[patch] improve locking mechanizm during staging process

[Stanislav Ievlev]

Current locking implementation during staging process will not properly work on network filesystems (we are using a shared cache for all machines in cluster).
I've replaced a simple touch() and unlink() calls with a more perfect solution.

My functions use both fcntl and mutexes. This technique prevents a race conditions both between a local threads in one process
and between different processes on different machines.

[Dmitrii Zagorodnov (dmitrii)]

Hi Stanislav,

Thanks a lot for the patch, I'm adding it to the 2.0.* code branch. I'll remove the

sem_v (disk_sem);
...
sem_p (disk_sem);

pair, because lock_file_aquire() is already called with sc_sem held and disk_sem is used to serialize long-running disk operations in storage.c. Does that make sense?

Regards,
Dmitrii

[Andy Grimm (agrimm)]

This issue is now being tracked upstream at http://eucalyptus.atlassian.net/browse/EUCA-2680

Please watch that issue for further updates.