>>  		goto out;
>>> @@ -436,6 +437,12 @@ void try_clean_remaps(int ns_fd)
>>>  			return;
>>>  		}
>>>  
>>> +		cwd_fd = open(".", O_DIRECTORY);
>>> +		if (cwd_fd < 0) {
>>> +			pr_perror("Unable to open cwd");
>>> +			return;
>>> +		}
>>> +
>>>  		if (setns(ns_fd, CLONE_NEWNS) < 0) {
>>>  			close(old_ns);
>>>  			pr_perror("`- Can't switch");
>>> @@ -453,6 +460,15 @@ void try_clean_remaps(int ns_fd)
>>>  		close(old_ns);
>>>  	}
>>>  
>>> +	if (cwd_fd >= 0) {
>>> +		if (fchdir(cwd_fd)) {
>>> +			pr_perror("Unable to restore cwd");
>>> +			close(cwd_fd);
>>> +			return;
>>> +		}
>>> +		close(cwd_fd);
>>> +	}
>>> +
>>>  out:
>>>  	if (ns_fd >= 0)
>>>  		close(ns_fd);
>>>
>>
> .
>