[libvirt] [PATCH v7 2/6] add fuse support for libvirt lxc

[Gao feng]

于 2012年11月07日 16:50, Richard W.M. Jones 写道:
> On Wed, Nov 07, 2012 at 01:45:27PM +0800, Wanlong Gao wrote:
>> On 11/07/2012 11:54 AM, Gao feng wrote:
>>> 于 2012年11月06日 20:55, Richard W.M. Jones 写道:
>>>> On Tue, Nov 06, 2012 at 07:50:33PM +0800, Gao feng wrote:
>>>>> 于 2012年11月06日 17:11, Richard W.M. Jones 写道:
>>>>>> On Tue, Nov 06, 2012 at 04:54:59PM +0800, Gao feng wrote:
>>>>>>> Because libvirt_lxc create a thread to mount fuse filesystem.
>>>>>>> and then libvirt_lxc will try to access the fuse filesystem.
>>>>>>>
>>>>>>> Without waiting,libvirt_lxc may access fuse before it is mounted.
>>>>>>>
>>>>>>> And I can't find a better way to make accessing fuse after the fuse
>>>>>>> being mounted,because fuse_main will return until fuse being unmounted,so
>>>>>>> I don't know how to pass message to tell libvirt_lxc whether the fuse is
>>>>>>> mounted success.
>>>>>>
>>>>>> Oh I see.  This is solvable, but I believe that you cannot use
>>>>>> fuse_main.  You have to look at the fuse sources (quite complex!)  and
>>>>>> do the individual fuse operations.
>>>>>>
>>>>>> Have a look at how libguestfs does it ...
>>>>>>
>>>>>>   https://github.com/libguestfs/libguestfs/blob/master/src/fuse.c#L911
>>>>>>   http://libguestfs.org/guestfs.3.html#mount-local
>>>>>>
>>>>>
>>>>>
>>>>> Thanks for the information.
>>>>>
>>>>> Can fuse_mount & fuse_new & fuse_loop avoid the waiting?
>>>>>
>>>>> It seems has the same problem.
>>>>
>>>> It waits, but it doesn't busy wait in a loop.  The first access to the
>>>> filesystem hangs until the thread which is handling the filesystem is
>>>> listening for requests.
>>>>
>>>
>>> It makes me confuse.
>>>
>>> There are two threads(A and B).
>>> thread A mounts fuse filesystem on mountpoint /mnt/fuse.
>>> thread B try to access file /mnt/fuse/meminfo. and we don't
>>> know which thread will execute first.
>>>
>>> The information I got from your comment is that because thread A
>>> doesn't mount fuse to /mnt/fuse, when thread B access to the
>>> /mnt/fuse/meminfo,it will hang. I am right?
>>
>> If I'm not wrong, I think Rich means that you should create the mountpoint
>> and the operations using fuse_mount and fuse_new before you create the thread B,
>> then make another thread A using fuse_loop to loop behind.  Then if thread B try
>> to access the mountpoint before thread A runs, B will hang until thread A runs into the loop.
> 
> This is right.
> 

With my test and reading fuse source,the thread B will return with error ENOTCONN.

Thanks
Gao