-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Fri, Feb 02, 2007 at 05:04:53AM -0200, Rodrigo Barbosa wrote:
I've got a complete lockdown on cfsd, and had to hardboot the machine. In both cases, I've lost nothing, and only the specific file I was copying, on the destination, was "lost" (partially copied). It gave me a VERY good impression of cfs' robustness.
That sounds Good. heh..
On the other hand, CFS is VERY succeptible to a nasty nfs related deadlock. The scenario is easy to imagine.
Say cfsd tried to write to the disk and has to wait. Then, you will get a nfs timeout. Since you have a nfs timeout, processed will stall. Since cfsd is stalled, you can't get out of the timeout.
I'm still trying to figure out the best way to solve this. Maybe multithreading cfsd, or maybe simply using O_NONBLOCK. If I can think of a good way to solve this, I might be able to patch it. I'm not sure about the O_NONBLOCK solution. It is kind of basic, and someone would have though of it before is it was all it takes, I guess.
This condition is VERY easy to trigger for me, if I copy a file from to the crypted filesystem, and both (crypted and non-crypted) as located on an external USB disk I have here.
Ok, I'm giving up on CFS, at least for now. Since I already use fuse for a lot of stuff (sshfs, obexfs and mysqlfs), I'm trying encfs now.
Will let you know how it turns out.
[]s
- -- Rodrigo Barbosa "Quid quid Latine dictum sit, altum viditur" "Be excellent to each other ..." - Bill & Ted (Wyld Stallyns)