Re: [fuse-devel] how to register a exit handler

[Hans B. <han...@gm...>]

On 2011-06-14 1:42, the...@gm... wrote:
> On Mon, Jun 13, 2011 at 6:05 PM, Hans Beckérus<han...@gm...>  wrote:
>> I stumbled into a design issue that forces me to leave the basic
>> fuse_main() approach and instead take the long turn using fuse_new() etc.
>> The problem I then got into is that eg. SIGTERM does not automatically
>> unmount my fs and I can not figure out how to register a exit handler.
>> To be able to umount the fs both mount point and channel is needed. How
>> can I easily register a exit handler that can pick up the data needed
>> for me to call fuse_unmount() and fuse_exit() ? Also, is it enough to
>> call fuse_unmount() and fuse_destroy() or should I use fuse_teardown()
>> instead?
> If you want your filesystem to call fuse_exit it will have to be done
> from either a separate thread or as part of signal handler. Once the
> main loop exits you can then call the unmount and destroy. When the
> umount goes through the kernel fuse_session_exit is called
> automatically by the channel receiver.
>

Thanks for the input. Actually I already knew that fuse_exit() needs to 
be called from another thread or signal handler.
I do have unmount and clean-up code after my event loops stops. But the 
problem is not when the event loop is
stopped in a controlled fashion. I need to call fuse_exit() also for eg. 
SIGINT (CTRL-C).
Registering my own signal handler for SIGINT could solve that, but then 
I do not know how to easily propagate the struct fuse*  to be able to 
call fuse_exit() for the correct instance. That is why I want to 
register a exit handler instead on session level using the 
fuse_session_ops::exit.
But I can not find a suitable API function that can do that? And I do 
not want to go through fuse_create_session() which means I need to start 
using the fuse_lowlevel API.

> fuse_teardown() calls  fuse_unmount_common() and fuse_destroy() for
> you. It will also call fuse_remove_signal_handlers() which resets HUP,
> INT, TERM and PIPE to their default values. The only warning I would
> give for usingfuse_teardown() is that it also tries to free
> mountpoint. If you don't want mountpoint to be freed then you are
> better off calling fuse_unmount() and fuse_destroy() separately.
>
Great. I will remember that.

>> I tried to use fuse_set_signal_handlers() but that just made things
>> worse :( Then I could not even break out of the foreground process using
>> CTRL-C.
>>
Hmm, actually this was not working as expected. Using a *very* dirty 
trick I managed to register my own exit handler in fuse_session_ops 
(that is why I need some help doing it the proper way). Then if I also 
call fuse_set_signal_handlers() and just leave the exit handler empty, 
fuse_exit() is called automatically at SIGINT and my event loop 
terminates nicely. But, If I do not register a exit handler and only 
call fuse_set_signal_handlers()  SIGINT does not work, eg. fuse_exit() 
is never called and clean-up is not possible! This I can not explain.


> -nate