Re: [sshfs] synchronisation bug in sshfs

[Florian Z. <fl...@fl...>]

Hi,

> On Sat, 26 Feb 2011, Florian Zumbiehl wrote:
> > while reading the source of sshfs, I noticed that there (almost certainly)
> > is a race condition between sftp_request_send() and process_one_request()
> > when the former is waiting on sshfs.outstanding_cond for the pipeline to
> > empty: If the ssh connection then disappears, the first sftp_read() in
> > process_one_request(), for example, could return -1, causing
> > process_one_request() to do the same, causing process_requests() to
> > close the fds, to reset sshfs.processing_thread_started, to clean the
> > request hash table, and to then terminate the thread, leaving noone to
> > wake up the thread sleeping in sftp_request_send(), thus resulting in
> > a deadlock.
> 
> Thanks for the report.
> 
> Could you please test the attached patch?  Does it fix the deadlock?

Erm, I don't have a clue how to provoke that deadlock, or at least I
haven't tried--I just noticed it while reading the source. And after all,
testing doesn't really work for concurrency problems anyhow ;-)

As I also don't know the source particularly well, I guess you should be
able to judge whether that would make things correct better than I
could.

Florian